Tests of six DDR3 memory kits: in pursuit of speed. Is the frequency of RAM important? RAM timings ddr3 2133

Oh, RAM. Everyone needs RAM. The market offers just a ton of options ranging from 2133 to 4266Mhz. Should you empty your wallet for the fastest memory on the market? Let's figure it out!

DDR4 is the most current generation random access memory to date. The RAM itself is used to seamlessly feed data to your processors, which turns all those zeros and ones into games you play, boring presentations or boring articles like the ones you read on this site \u003d).

Therefore, the faster the memory, the better performance, right?

Not really. High-frequency memory can throw a huge amount of data on the processor, but there will come a point when the processor stops handling it fast enough for you to notice the difference. So, we just need to find a middle ground.

There are uncles and aunts who are members of the Semiconductor Engineering Standardization Committee, or JEDEC for short. They set the standard frequency for DDR4 - 2133Mhz. At the same time, on Z270 and x99 platforms, anything higher than DDR4 2400 is no longer included in the official specification.

AMD, in turn, has also set the standard memory frequency around 2400Mhz.

Then why the hell are companies like G.SKILL selling RAM at nearly TWO times the standard?

There is only one answer - authorized overclocking. Intel's Extreme Memory Profile, or XPM, has become a kind of standard means of allowing high-end RAM to be used at the stated frequencies using certified motherboards.

Every stick of RAM, even the fastest, is programmed with the aforementioned standard frequency, defined by JEDEC. In addition, a faster XMP profile is stored in the high end memory on a separate chip. In order to ensure full compatibility, the memory initially operates at the standard JEDEC frequency, and in order to increase the frequency to the one declared by the manufacturer, you must purposefully enable XMP in the BIOS settings.

All this bright and promising theory crashed on the harsh stones of reality. For example, despite all assurances from ASUS, AMD or GSKILL, many enthusiasts were unable to overclock the memory above 2666Mhz during the release of Ryzen 7.

A lot of time has passed since then, fixes and updates for motherboards were released, but overclocking remained overclocking, albeit sanctioned, the success in which is difficult to predict. It all depends on luck as well as configuration. Nevertheless, even a successful stable overclocking can sometimes give such surprises:

Tests

Intel	AMD
i7-7700K	R7 1700X
ASUS Maximus IX	ASUS Crosshair VI
G.SKILL Trident RGB DDR4-3600	G.SKILL FLAREX DDR4-3200
GTX 1080Ti	GTX 1080Ti
Kingston KC400 1TB SSD	Kingston KC400 1TB SSD

AMD managed to squeeze out 3200Mhz, Intel - 3600Mhz. There was no point in overclocking above, you will soon find out why.

In games, the JEDEC standard of 2133Mhz was used as a basis for counting indicators. Deus Ex and Tomb Raider have been tested in DX11 and DirectX 12 modes.

Obviously, some games do not pay attention to the frequency of the RAM. Others, in turn, were found to be more frequency sensitive than expected.

In For Honor, the average frame rate was within the margin of error, but a closer look at the results shows that the minimum frame rate of the 97th percentile was actually affected by the RAM frequency.

In synthetic tests, more uniform results were obtained.

We can already conclude that in almost all cases, the most noticeable increase in performance occurs when upgrading from 2133Mhz to 3200Mhz. Therefore, we can conclude: on both platforms, without taking into account marketing and synthetic tests, it makes no sense to climb higher, even if there is an opportunity and funds for it.

Hence, the sweet spot is 3200Mhz RAM? Well, not really.

It all depends on your budget. If you are focusing only on the best hardware, then for the sake of these few percent of performance it makes sense to pay extra.

If your budget is tight, here's a tip: the difference in price between 2133 and 3600 RAM is about $ 64, which can be spent on upgrading the same processor. For example, you can take the 7600k instead of the i5 7500. This simple operation will give you a higher clock speed as well as overclocking capability. In the case of Ryzen, this amount can be upgraded from Ryzen 5 1500x to 1600x, which will give you TWO additional cores.

With the financial component sorted out. It remains to talk about the possible time spent on overclocking. Let's look at the numbers again: when you upgrade the standard JEDEC 2133Mhz RAM to the fastest one discussed in this article, you will pay $ 64 for a 4.7% increase in performance, with no guarantees that it will all work as stated.

This does not mean that we do not recommend it. Logically, you might think that you should only do this when everything that really improves the performance and usability of your build has already been purchased. Keep in mind that careful incremental overclocking takes time and skill, not to mention a lot of patience.

Reviews of new memory modules appear on our site quite regularly. This time we will test high-speed dual-channel DDR3 memory kits with a total capacity of 16 GB. A distinctive feature of all these kits is the presence of Intel XMP profiles (Extreme Memory Profiles), which can be used on motherboards for Intel processors with support for XMP profiles.

Instead of a preface to this review, I would like to make a few remarks about modern DDR3 memory.

As you know, almost all manufacturers of memory modules offer a very wide range of products aimed at various categories of users. This is common memory, gaming memory, and memory for overclockers. Recall that there are not so many manufacturers of memory chips themselves: such companies as Samsung, Micron and Hynix are the industry leaders. It is clear that the manufacturers of modules have not that great choice. So where does such a wide range of products come from?

Of course, all these different memory series are pure marketing. Memory modules belonging to different series may have exactly the same characteristics (and even the same memory chips) and differ only in the color of the heatsink. By the way, the heatsinks themselves on memory modules are purely decorative and, by and large, pointless thing. Well, the memory chips do not heat up so much that they need cooling using radiators! Let's not be unfounded and confirm what we have said with facts.

In order to demonstrate the senselessness of heat sinks on memory modules, we used a pyrometer, which allows us to remotely determine the temperature change. Once we used a DDR3-2400 memory module with a heatsink, and another time without. The supply voltage was 1.65 V (standard supply voltage was 1.5 V). To load memory, we used the Stress System Memory stress test in the AIDA64 utility. Our measurement results are as follows. When the memory is working with a heatsink, the heatsink temperature increases by 7-8 ° C in the memory loading mode compared to the temperature in idle mode. When the memory module is operating without a heatsink, the temperature of the memory chips increases by 15-16 ° C in the memory loading mode compared to the temperature in the idle mode. It would seem that the difference of 7 ° C is not so little. But the point is that the absolute temperature of memory chips in the stress mode is only 45-46 ° C, which is absolutely uncritical for a microcircuit.

Of course, you can try to overclock the memory even more by applying a higher voltage and increasing the frequency. But even if the memory starts up at this higher frequency, in terms of heating, this will not give a significant increase. So, we note once again that modern memory modules do not need heatsinks.

In general, heatsinks on modern memory modules do not so much function as a heat sink, but rather allow manufacturers to simply expand their product range. I painted the radiator black - here's a new memory line for overclockers; installed pink heatsinks - got a new line of memory for girls ... Besides the possibility of obtaining different lines of memory, heatsinks are also a sign that we are talking about high-speed memory modules that operate at a higher frequency not specified in the JEDEC specification.

Recall that, according to the JEDEC standard, the maximum (effective) frequency of DDR3 memory is 1333 MHz with timings of 9-9-9 and a supply voltage of 1.5 V. Naturally, any modern DDR3 memory will operate at 1333 MHz at 1.5 V , however, all memory manufacturers also produce faster modules (DDR3-1600 / 1866/2133/2400/2600), guaranteeing their stable operation in such overclocking mode. Memory operation at higher frequencies can be implemented both through the XMP profile, in which the frequency, supply voltage and timings are prescribed, and by setting all the listed parameters in manual mode (if the BIOS of the board does not support work with XMP profiles). However, do not forget that the ability of memory to run at a higher speed than the JEDEC specification depends not only on the module, but also on the memory controller integrated into the processor. For newer 4th Generation Intel Core processors (codenamed Haswell), the memory controller officially only supports DDR3-1600 memory. Naturally, he is able to support faster memory, but without any guarantees (here you are, how lucky). As practice shows, most Haswell processors can support DDR3-1866 / 2133/2400/2600 memory without any problems.

Increasing the memory frequency, as a rule, requires changing other parameters as well - timings, the supply voltage of the memory modules themselves and the supply voltage of the memory controller. The memory supply voltage, of course, does not affect the system performance in any way, but an increase in timings with a simultaneous increase in clock frequency can lead to the fact that DDR3-2133 memory with lower timings will be more efficient in performance than DDR3-2400 memory with higher timings. Therefore, it is far from always worth chasing higher clock frequencies.

As for the influence of memory speed characteristics on the system performance as a whole, everything is very ambiguous here. In general, custom applications that would receive tangible an increase in performance (speed of task execution) from an increase in memory frequency simply does not exist. That is, the fact that you are doubling the memory frequency does not mean that there will be such applications in which the speed of task execution will also double. In some applications, such an increase in clock frequency will not affect the speed at all, while in others the increase in speed will be very modest. An increase in the clock frequency in the processor in many (but also not in all) applications leads to an adequate increase in the speed of executing tasks, but with memory everything is a little different. However, we have already talked about this many times. Let's make a reservation that such reasoning is valid provided that the memory operates in [at least] two-channel mode, but in modern systems this condition is almost always met. And even single-channel memory (such options can be found in some laptops) will not double the speedup when the operating frequency is doubled. On the other hand, even if in some applications the performance gain from using faster memory is 5-7%, why not? Especially if you consider that the difference in cost between the usual (DDR3-1333) and high-speed memory of the same size is not that great.

Next, we will look at several dual-channel sets of modern high-speed DDR3 memory with a total volume of 16 GB. These are kits of two or four memory models: if the kit consists of four modules, it was installed into the test system with two modules per channel, and in the case of two modules, one module per channel. So, let's start with a more detailed acquaintance with the participants in our testing.

Kingston HyperX Predator KHX24C11T2K2 / 8X

Kingston HyperX Predator KHX24C11T2K2 / 8X memory refers to the Kingston HyperX Predator series overclocking gaming memory. You can read the following user caution regarding this memory series: “Users may experience extreme motion sickness and / or complete loss of orientation due to the extremely high speeds achieved with HyperX Predator modules. They are not intended for children, faint-hearted people, people who are in no hurry, and for all those who can be content with little. Memory modules up to 2666 MHz, new heatsink for improved heat dissipation, Intel XMP support, compatible with motherboards All major brands and legendary Kingston reliability. We would even recommend using a helmet. "

This is, of course, a joke, but it definitely characterizes the audience for which these memory modules are targeted.

HyperX Predator KHX24C11T2K2 / 8X memory is a set of two DDR3-2400 modules with a total capacity of 8 GB. Let's make a reservation right away that we used two sets of HyperX Predator KHX24C11T2K2 / 8X memory so that the total volume was 16 GB.

These memory modules are labeled KHX24C11T2K2 / 8X. Recall that for Kingston HyperX memory modules the following decoding is used. The first three letters - KHX - indicate that this is Kingston HyperX memory. The next two digits define the memory clock speed. In our case, this is 24, which corresponds to a clock frequency of 2400 MHz. Next, the CAS Latency value is set. Here C11 indicates that the CAS Latency is 11 clocks. The next two characters (in our case T2) define the type of memory within the Kingston HyperX series. The following is the number of memory modules included in the kit. So, K2 corresponds to two memory modules. The slash indicates the total memory capacity for the kit in gigabytes, and the presence of the letter X indicates the memory compatibility with Intel XMP profiles (eXtreme Memory Profiles).

Thus, the KHX24C11T2K2 / 8X marking means that we are talking about a set of two DDR3 Kingston HyperX Predator memory modules with a clock frequency of 2400 MHz and a CAS Latency value of 11 cycles. The total memory capacity is 8 GB, in addition, the memory is compatible with Intel XMP profiles.

According to the specification, KHX24C11T2K2 / 8X memory modules support operation at a frequency of 1333 MHz with a supply voltage of 1.5 V and timings of 9-9-9 (JEDEC specification), as well as two XMP profiles. The first profile corresponds to a clock frequency of 2400 MHz, and the second to a frequency of 2133 MHz. For the first XMP profile, the supply voltage is 1.65 V, and the timings are 11-13-13. For the second XMP profile, the supply voltage is 1.60 V, and the timings are 11-12-11.

It remains to add that the KHX24C11T2K2 / 8X memory modules have proprietary heatsinks for efficient heat dissipation, and the height of the memory module with the heatsink is 53.9 mm, and its thickness is 7.24 mm.

On our test bench (see below), the Kingston HyperX Predator DDR3-2400 KHX24C11T2K2 / 8X memory started up without problems when using the XMP profile at 2400 MHz (timings 11-13-13). The frequency of 2600 MHz, with unchanged timings, was beyond the capacity of Kingston HyperX Predator DDR3-2400 KHX24C11T2K2 / 8X memory modules. However, they are not required to operate at this frequency.

Below are the test results of the Kingston HyperX Predator DDR3-2400 KHX24C11T2K2 / 8X memory module kit at 1333 MHz (9-9-9-24) and 2400 MHz (11-13-13-30) in the AIDA64 program. Let us remind you once again that during testing we used two sets of Kingston HyperX Predator DDR3-2400 KHX24C11T2K2 / 8X memory.

Kingston HyperX Beast KHX21C11T3K2 / 16X

The Kingston HyperX Beast KHX21C11T3K2 / 16X memory belongs to the Kingston series overclocking gaming memory.

A distinctive feature of the memory modules in this series is that they use black printed circuit boards and a black aluminum heatsink.

The manufacturer's website notes that this design was made at the request of HyperX fans "to aggressively improve any system enthusiasts." It is not very clear what is meant (apparently, these are the features of the translation), but "at the request of HyperX fans" - it's just like in the USSR, when prices were raised at the request of workers.

Again, according to the manufacturer's website, the HyperX Beast series memory modules are designed to work with third generation Intel Core i5 and i7 processors and AMD processors.

Actually, there is only one comment - this information is already outdated, and the memory modules of this series are perfectly compatible with fourth-generation Intel Core processors.

We also add that HyperX Beast series memory modules are available in dual-channel and quad-channel kits with capacities from 8 to 64 GB and frequencies up to 2400 MHz. The modules of this series are provided with a lifetime warranty.

Kingston HyperX Beast KHX21C11T3K2 / 16X is a dual channel set of two memory modules with a total capacity of 16 GB (2 × 8 GB). As it follows from the KHX21C11T3K2 / 16X marking, the modules of this memory can operate at a clock frequency of 2133 MHz, and the CAS Latency value is 11 clock cycles.

According to the Kingston HyperX Beast KHX21C11T3K2 / 16X memory modules support operation at a frequency of 1333 MHz with a supply voltage of 1.5 V and timings 9-9-9 (JEDEC specification), as well as two XMP profiles. The first profile corresponds to a clock frequency of 2133 MHz, and the second to a frequency of 1600 MHz. For the first XMP profile, the supply voltage is 1.60 V, and the timings are 11-12-11. For the second XMP profile, the supply voltage is 1.5 V, and the timings are 9-9-9.

On our test bench, the Kingston HyperX Beast KHX21C11T3K2 / 16X memory started up without problems when using the XMP profile at 2133 MHz (timings 11-12-11-30).

In addition, as it turned out, the Kingston HyperX Beast KHX21C11T3K2 / 16X memory kit works without any problems at 2400 MHz, and with the same timings as at 2133 MHz.

Geil Evo Veloce Frost White GEW316GB2400C11ADC

Geil Evo Veloce Frost White GEW316GB2400C11ADC dual-channel memory kit belongs to the series announced by the company in 2012. Memory kits in this series are equipped with Maximum Thermal Conduction & Dissipation heatsinks in red or white. Memory modules with white heatsinks are called Frost White, and those with red heatsinks are called Hot-rod Red.

In general, I must say that Geil's assortment simply contains a huge number of different series of DDR3 memory, and in each series there are several variants of memory modules. Why such a huge range of products is needed is not very clear. After all, it is obvious that if we discard all the marketing "nonsense", it turns out that the memory modules hiding behind heatsinks of different colors and belonging to different series are essentially the same thing.

For example, dual channel DDR3-2400 memory kits belonging to the Geil Evo Veloce Frost White, Geil Evo Veloce Hot-rod Red and Evo Leggera series differ, in fact, only in the color of the heatsink and marketing positioning. Each of these series contains sets of memory modules with the same timings and the same size. And, most likely, the memory chips themselves in these modules are the same. However, let's return to the consideration of the Geil Evo Veloce Frost White GEW316GB2400C11ADC dual-channel memory module kit.

So, we are talking about a set of two DDR3-2400 memory modules with a total volume of 16 GB (2 × 8 GB). The memory modules are equipped with white heatsinks, that is, they belong to the Frost White series. In general, it should be noted that the radiators on the memory, although they have their own brand name, do not look impressive, let's say. The thickness of the plates from which the radiator is made is only 1 mm. The memory module with a heatsink is 47 mm high and 16.8 mm thick.

According to the information, at a frequency of 2400 MHz, Geil Evo Veloce Frost White GEW316GB2400C11ADC memory modules can work with timings 11-12-12-30 at a supply voltage of 1.65 V.

Moreover, this mode of operation of memory modules is provided upon activation of the Intel XMP profile and is guaranteed by the manufacturer only on motherboards with Intel X79 and Intel Z77 chipsets, as indicated by the corresponding sticker on the packaging of memory modules.

The guaranteed compatibility with Intel X79 and Intel Z77 chipsets is explained by the fact that motherboards based on these chipsets support Intel XMP memory profiles. Naturally, today support for XMP profiles is provided by a large number of chipsets (in particular, Intel 8 series chipsets), so that it is possible to guarantee the operability of this memory with an XMP profile on motherboards with the Intel Z87 chipset.

However, recall that Intel XMP profiles are not supported on motherboards with AMD chipsets, and to run this memory in overclocked mode, you need to set the frequency, voltage and timings in manual mode.

Note that the Geil Evo Veloce Frost White series of dual-channel DDR3-2400 memory also includes 8 and 16 GB memory kits with timings 9-11-10-28 (GEW38GB2400C9DC / GEW316GB2400C9DC), 10-11-11-30 (GEW38GB2400C10DC / GEW316GB2400C10DC) , 10-12-12-30 (GEW38GB2400C10ADC / GEW316GB2400C10ADC), 11-11-11-30 (GEW38GB2400C11DC / GEW316GB2400C11DC). So the GEW316GB2400C11ADC memory kit has the least aggressive timings in the DDR3-2400 Geil Evo Veloce Frost White line, that is, it is the youngest model in the series.

On our test bench, the Geil Evo Veloce Frost White GEW316GB2400C11ADC memory started up without problems when using the XMP profile at 2400 MHz.

The 2600 MHz frequency, with unchanged timings, turned out to be beyond the capacity of these memory modules. However, increasing the main timings by one stop makes it easy to run this memory at 2600 MHz.

Corsair Vengeance CMZ16GX3M2A1866C9

The Corsair Vengeance CMZ16GX3M2A1866C9 is a dual channel DDR3-1866 memory kit with a total capacity of 16GB (2x8GB).

This memory kit also belongs to the Corsair Vengeance series aimed at overclockers.

In terms of the design of aluminum heatsinks, the modules of the Corsair Vengeance CMZ16GX3M2A1866C9 dual-channel memory kit practically do not differ from the modules of the Corsair Vengeance CMZ16GX3M4X2133C11R quad-channel memory kit. The only difference is the color of the heatsink. In this case, it is black.

According to the information, Corsair Vengeance CMZ16GX3M2A1866C9 memory modules support a frequency of 1866 MHz with timings of 9-10-9-27 and a supply voltage of 1.5 V.

Naturally, this mode of operation corresponds to the XMP profile. Well, in the standard operating mode, the memory operates in DDR-1333 mode with timings of 9-9-9-24.

On our test bench, the Corsair Vengeance CMZ16GX3M2A1866C9 memory started up without problems using the XMP profile at 1866 MHz.

However, as it turned out, the 1866 MHz frequency is not the limit for this memory and it can be easily overclocked to 2000 MHz with the same timings as for the 1866 MHz frequency.

Corsair Vengeance CMZ16GX3M4X2133C11R

The Corsair Vengeance CMZ16GX3M4X2133C11R is a set of four DDR3-2133 memory modules with a total capacity of 16GB (4x4GB).

This memory kit belongs to the Corsair Vengeance series aimed at overclockers. According to the information, Corsair Vengeance series memory modules use memory chips specially selected for high performance potential.

The modules in this kit are equipped with heat sinks, which not only provide heat dissipation, but also serve as an aggressive design element that is great for gaming computers. The heatsink on the memory module consists of two aluminum plates (one plate on each side of the module) 1 mm thick, painted burgundy and have stickers indicating the series and characteristics of the module. The height of the memory modules, taking into account the heatsink, is 53 mm, and the width is 17 mm.

Note that the Corsair Vengeance series includes one-, two-, three- and four-channel memory kits with volumes from 4 to 16 GB, which differ in timings, color, and even the shape of the radiator.

The Corsair Vengeance CMZ16GX3M4X2133C11R kit, as already noted, consists of four 4GB memory modules. Accordingly, this kit can be used in two-channel or four-channel memory modes.

According to the information, Corsair Vengeance CMZ16GX3M4X2133C11R memory modules support a frequency of 2133 MHz with timings 11-11-11-27 and a supply voltage of 1.5 V.

Naturally, this mode of operation corresponds to the XMP profile. Well, in the standard operating mode, the memory operates in DDR3-1333 mode with 9-9-9-24 timings.

True, according to the results of the diagnostic test in the AIDA64 utility, it turned out that slightly different timings are registered in the XMP profile of this memory: not 11-11-11-27, but 11-11-11-30. The difference, of course, is not significant, but it is.

On our test bench, the Corsair Vengeance CMZ16GX3M4X2133C11R memory started up without problems when using the XMP profile at 2133 MHz with timings 11-11-11-30.

Moreover, it turned out that with unchanged timings, this memory runs without problems at 2200 MHz.

Corsair Vengeance Pro CMY16GX3M4A2400C10R

The Corsair Vengeance Pro CMY16GX3M4A2400C10R is a dual channel DDR3-2400 memory kit with a total capacity of 16GB (2x8GB).

This memory kit belongs to the Corsair Vengeance Pro series aimed at overclockers. noted that Corsair Vengeance Pro series memory kits are specially designed for Intel processors Core third and fourth generations.

The memory modules of this series use aluminum heat sinks of various colors. The height of the memory modules, taking into account the heatsink, is 46 mm, and the width is 17.5 mm.

The Corsair Vengeance Pro series includes kits of two or four memory modules with a total capacity of 8 to 32 GB and frequencies from 1600 to 2400 MHz.

The Corsair Vengeance Pro CMY16GX3M4A2400C10R memory kit, as noted, consists of two 8GB memory modules. These memory modules are equipped with black aluminum heatsinks with a decorative burgundy insert. On one side of the heatsink there is a sticker with information about the memory series (Vengeance Pro), and on the other side there is a sticker with information about the characteristics of the memory module (frequency, timings, supply voltage).

According to the information, Corsair Vengeance Pro CMY16GX3M4A2400C10R memory modules support a frequency of 2400 MHz with timings of 10-12-12-31 and a supply voltage of 1.65 V.

Naturally, this mode of operation corresponds to the XMP profile. Well, in the standard operating mode, the memory operates in DDR-1333 mode with timings 9-9-9-24.

As it turned out during testing, with the Corsair Vengeance Pro CMY16GX3M4A2400C10R modules, everything was not easy.

The fact is that the declared XMP-profile for 2400 MHz is absent. Instead, there is an XMP profile at 1866 MHz with timings of 9-10-9-27. But even when this profile is activated in the BIOS, the memory operates at 1800 MHz, not 1866 MHz.

Nevertheless, if you set the memory frequency, supply voltage and timings manually in the BIOS (2400 MHz, 1.65 V, 10-12-12-31), then the memory will work as it should.

Testing

So, in total, six memory kits took part in our testing, each of which was tested in two operating modes:

• Corsair vengeance pro
  • Corsair CMY16GX3M2A2400C10R @ 1800 MHz 9-10-9-27
  • Corsair CMY16GX3M2A2400C10R @ 2400 MHz 10-12-12-31
• Corsair Vengeance (DDR3-1866)
  • Corsair CMZ16GX3M2A1866C9 @ 1866 MHz 9-10-9-27
  • Corsair CMZ16GX3M2A1866C9 @ 2000 MHz 9-10-9-27
• Corsair Vengeance (DDR3-2133)
  • Corsair CMZ16GX3M4X2133C11R @ 2133 MHz 11-11-11-30
  • Corsair CMZ16GX3M4X2133C11R @ 2200 MHz 11-11-11-30
• Geil evo veloce
  • Geil GEW316GB2400C11ADC @ 2400 MHz 11-12-12-30
  • Geil GEW316GB2400C11ADC @ 2600 MHz 12-13-13-32
• Kingston hyperx beast
  • Kingston KHX21C11T3K2 / 16X @ 2133 MHz 11-12-11-30
  • Kingston KHX21C11T3K2 / 16X @ 2400 MHz 11-12-11-30
• Kingston HyperX Predator
  • Kingston KHX24C11T2K2 / 8X @ 1333 MHz 9-9-9-24
  • Kingston KHX24C11T2K2 / 8X @ 2400 MHz 11-13-13-30

For testing, we used a stand with the following configuration:

• processor - Intel Core i7-4770K;
• motherboard - ASRock Z87 OC Formula;
• chipset - Intel Z87;
• drive - Intel SSD 520 Series (240 GB);
• operating system - Windows 8 (64-bit).

Perhaps the most non-trivial task when testing memory is to find those applications and tasks in which you can really see the difference in performance for memory with different frequencies.

Naturally, we used the synthetic test AIDA64, which allows us to determine the speed of reading, writing and copying data, as well as memory latency. The results of this synthetic test are shown below.

As a basis, we took the Kingston HyperX KHX24C11T2K2 / 8X memory in 1333 MHz mode with 9-9-9-24 timings, which corresponds to the JEDEC specification.

As you can see, here you can easily see the difference between DDR3-1333 memory and memory with a higher clock speed.

However, this test is synthetic. Now let's see what will happen in tests based on real applications.

As we have already said, not all applications are "sensitive" to the memory speed - more precisely, the DDR3-1333 bandwidth is sufficient for most applications, and further increase in the memory frequency becomes meaningless. Nevertheless, we managed to find a number of benchmarks based on real applications, in which we can fix the difference in system performance when using memory modules with different frequencies.

As a result, we selected the following set of applications for testing:

• MediaCoder x64 0.8.25.5560;
• Adobe Premiere Pro CC
• Adobe after Effects CC;
• Adobe Photoshop CC;
• Adobe Audition CC;
• Photodex ProShow Gold 5.0.3276;
• WinRAR 5.0.

In the application MediaCoder x64 0.8.25.5560 A 3:35 HD video is transcoded to a different format with a lower resolution. The original video is recorded in H.264 format and has the following characteristics:

• size - 1.05 GB;
• container - MKV;
• resolution - 1920 × 1080;
• frame rate - 25 fps;
• video bitrate - 42.1 Mbit / s;
• audio bitrate - 128 Kbps;
• number of audio channels - 2;
• sampling frequency - 44.1 kHz.

The parameters of the resulting video file are as follows:

• size - 258 MB;
• container - MP4;
• video codec - MPEG4 AVC (H.264);
• resolution - 1280 × 720;
• frame rate - 29.97 fps;
• video bitrate - 10000 Kbps;
• audio codec - AAC;
• audio bitrate - 128 Kbps;
• number of channels - 2;

The result of this test is the conversion time.

Adobe Premiere Pro CC a video is created from ten video clips with a total volume of 1.48 GB. The video clips (MOV container) were shot with a Canon EOS Mark II 5D at 1920 × 1080 resolution and 25 fps. Transition effects are created between all video clips, after which the stage is rendered and the video file with the preset is exported Apple iPad 2, 3, 4, Mini; iPhone 4S, 5; Apple TV3 - 1080p 25... The length of the finished film is 4:25 and the volume is 163 MB.

• container - MP4;
• resolution - 1920 × 1080;
• video codec - MPEG4 AVC (H.264);
• video bitrate - 5 Mbit / s;
• frame rate - 25 fps;
• audio codec - AAC;
• idiobitrate - 160 Kbps;

The result of this test is the total time it takes to render and export the movie.

In a test using an app Adobe After Effects CC a 30-second video clip (MOV container) of 164 MB in size is processed, shot with a Canon EOS Mark II 5D camera with a resolution of 1920 × 1080 and a frame rate of 25 fps, followed by rendering without compression (AVI container) using the built-in render.

Processing consists of adjusting the white balance, applying a Cartoon filter and overlaying 3D titles with various effects (explosion, blur, etc.)

The output file parameters are as follows:

• resolution - 1920 × 1080;
• video codec - no (uncompressed video);
• container - AVI;
• video bitrate - 1492 Mbit / s;
• frame rate - 30 fps.
• audio codec - PCM;
• audio bitrate - 1536 Kbps;
• number of channels - 2 (stereo);
• sampling frequency - 48 kHz.

The output video file size is 5.21 GB. The result of this test is the video rendering time.

Photodeх ProShow Gold 5.0.3276 determines the speed of creating an HD video (slideshow) with a resolution of 1920 × 1080 (MPEG-2 format, 59.94 fps) from 24 digital photos captured with an EOS Canon Mark II 5D camera and converted to TIFF format. Each photo is 60.1 MB in size. In addition, music is superimposed on the film. The movie itself is created using the Wizard of the Photodex ProShow application. Various transition effects are applied between individual slides, and some of the slides are animated.

The result of the test is the total time for creating a slideshow project, including the time for loading photos and music and applying special effects, as well as the time for exporting the project to a movie.

In an application test Adobe Photoshop CC 24 photos taken with the EOS Canon Mark II 5D in RAW format (25 MB each) are processed in batch. For each photo that is opened in 8-bit format, the following steps are performed sequentially:

• the color depth changes from 8 to 16 bits per channel;
• the Smart Sharpen adaptive filter is applied;
• shake Reduction filter is applied;
• the Reduce Noise filter is applied;
• lens Correction filter is applied;
• the color depth changes from 16 to 8 bits per channel;
• the photo is saved in TIFF format.

The result of this test is the batch processing time for all photos.

In an application test Adobe Audition CC a six-channel (5.1) FLAC (lossless compressed) audio file is initially processed and then converted to MP3 format. The processing of the original file is to apply an Adaptive Noise Reduction filter to it. The result of the test is the total time of processing and converting an audio file. The original test audio file is 1.65 GB in size. The parameters of the resulting MP3 file are as follows:

• bit rate - 128 Kbps;
• sampling frequency - 48 kHz.

In a test using an application app WinRAR 5.0 (64-bit version) archives an album of 24 digital photos in TIFF format (size of each photo - 60.1 MB). In the WinRAR 5.0 archiver, data compression uses the RAR5 format, the Best compression method (maximum compression) and the dictionary size is 32 MB.

The test result is the archiving time.

When testing memory, all tests were run three times, between each launch the computer was rebooted.

Test results

Now let's turn to the test results. As before, we took the Kingston KHX24C11T2K2 / 8X memory in 1333 MHz mode with 9-9-9-24 timings as a basis.

Let's start with a video transcoding test using MediaCoder x64 0.8.25.5560 application. As you can see, this task is not very sensitive to memory performance. The worst result (112.4 s for DDR3-1333 memory) differs from the best (109.1 s for DDR3-2400 memory) by only 3%. But there is practically no difference in the speed of the test between DDR3-1866 and DDR3-2400.

Adobe Premiere Pro CC is a little more memory-sensitive, with a 6.5% difference between the worst and the best in our test. Well, that's already something.

But in the test based on the Adobe After Effects CC application, the difference between the worst and the best result again does not exceed 3%.

Photodex ProShow Gold is slightly more sensitive to memory speed, with a 6% difference between the worst and the best in our test.

Adobe Photoshop CC was found to be even more sensitive to memory speed. Here we finally saw something that can really be called the difference: 11% between the best and worst result. However, the worst here, of course, is the memory indicator DDR3-1333, and if we take DDR3-1800 as the base indicator, the difference, alas, is reduced to 5%.

The test results based on the Adobe Audition CC application from our methodology are presented not so much to demonstrate the advantages of high-speed memory, but to demonstrate the absence of these advantages in many, many applications. In our test based on this application, the difference between the worst and the best results is only 2%, that is, there is practically no difference at all.

But the data compression test based on the WinRAR 5.0 application is very sensitive to the memory speed. The Photoshop record was not achieved here, but the difference between the worst and the best result is quite a respectable 9.5%, which is very good.

conclusions

Actually, the conclusions that can be drawn from our testing are quite predictable. High-speed memory doesn't make much sense today, and DDR3-1333 memory is enough for most user applications. The maximum performance gain that can be obtained by using high-speed DDR3-2400 or DDR3-2600 memory instead of standard DDR3-1333 memory can barely exceed 10%, and tasks that reveal such an advantage of high-speed memory still need to be looked for.

As for a variety of bizarre heatsinks on high-speed memory modules, which, according to marketers, can improve the efficiency of heat dissipation, this is nothing more than fiction. Modern memory with a frequency of 2400 and even 2600 MHz with a supply voltage increased to 1.65 V does not need heatsinks at all, which was confirmed by the numbers in the preface to this review.

Now about the cost. On average, a set of high-speed DDR3-2400 memory with a volume of 16 GB costs about 7-8 thousand rubles (you can find it more expensive - it all depends on the brand, model and the conscience of the seller). A set of DDR3-1333 memory of the same size (and of the same brand) will cost about 5-6 thousand rubles.

If we are talking about a top-end high-performance PC based on a processor, for example, Intel Core i7-4770K and a motherboard on the Intel Z87 chipset, then even a few percent of additional performance due to the use of high-speed memory may not be superfluous, and then there is no point in saving on memory. especially since the difference in cost between high-speed memory and standard memory is very small (compared to the cost of such a computer as a whole, of course). If we are talking about an ordinary inexpensive or office PC, then there is no point at all in high-speed memory.

As for the question of choosing a specific manufacturer (Kingston, Corsair, Geil, Samsung, etc.), we would like to remind you that all memory modules use chips made by Samsung, Micron and Hynix. And by and large, it is absolutely unimportant who exactly is the manufacturer of the memory module. Perhaps this is the last thing worth paying attention to.

Actually, this post was supposed to be short, but since the tests took me a certain amount of time, let it be a little longer.

It all started with the transition to the motherboard from DDR3. I don't even remember how many years ago it was. Then, with the motherboard, I immediately took two Patriot strips of 2 GB each with a frequency of 1600 MHz in "Trash.net" (by the way, they did not overclock at all - it was impossible to take more than 1600).

The total is 4 GB. It was "a lot" in those days. Then I gradually changed three motherboards and 4-gig sticks began to appear on sale. And I wanted to "upgrade". It just so happened that at the moment when I took a couple of 4-gig bars, there were no optimal prices for 1600-frequency ones and I took the usual Patriot at 1333 MHz from the Monitor (now 8 GB). A month later, for the sake of experiment, I took the same pair of slats (it became 16 GB). Fortunately, the motherboard supports up to 32 GB. These 1333 bars were consistently overclocked without raising the timings, without raising the voltage to 1600 MHz. But not higher.

And on Saturday I bought two 4GB DDR3 strips from Hynix (Hyundai Electronics). They ended up with the party number HMT351U6CFR8C-H9, which was noted on the overclockers.ru forums

http://forums.overclockers.ru/viewtopic.php?f\u003d111&t\u003d292041

The batch is HMT351U6CFR8C-H9.

This batch is famous for the fact that in most cases it is capable of overclocking from "native" 1333 MHz to stable 2133 MHz.

To my delight, it was this batch that we found on free sale in one of the shops in Yakutsk.

The bar is produced without heatsinks, but it is interesting to note that it does not heat up at all (including at 100% test load) neither at the nominal frequency of 1333 MHz (1.50 V), nor at 2133 MHz (1.59 V) ... It turns out that in this case, radiators are not needed here.

For example, here is what the overclockers forum wrote about this memory:

Hynix Original DDR3 PC3-10600

Rating: (1333 MHz) 9-9-9-27 1.50v

Acceleration: (1866 Mhz) 9-10-9-30 1.55v

Acceleration: (2133 Mhz) 10-12-11-30 1.57v

Overclocking MAX: when raising vccio to 1.120v takes (2271mhz) 11-12-11-30 1.59v

I placed two Hynix bars in dual channel mode. And immediately before the first launch, I cleared the CMOS. I went into Windows - looked in CPU-Z - the memory was detected correctly.

Then I rebooted and set the memory frequency to 2133 MHz at a voltage of 1.59 V (they wrote on the forum that this sponge would be enough in most cases) and "sparing" timings 11-12-11-30-2T (again, in most cases, the data I set the timings with a "margin"). By the way, I should immediately write that I will only "work" on 4 main timings, which most strongly affect memory performance.

Everything started - I went into Windows. Since I only had LinX out of the fresh stress tests using RAM, for the first time I decided to use only it in tests. However, it is common knowledge that no program can UNIMINELY and quickly detect memory errors. Be sure to test for a long time on different types stress programs that make good use of the operative in their work. For example, in the "Memory Testing Methodology" branch on overclockers.ru they recommend - [further - the list of programs]

In the test, LinX used 6500 MB of memory out of the installed 8 GB.

Duration - 10 passes.

Control of all voltages, all kinds of frequencies, memory timings and temperatures - CPU-Zx64 1.59, ASRock eXtreme Tuner 0.1.54, Core Temp 0.99.8.

Then I sharply reduced all timings by one.

10-11-10-29-2Т - pass # 7 linpak did not pass.

10-11-10-30-2T - all 10 passes of the linpak passed.

9-11-10-30-2T - does not enter.

10-10-10-30-2Т - does not enter.

10-11-9-30-2Т - when approaching, there is a BSOD.

10-11-10-30-1Т - pass No. 4 linpak did not pass.

So I found out that in the case of a stress load with a linpak, the minimum timings at a frequency of 2133 MHz, ensuring stability, are the following figures: 10-11-10-30-2T.

On Sunday I went to the store and bought a second set of the same slats: 2 pcs. x 4 GB.

His data:

HMT351U6CFR8C-H9 - the same batch as in the upper set, differ only in the week of release.

I'm testing the second set.

Nominal: 1333 MHz, 9-9-9-25, 1.50 V.

I increased the voltage for overclocking by timings to 1.59 V.

The frequency is 2133 MHz.

11-12-11-30-2Т - enters Windows, has not tested

10-11-10-30-2Т - does not enter

11-11-10-30-2Т - does not enter

11-11-11-30-2Т - does not enter

11-12-11-30-2Т - passed all passes stably.

Now I will test all four strips: "set_1" + "set_2"

At par: 9-9-9-25, 1.50 V - everything is stable. Everything comes in and works, however, it should be so - since there is no overclocking.

Also made clear_cmos after installing all 4 boards on the motherboard.

I again increased the voltage for overclocking by timings to 1.59 V.

The frequency is 2133 MHz. Timings - 11-12-11-30-2Т. Don `t come in.

The frequency is 2133 MHz. Timings - 12-13-12-35-2Т. Don `t come in.

Frequency overclocking removed to native 1333 MHz.

1333 MHz, 11-12-11-30-2Т. It still does not come in.

1333 MHz, 10-11-10-30-2Т. Goes into Windows.

1600 MHz, 10-11-10-30-2Т. Goes into Windows.

1866 MHz, 10-11-10-30-2Т. Goes into Windows.

An attempt to increase the first timing by one leads to a complete stop. The computer goes into "unconscious".

The same garbage when changing the second and third timings. Those. they cannot be increased. You cannot reduce it either. Increasing the memory voltage to 1.7 does not help in any way. The fourth timing was not touched at all - let it remain so.

Thus, overclocking for four slats came to only one end:

1866 MHz, 10-11-10-30-2Т. I reduced the voltage from 1.59 to 1.56 V. I'll try to test at this voltage. It seems so far everything is fine.

Test computer configuration:

Intel Core i5 2500K, 4600 MHz, 1.350 V;

Cooler ThermalRight Silver Arrow, TR TY-140 х 2 pcs. x 1300 rpm;

ASRock P67 Extreme6 P67 motherboard (bios P1.60);

DDR3 memory, 2 x 4 Gb 1333 MHz, Hynix HMT351U6CFR8C-H9, 1.50 V, 9-9-9-25;

Hard 500Gb, WD5000AAKS (SATA2, 7200 rpm, 16 Mb);

Lian Li PC-A70FB case, and 4 pieces of native built-in fans + open side wall;

Reobas Zalman ZM-MFC1 Plus;

PSU AeroCool Strike-X 1100 (1100 W, 80+ Gold);

Video Inno 3D Geforce GTX570 (732/1464/3800, 1,000 V) - the reference CO is replaced by DeepCool V6000;

Monitor 24 "Acer P246H 1920 * 1080.

Calculations:

2133/1333 \u003d 1.6 - 60% increase in frequency.

1866/1333 \u003d 1.4 - 40% increase in frequency.

Overclocking from 1333 MHz to 2133 is technically and practically justified, despite a slight increase in timings. Causes:

Increasing the frequency by 40% -60% affects the increase in memory performance more than an increase in timings;

An increase in voltage from "native" 1.50 to 1.55-1.59 V does not lead to heating of memory microcircuits (even with long-term loads);

The cost of the standard DDR3 memory in 4 GB trims from Hynix is \u200b\u200bextremely low (700 rubles). This is especially noticeable when compared with proprietary overclocked memory strips, which have an increased voltage (up to 1.65 V and higher) and prices are at least twice as high (especially strips with frequencies of 1866-2133 MHz). Although they can be justified by writing:

a) a well-fitted radiator that is both beautiful and provides proper heat dissipation;

b) selection of planks in factories - i.e. theoretically they have a higher probability of higher overclocking.

Rumors that obviously overclocking strips, when installed in four pieces, lose their frequency potential - have been confirmed. This is actually the case.

III. Of course, a 60% increase in frequency is a very good result.

But is he so good "in life", so to speak? ..

In synthetics, there will definitely be an increase, but in practical applications, I think, there will be, but not 60%, of course.

In general, let's see. Unsubscribe after the tests.

Now comes the question - "What will I do with the planks?"

A mandatory component that is required when assembling any computer device is RAM. If you look closely at the range of stores, you will notice several clear trends. First, as regards desktops, it is quite possible to assert that DDR3-1333 strips are already leaving the scene, and the most "popular" memory clock frequency has become 1600 MHz (PC3-12800). It is among the PC3-12800 kits that there is now the largest number of offers at different prices. Secondly, now in the average computer they put not 4 GB of RAM, but 8 GB. The second factor is caused not so much by the increased requirements of the software as by the general desire to put more memory in your computer. At the same time, in gaming and professional computers, 16 GB and 32 GB of RAM are more and more often found. But a whale of two 4 GB modules will remain a "classic" for a home PC for a long time, and constant demand contributes to lower prices.

Therefore, we decided to test several kits, not limiting ourselves to 1600 MHz, but, on the contrary, to take the fastest whales with 2133 MHz and 2400 MHz memory clock frequencies that are relevant for modern platforms.

The main requirements for the kits will be operability and high performance when setting the parameters declared by the manufacturer (timing values \u200b\u200brecorded in the SPD). We will also appreciate the ease of installation (which can be extremely difficult due to high radiators), the appearance and quality of the packaging. We will also carry out a series of tests with parameters that are overestimated relative to the nominal value, in order to determine the suitability of the chips for overclocking and the expediency of this venture as such.

In general, are these megahertz needed?

This question was raised by a noticeable difference in the cost of memory with "budget" and "overclocking" clock frequencies. To date, a kit DDR3 2 x 4 GB with a frequency of 1600 MHz costs 1300 rubles in retail, high-speed DDR3-2133 memory is already 1900 rubles. and higher. So you wonder which is better, to give preference to a faster set of smaller volume (it's cheaper) or to pay the required amount and take four 4 GB slats? You can argue and look for the truth for a long time, but one thing is clear that 16 GB of RAM will always cost significantly more than 8 GB, and if you do not professionally engage in cryptography, data archiving and video editing, most of the RAM will not be used. Therefore, it is most reasonable to stop at an 8-gig set, the modules in which, if not matched for joint work, are at least assembled from microcircuits of the same batch, which means, at least a little, but less likely to encounter incompatibility. Plus, branded kit packaging is more durable than an anti-static OEM memory bag, and has a higher chance of surviving the straps during transit. So, let's see what sets have come to our testlab.

Hope for XMP, but don't do it yourself

A little warning. Even if you like some memory, you will read a lot of tests and positive reviews on it, do not rush to the store. One important nuance must be taken into account. You probably know that the lion's share of sudden "whims" of computers is due to memory failure? Manufacturers of components, in particular, motherboards, are well aware of this, therefore, when preparing a product for entering the market, they prepare a compatibility sheet for a specific motherboard model with various components, mainly with processors and RAM. Moreover, if support for more modern CPUs appears in new BIOS revisions, then the expansion of the range of compatible memory occurs much less often.

When assembling a computer, the reliability of which has increased requirements, be it a mini-server, or an HTPC with a tight case and a weak power supply, you should choose any memory model from the QVL (Quality Vendor List) list, but if the main goal is overclocking, you can take a chance. installing memory that the vendor did not have time to test and approve, but in this case, you will have to adjust the timings and supply voltage of the modules according to their specification, without relying on SPD and XMP. And in case of a difficult launch of the system or the appearance of "blue screens" do not forget to go through the full Memtest test cycle.

DDR3 Memory Tests | Goodram Pro DDR3-2133 2 x 2 GB (GP2133D364L10 / 4GDC)

This 2 x 2 GB DDR kit is manufactured in Poland by Wilk Elektronik. Memory strips are hidden in an individual transparent blister package, and the required number of blisters for double or triple whales is tightened with paper tape, on which the kit name, its article number and the timings required for operation at a frequency of 2133 MHz (10-10-10-30) are printed ... There is no information about the required supply voltage, there is no information about the "correct power supply" and on the sticker on the radiators. There is no specification on the manufacturer's website either, but there is a piece of paper inside the package reminding you to set the correct operating frequency in the BIOS.

It is not difficult to do this because the XMP profile system is supported, but in fact only one set of XMP parameters for the 2133 MHz frequency, given above, is "hard-wired" into the EEPROM. This makes it possible, without going into settings, to make the memory work at the optimal frequency by changing literally one line in the BIOS, of course, provided that the platform is based on the Intel chipset. Is the system healthy with memory Goodram pro on the parameters chosen by the manufacturer and whether it is suitable for overclocking platforms, we will find out below.

To protect against overheating (and partly against mechanical damage) modules Goodram pro DDR3 is protected by thin (1mm) aluminum plates for some reason, blue, despite the fact that the corporate color is orange. Symbolism Goodram pro milled on one of the radiator halves, on the other there is a sticker with module parameters and a barcode. The bank of DDR3 chips is located on one side of the PCB, there are 8 of them. On the other hand, instead of microcircuits, a thick gasket is glued. Heat removal from the microcircuits is organized through a thin thermal tape, so the removal of metal plates without their preliminary technological heating is not recommended - you can tear off the chips "with meat", precedents are known. Although why take them off, from practical experience we can say that memory heating is insignificant and does not affect anything at home; and overclockers themselves know how to "bare" the chips and put on them, for example, a water block or a more efficient heat dissipator without thermal tape, directly.

The heatsinks are only two millimeters higher than the PCB, so a mechanical conflict between the strips and the "spreading" heatsink of the CPU is practically excluded. There is no need to talk about any heating of these radiators in normal modes, the temperature over the entire aluminum area did not exceed 35 ° C, even during calculations in the Super PI 32M benchmark, one must assume that in a closed case, the heating of chips is unlikely to reach any dangerous values.

It should be added that these modules are not yet widely available, and the volume of the 2 x 2 GB kit is too small to attract the attention of an enthusiast even for a moment. So we are waiting for new capacious and fast samples from Poland.

Goodram Pro Specifications
Price	n.d.
A type	DDR3-2133 SDRAM
Volume	2 x 2 GB
	2133
Nominal timings	10-10-10-30
Supply voltage, V	n.d.
Module height, mm	33
useful links	Description Goodram Pro DDR3-2133 2 x 2 GB on the manufacturer's website

DDR3 Memory Tests | Apacer Armor Series DDR3-2133 2 x 4 GB (78.BAGGL.AFK0C)

RAM from this Taiwanese manufacturer is not often found in russian stores, but flash drives and card readers have been known for a long time. Armor Series is the top-end overclocking Apacer memory series, which includes DDR3 kits with clock frequencies of 1600 MHz, 1866 MHz, 2133 MHz, and it was the last version that appeared in our testlab. Memory Apacer armor series comes in a beautiful printed packaging with a window, inside which is a plastic blister with a couple of modules. The color of the radiators can be black, red, yellow, turquoise, you can choose it to your liking. The information on the label is extremely scanty, it is absolutely impossible to understand what supply voltage the memory chips are designed for, but if you look closely at the factory tag glued to the strips, you can see the line 4GB UNB PC3-17000 CL11-11-11-30. That is, the nominal memory frequency should be 2133 MHz with worse timings than the Goodram product. How much this fact affects performance, practical tests will show.

Similar to the Goodram product, the cooling radiators do not have developed fins and are painted aluminum plate-plates, glued to the thermal tape on both sides of the printed circuit board. It's even a pity to hide such beauty in the body. The thermal interface layer is very thin, noticeably thinner than that of Goodram.

The testlab had two sets of identical memory at its disposal, not right away, but it was noticed that the heatsinks were poorly glued to the microcircuits in all four samples. True, this did not affect the performance in any way, which only emphasized the decorative purpose of these plates.

The heatsinks of Apacer memory barely extend beyond the top of the PCB, so here too there will be no contact with the powerful (wide) cooler, the bars are very compact. The total height is 32 mm, this value can be taken as a reference, and compare "colleagues" with it.

Apacer Armor Series Specifications
Price	n.d.
A type	DDR3 SDRAM
Volume	2 x 4 GB
Nominal clock frequency, MHz	2133
Nominal timings	11-11-11-30
Supply voltage, V	1,65
Module height, mm	32
useful links

DDR3 Memory Tests | Corsair Vengeance 8GB DDR3-2133 2 x 4 GB (CMZ8GX3M2X2133C9R)

Vengeance memory of the well-known overclocker brand Corsair attracts attention, first of all, with its graceful appearance, and the outer packaging is made as bright as possible. Photographic quality printing with the image of a fragment of a printed circuit board, where there are a couple of RAM modules in the slots, bright red wherever possible, familiar Intel and AMD badges in plain sight. But the variegation and brightness did not go to the detriment of the information content, on the reverse side in the window you can see the strips themselves, on the tag of which both the voltage (1.5 V) and the operating frequency (2133 MHz) are indicated, and very encouraging delays 9-11-10-30 for this frequency. The CAS # Latency parameter (the first value) sometimes has a stronger impact on performance than other parameters, so even without opening the kit, you can be sure that the chips were selected by the manufacturer to achieve stable operation at short latencies.

Inside the cardboard shell there are separate blisters for each module, thus, the requirements for reliable protection of products during transportation are fully met.

Taking the slats out of the packaging, we again see that the manufacturer paid a lot of attention to design. The radiators of the microcircuits not only have an unusual, recognizable shape, but are also painted with a metallic paint of a bright cherry color. The chubby radiator ridge sticks out great, therefore, choosing Corsair vengeance you need to know exactly how much space the processor cooler left you in your computer, or be ready to install RAM in the slots farthest from the socket.

A visual demonstration of the height of Corsair Vengeance modules installed in conjunction with the Thermaltake Frio Extreme cooler. From left to right: Apacer Armor Series (32mm), Geil Evo Corsa (42mm), Corsair Vengeance (52mm). The first (from the processor) slot is not occupied.

The Thermaltake Frio Extreme cooler turned out to be too wide for the Corsair, blocking its way to the first slot from the processor.

Corsair Vengeance Specifications
Price	3000
A type	DDR3-2133 SDRAM
Volume	2 x 4 GB
Nominal clock frequency, MHz	2133
Nominal timings	9-11-10-30
Supply voltage, V	1,5
Module height, mm	52
useful links	Description Apacer Armor Series DDR3-2133 on the manufacturer's website Apacer Armor Series DDR3-2133 specifications on the manufacturer's website Prices for Corsair Vengeance 8GB DDR3-2133 on market.yandex.ru Prices for Corsair Vengeance 8GB DDR3-2133 at price.eu

DDR3 Memory Tests | ADATA XPG Xtreme Series DDR3 2133 2 x 4 GB (AX3U2133XC4G10-2X)

ADATA, like Apacer, has never been a big player in the RAM market, but the recent expansion of the SSD and flash storage portfolio has paid off and the company has decided to try its hand at the RAM market as well.

Unlike the typical form of packaging (inner strong case and outer cardboard shell) ADATA memory strips are packed only in a blister. And due to the fact that the box is sealed in a circle, its integrity is immediately visible. The disadvantage of this solution is that it is difficult to clean the cut package. The parameters of the memory microcircuits are not indicated on the packaging, but the timings, frequency and supply voltage can be seen on the tags stuck to the memory heatsinks, which are clearly visible through the transparent packaging.

Before us again are two "low-profile" modules, the radiators of which are thin metal plates attached to a thermal tape. The bar is 30 mm high; most coolers will not hurt to put this memory in the first slot. The recommended parameters in this RAM are alarming: the supply voltage is 1.65V - clearly higher than the "hospital average" value of 1.5-1.6 V, while the set of timings does not look optimal - 10-11-11-30. The first impression is that microcircuits of a lower performance class, not capable of taking higher frequencies, were rejected by selection from the batch, and suitable copies went to complete the memory ADATA XPG Xtreme Series , and the voltage is raised up to achieve greater stability in non-standard chip modes. If so, then you shouldn't really hope for further overclocking. But we will definitely try.

ADATA XPG Xtreme Series Specifications
Price	n.d.
A type	DDR3-2133 SDRAM
Volume	2 x 4 GB
Nominal clock frequency, MHz	2133
Nominal timings	10-11-11-30
Supply voltage, V	1,65
Module height, mm	30
useful links	Description ADATA XPG Xtreme Series DDR3 2133 on the manufacturer's website ADATA XPG Xtreme Series DDR3 2133 specifications on the manufacturer's website Prices ADATA XPG Xtreme Series DDR3 2133 on market.yandex.ru

DDR3 Memory Tests | Geil Evo Corsa 2400MHz 2 x 4 GB (GOC38GB2400C11ADC)

The Geil memory in today's test belongs to a higher speed class than the previous participants. Geil manufactures memory chips independently. The product lines include both ordinary memory and overclocking memory with advanced features. Extensive sales of DDR modules from this manufacturer in Russia have not been noticed, but they are well known to enthusiasts. Colleagues have repeatedly tried to draw readers' attention to inexpensive memory, which sometimes showed very high results in tests, despite the low price.

The Evo Corsa packaging is standard both for Geil devices themselves and for many other similar kits from other factories: the inner plastic of the blister and the cardboard outer lining. There are no complaints about the design of the pack, everything is modest and even tasteful, and the parameters (timings and operating frequency) are not hidden in the far corner, but are shown in a white plate in a conspicuous place. Their duplicate is clearly visible through the window in the package. By design and appearance memory Geil evo corsa very similar to Corsair vengeance , a slightly different form of heatsinks, green PCB boards, but there are definitely similarities. The height of the slats is 47 mm, which is a lot, that is, conflicts with the overall cooling systems are possible again. It is stated that the memory is capable of starting at 2400 MHz with timings of 11-12-12-30 at a supply voltage of 1.65 V. In this situation, the numbers look quite adequate, since the clock frequency is quite high, and the CAS # Latency value \u003d 11 at 2400 MHz - fine.

Specifications Geil Evo Corsa
Price	2600
A type	DDR3-2133 SDRAM
Volume	2 x 4 GB
Nominal clock frequency, MHz	2400
Nominal timings	11-12-12-30
Supply voltage, V	1,65
Module height, mm	42
useful links	Description Geil DDR3 Evo Corsa 2400MHz on the manufacturer's website Geil DDR3 Evo Corsa 2400MHz specifications on the manufacturer's website Prices Geil DDR3 Evo Corsa 2400MHz on market.yandex.ru Prices Geil DDR3 Evo Corsa 2400MHz at price.in

BGA-chips are covered with an innovative coating consisting of carbon and silicon, which dissipates excess heat.

We'll talk about its efficiency later, we will also try to determine the temperature of the chips.

The blue packaging of the modules, the windows in which are made in the form of arrows, are both beautiful and informative. On the back there is a detailed specification, indicating the permissible supply voltage - 1.5-1.8 V, and the CAS # Latency timing values. And inside the cardboard packaging there is a rigid plastic trough, in which the modules are held very tightly. You have to make significant efforts, bending the plastic and the modules themselves, in order to get them out of the "captivity".

There is practically nothing to add to the above, we have a printed circuit board with 16 BGA chips painted in turquoise. The heat dissipating coating is of the greatest interest. Under a strong twelve-fold loupe, the patented dusting looks like a layer of baked enamel, the surface is very heterogeneous, bumpy, but at the same time smooth.

The manufacturer's declared characteristics are promising, you can try both ways to increase performance: both short timings and increased clock frequency (up to 2400 MHz). In any case, it will be very interesting to check the capabilities of "nanomemory".

Kingmax Nano Gaming RAM Specifications
Price	n.d.
A type	DDR3-2400 SDRAM
Volume	2 x 4 GB
Nominal clock frequency, MHz	2400
Nominal timings	10-11-10-30
Supply voltage, V	1,7-1,8
Module height, mm	30
useful links	Description of Kingmax Nano Gaming RAM DDR3-2400 on the manufacturer's website

DDR3 Memory Tests | Test stand, methodology and software

The memory test took place at the stand, the list of components from which it was assembled is given in the table:

• CPU AMD FX-8350 (Vishera)
• ASUS ROG Crosshair V Formula motherboard (AMD 990FX / SB950 chipset), BIOS version 1703
• Point of View GeForce GTX 580
• Cooler Master 1200W Power Supply
• Cooling Air Cooler
• Thermaltake Frio Extreme
• 240GB Corsair Neutron GTX SSD
• Operating windows system 7 x64 Ultimate

The AMD platform was chosen only on the basis of the presence of a corresponding configured motherboard, which previously proved to be absolutely stable, accurately withstanding the voltage and frequency of the tested components, at the same time, it has a huge number of adjustable parameters in the BIOS. AMD processors of the latest Vishera microarchitecture have significantly increased their speed in comparison with their predecessors. It was decided to abandon the CPU overclocking altogether, the bus frequencies and multiplier did not change from test to test, but in a real situation it is better to overclock this processor, the performance gain is decent.

Information of the main window of the CPU-Z utility. The processor frequency was set automatically in the BIOS, so there were still some deviations from round numbers to a larger side. This is a feature of the motherboard sample, or rather, this is how ASUS configured it

Choosing the AMD platform, I had to abandon the Intel XMP profile system, relying on Jedec standards and profile settings in SPD. And they were so strange for some kits that it was necessary to adjust the main parameters, since they not only did not coincide with the passport ones, but, in most cases, were not optimal.

The set of tests included purely synthetic benchmarks: these are "graphical" 3D Mark 11 (with "Performance" settings) and "platform" PCMark 7 (points taken into account in the "Creativity" and "Productivity" tests). The Creativity tests evaluate the performance of the main components when processing photo and video materials, and Productivity tests the speed when processing Internet tasks and office suites... Having received data on the effect of memory settings on the results obtained in these benchmarks, it will be possible to indirectly judge the effect of memory performance on overall PC performance.

It seemed interesting to include SiSoftware Sandra Personal 2013.01.19.23 in the tests and use it to evaluate the performance of the "processor-memory" link in cryptographic tasks, since there is such a module in the program. On the modern Internet, cryptography is used very often: content encryption modules, secure Web pages, Internet banking, remote access systems. In addition, this test responds very well to changes in memory frequencies.

The traditional AIDA Cache & Memory Benchmark will clearly show latency and bandwidth. Of the real applications, archivers, including their built-in benchmarks, are still the most sensitive to memory parameters, so let's take two popular packages: WinRAR 4.20 and 7-Zip 9.20, 64-bit versions. I had to take a couple of softwares at once, because 7-Zip performs multi-threaded calculations more accurately than WinRAR, and the estimation accuracy is higher, since the compression algorithms are slightly different.

As with any component stress test, instability can be encountered, and it is not known how and when it will manifest itself. In order not to waste time collecting the main results, at first we ran "heavy" tests, which, with incorrectly set timings, are guaranteed to lead to a hang or BSoD. This is, first of all, 3D Mark 11, especially its subsections using calculations physical models, "calculator" Super PI / mod 1.5 XS 32M and one of the archivers. If all these tests passed without hesitation, as a rule, there were no further problems. The most important thing was to get not some "nice" figure showing the performance of the AMD platform, but to compare the performance of the memory sticks with each other, providing them with the most similar modes with the same set of tests. It is clear that the operating speed of the memory controller built into the Intel CPU is higher, but in percentage terms the difference between the results of different memory kits in the same tests will be almost the same on both platforms.

DDR3 Memory Tests | Factory settings: a lot of strange

Almost every user who has bought a set of RAM is sure that installing divisors and multipliers on Auto, not only what is recommended by the manufacturer, but also eliminates the hassle with a long selection of parameters, so it is not surprising that such a scenario is most often played out. Let's see what happens in this case.

Goodram pro

Goodram Memory has proven itself to be a great reinsurer. At Auto, a frequency of 1333 MHz was set with a delay formula of 8-8-8-22, but this applies to AMD platforms, since the XMP profile fully corresponds to the information on the package: 1066 MHz and timings 10-10-10-30. The tRC parameter of 40 is too big, although it is not the most important, but it could be reduced to at least 36.

Goodram Pro default memory timings table

If, for some reason, you choose 1333 MHz, for example, the old chipset will not pull out more, then it is advisable to try to lower the timings as much as possible. It turned out that the Polish kit easily starts up at 1333 MHz with delays of 7-7-7-22, having passed all stability tests, but even after the final "polishing" of the secondary timings, it is unlikely that it will be possible to achieve high performance of the memory subsystem, and we will see this in comparative diagrams.

Goodram Pro minimum memory timings at which it is still stable

The situation with overclocking in frequency looks not so rosy. The computer starts perfectly at 2400 MHz with the set delays 11-11-11-28 CR2, but categorically refuses to pass the "physics test" in 3D Mark 11, and also hangs after 15-20 minutes of running the built-in 7-Zip benchmark. There is no point in raising the first CAS # Latency value to "12", yes, stability may appear, but in terms of speed it will be the slowest set.

At rated frequency \u003e Goodram Pro worked as expected and even allowed to slightly improve performance by reducing delays. The combination of 2133 MHz clock frequency with 9-10-10-26 timings became the best in all tests.

Excellent stability, fairly high read / write speeds, low latency at 2133 MHz are achieved with careful tuning of memory parameters in BIOS. The factory settings are too coarse and need to be adjusted.

The operation of the Polish kit at 1600 MHz did not raise any questions either, but even here it turned out to be possible to "twist the knobs" a little, the CAS # Latency parameter should be set to 8 instead of 9. This gives a sharp increase in bandwidth and reduces latency.

Apacer armor series

As mentioned above, the set of settings for the frequency of 2133 MHz was chosen by the manufacturer in vain. The 11-11-11-30 CR2 formula is clearly not for victories, but it is precisely it that is "sewn" in the form of an XMP profile. And the recommended voltage of 1.65 V is too high.

Please note that at 1600 MHz, almost the same timings will be set as at 2133 MHz. This will lead to a noticeable decrease in speed in processor-dependent applications.

It would be strange not to try to correct the situation, but they did not succeed much in this. It seems to be a successful start of the kit at 2400 MHz with the 11-12-11-30 CR2 formula, even passing the physics test in 3D Mark 11, but a repeated fatal error in the archivers says that we have climbed too high.

As a result, the results collected at 1600 MHz and 2133 MHz were taken into account, and in both cases we managed to slightly reduce the timings, which could not but improve the results.

The difference at the frequency of 2133 MHz is immediately visible, the upper result was obtained at nominal settings, the lower one - when the timings were set to 10-11-10-30.

At 1600 MHz, the memory passed all tests with delays of 8-8-8-24 and the results were included in the summary tables.

An interesting observation: the first test was rejected, because the modules were inserted into slots in a single-channel configuration by mistake, which caused Apacer's memory to consistently occupy the worst places in the tables. After the error was discovered, all tests were repeated, and the same RAM with the same working timings began to claim the top lines in the diagrams. The performance impact of two-channel mode is extremely high.

Corsair vengeance

Even without putting this memory on the stand, one can confidently predict victory for it. The timing formula indicated on the package is impeccable, there is practically nothing to add or subtract. But a detailed study of the contents of the SPD showed that, albeit in trifles, it will have to be corrected.

Judging by the numbers in the rightmost column of the SPD table, the kit is inoperable when the XMP profile is selected. Have not yet come up with DDR3 microcircuits capable of operating at 2133 MHz with CAS # Latency \u003d 7, and even at a reduced voltage of 1.5 V.

And now the fun part is overclocking. Without a noticeable increase in the supply voltage (to guarantee its value was 1.55 V), the Corsair kit started up at 2400 MHz and timings of 10-11-11-30. And all tests passed without errors!

Judging by the results of all AIDA tests, it is the Corsair kit that has the highest speed characteristics.

There were no complaints about the memory operation at 1600 MHz with timings of 8-8-8-24, and the bandwidth turned out to be very, very high. We can safely recommend memory Corsair vengeance enthusiasts.

ADATA XPG Xtreme Series

The first set of memory that can work without additional study of settings, if the motherboard allows you to use the XMP profile. The settings saved in the nonvolatile memory for the frequency of 2133 MHz are fully consistent with the recommended ones, and, what is most important, the computer starts up perfectly.

"Passport" of the ADATA XPG Xtreme Series kit. The memory will remain operational at any frequency, the timings can be safely left in Auto, if you do not try to squeeze out the maximum speed

Initially, attention was paid to the ADATA memory sticks. The combination of timings 10-11-11-30 at 2133 MHz seemed "strained" with the last bit of strength. But practice has shown that this is not entirely true, but there is some truth in this statement. And this was confirmed in experiments with the Command Rate parameter. By default, it is supposed to put "two", it will also be selected if you specify Auto in the CR settings. The ADATA memory worked with CR \u003d 1, but some of the computational tests failed even at a frequency of 2133 MHz, without any overclocking.

The Super PI 32M "calculator" was launched four times, but stopped with an error on pass 10-11. Neither lowering the voltage to 1.6 V nor raising it to 1.7 V improved the weather, so it was decided to leave CR \u003d 2, as recommended by ADATA

It turns out that these levels will not allow overclocking, since the parameters are extremely tightened? We thought so too, but in reality the situation turned out to be much better. A lot of surprise, we observed all tests passing at 2400 MHz, and the latency did not even have to worsen: the memory worked best with a combination of timings 10-11-11-30 CR2.

After small manipulations with the settings, the ADATA kit began to deservedly claim victory in today's competition

It turns out that the first assumption about the overclocking potential of the ADATA memory kit turned out to be wrong, there are considerable reserves hidden in these bars. Needless to say, 1600 MHz with 8-8-8-24 timings were played like clockwork, the tests were passed with very interesting results, which will be discussed a little later.

Geil DDR3 Evo Corsa

For Geil memory, the frequency of 2400 MHz is shown by the manufacturer as nominal. At least that's what the packaging and the booklet say. Let's see what parameters are recommended by the manufacturer and do they match the declared ones?

The XMP profile is written perfectly, and the frequency, timings, and supply voltage correspond to the declared values. Arguably, with the Geil memory kit, there will be no problems with the first launch.

But aren't the "12" values \u200b\u200blarge in the RAS # to CAS # and RAS # Precharge lines? It is quite possible to drop them by 1 point, only after that you will have to check the platform for stability, which can be entrusted to our set of tests. The result of timing selection is as follows:

Walk, so walk! The CAS # Latency value was also reduced to 10, which did not lead to system instability

Great result. This memory works ideally at 2400 MHz and the test results only confirm this. Unfortunately, the same cannot be said about stability and performance. Geil evo corsa at a frequency of 2133 MHz. To maintain the optimal balance of speed and reliability, the correct timings had to be looked for almost blindly and it took a lot of time. The tRAS parameter unexpectedly showed its essence. Until we set it to 28, the performance was very poor. The final results of tests at a frequency of 2133 MHz went to the results Geil evo corsa with parameters 9-10-10-28 CR1, almost like the memory of the "Corsair".

It took a lot of time to improve latency so much.

According to AIDA, the speed gain after overclocking is not as noticeable as we would like. It will be possible to see it only in comparative diagrams, in real work it is unlikely

Let's sum up the preliminary result. Geil has again got a perfect set of RAM, which allows you to try overclocking by all available methods, both in frequency and by reducing delays. With such superior performance, you can count on Gale to take the top spots in today's test. It's strange to use such a high-speed memory at 1600 MHz, but nevertheless we tested this variant too, having carried out a set of tests with timings of 8-8-8-24. This will provide an opportunity to compare Geil trims with their competitors.

Kingmax Nano Gaming RAM

Here we come to the most interesting memory kit. First, let's look at the default timings table and see its contents.

The XMP profile is registered correctly, but the high voltage of 1.8 V is alarming. Intel does not recommend setting it higher than 1.7 V to avoid damage to the memory controller in the processor

The first starts at a relatively safe voltage of 1.7 V ended in failure, the bench computer refused to start. Even raising the voltage to the maximum 1.8 V and higher to 1.85 V, it was not possible to make the memory work at a frequency of 2400 MHz.

Even our overclocker-tolerant motherboard found 1.8 V to be too much, highlighting the line in yellow.

After several dozen experiments with raising and lowering timings, voltage adjustments, the computer began to run stably at a frequency of DDR 2400 MHz, with timings of 10-11-11-31 CR2, but the timing values \u200b\u200bwere incorrectly displayed in the CPU-Z utility, and test runs utilities AIDA, Super PI 32M, archivers, showed that the results are very low, the same memory at a frequency of 2133 MHz is faster, so it was decided to leave in the table the results obtained at a frequency of 2133 MHz, and the test Kingmax Nano Gaming RAM do not run at 2400 MHz. Kingmax commented on this situation as follows: the company did not conduct a detailed test of this memory on the platform we used, and Kingmax does not guarantee the stability of the Nano Gamning RAM, and recommends using Kingmax Nano Gaming RAM on Intel platforms. However, we also noticed that there is no stability, and the two-channel mode is completely inoperative. We did not increase the memory voltage to 1.9 V, fearing for the health of the stand components.

The optimal timing formula for operating at 2133 MHz turned out to be 9-10-10-24. She also had to be handpicked. The already familiar "three eights" at 1600 MHz were given without effort.

Now for the wonderful chip coverage. At such a high voltage at maximum load (archivers, Super PI 32M), the temperature of the microcircuits reached 37 - 39 degrees. In a closed stand, the heating will be higher, but not fatal. Nanoradiators work, cope with their tasks, and what else is needed from them? You still can't see beauty in a closed system unit.

Feelings from Nano Gaming RAM remained ambiguous. On the one hand, this is a really fast overclocking (gaming) memory with modern chips. But in order to make RAM work at the rated frequency, you need to have a lot of experience in BIOS settings, at least in general terms to represent the RAM operation scheme and know the typical values \u200b\u200bof delays. Moreover, there is absolutely no guarantee that your memory will eventually work.

And with the supply voltage is generally a rebus. The supply voltage range of 1.5-1.8 V is indicated on the module packaging. According to the specification, 1.8 V is required to start at 2400 MHz, but with decreasing the frequency, the supply voltage can be reduced. All on a whim, since Kingmax did not leave us exact instructions.

DDR3 Memory Tests | Test results

In order not to clutter up the article with a huge number of screenshots and not force the reader to independently search for the necessary numbers, all test results were summarized in several diagrams, in which a certain color was assigned to memory kits from one manufacturer, selected by the color of the radiator or packaging. It is also necessary to note that in all diagrams the scale of values \u200b\u200bis relative, so that a twofold difference in the length of the bars does not mean a twofold difference in speed. Results are grouped according to the applications in which the memory was tested. And for each participant, his clock frequency and a set of basic timings at which he was tested are indicated. Deliberately incorrect results are not shown in the final diagrams, and the test results in the Super PI 32M benchmark did not go into action due to poor repeatability of the results.

Archivers

The built-in WinRAR benchmark gave preference to the Geil kit, and in general it is clear that a high memory frequency has a beneficial effect on performance. So almost all kits that were able to pass the tests at 2400 MHz were the winners.

At the bottom of the list, as expected, was Goodram memory at 1333 MHz with short timings, as well as Nano Gaming RAM, which worked in single-channel mode at 2400 MHz, which is why it was removed from the distance and removed from the diagrams. The results of the newcomer to the memory market, Apacer, are predictable, but pay attention to what a noticeable increase in Apacer's performance after timing selection! Another interesting conclusion can be drawn from the same diagram: if your platform does not support working with memory at a frequency higher than 1600 MHz, there is nothing to be upset about, the difference between the results at 1600 MHz and 2133 MHz is small. Having an outdated motherboard, it makes sense not to chase megahertz, but to shorten the delays.

7-Zip has different compression algorithms, so the results will be different. But the overall trend is similar. The best results are collected at 2400 MHz, and the positive effect of short timings (Geil, ADATA, Corsair) is very noticeable. Let's take a look at the results of the 7-Zip embedded test.

In the laggards again, the slow Goodram at 1333 MHz, which managed to bypass itself at 1600 MHz. Obviously short timings for 7-Zip are not critical. But then there is nothing to explain why Corsair did not take the first places, losing to Geil and ADATA.

In the unpacking diagram of the same built-in 7-Zip test, the results are even more interesting.

Outsiders are drawn to Olympus, and former winners are bored in the rearguard. Apacer at an average frequency of 1600 MHz took the third place, Goodram took the lead at 2133 MHz. It is absolutely impossible to understand the reason for such an arrangement of test participants, so it is worth looking at the table with the final test rating of 7-Zip.

The 7-Zip results are calculated in MIPS, which is an abbreviation of the English phrase Million Instructions Per Second.

Now, the situation has become noticeably clearer. High-frequency memory gains maximum "Mips", lowering the clock frequency leads to a uniform decrease in performance. The results show the successes of ADATA, Geil, Corsair, Goodram memory. Kingmax is not among the leaders, having consolidated in the middle. Apacer does not take the first places, short timings at 1600 MHz turned out to be more preferable for its microcircuits. Anyway, the effect of delays is noticeable. Look at Geil evo corsa , what is the gap at 2400 MHz with timings 11-12-12-30 and 10-11-11-30. Here you can clearly see which memory at a frequency of 2133 MHz copes with a bang, and which is trying with all its might.

AIDA: cache and memory test

This benchmark allows you to evaluate the performance of the memory subsystem in read / write / copy operations, as well as measure latency, which strongly depends not only on the memory used, but also on the DDR controller built into the CPU. The lower the latency, the noticeably higher the computer's results in all tests. The AMD platform has a latency of 45 - 60 ns, the latest Intel processors demonstrate 38 - 45 ns in tests. There are several ways to reduce the latency: by overclocking the CPU, shortening the timings, and increasing the memory frequency. Since we are not touching the processor parameters today, the latency will depend solely on the properties of the chips and the set memory parameters. But we will not start the analysis with the latency, but with the linear reading test.

We have already seen something like this. It is possible that AIDA was the most truthful in terms of memory performance. But we figured out the obvious leaders in advance, it is more interesting to look at the lagging behind. There we can see that Apacer's operation at 1600 MHz is not ideal, but if we go to 2133 MHz, the results are clearly better. The same picture with "Gale". But Kingmax Nano Gaming RAM clearly does not break into the leaders, taking only the sixth place at 2133 MHz.

We can see the same trend in the write speed test.

And again ADATA, Geil, Corsair won prizes, but look how high Goodram has climbed (1333 MHz)! Of the other kits at a low frequency of 1600 MHz, the ADATA kit still works well, our "dark horse" is very confident, if not climbing to an inaccessible height, then occupying very prominent places. A Apacer does not add much in speed with shorter timings, but it was the first in the "2133" group in the nominal mode. You can also immediately see the division of all participants by clock frequency.

And here is not the most obvious copy test, in which the contents of one cell are overwritten to another. Its impact on the performance of real applications is controversial. Let's not jump to conclusions, just see what happened.

As they say, "all the same faces." ADATA is in the lead along with Corsair and Geil, Apacer even at 2133 MHz at 10-11-10-30 timings is not faster than with 11-11-11-30 timings. In class "1600" ADATA performed on an equal footing with Apacer, and Goodram did not bring any surprises this time. From the tests already done, everything is quite clear, but we still have a lot of interesting things in store. Finally, we got to the latency test. Since the lower the latency, the better, the diagram is inverted, but thanks to this, the winner should still be looked for at the bottom of the diagram.

Goodram DDR3-2133 was again included in the list of winners, but competitors with a frequency of 2400 MHz did not give it to the top. In addition, the work of the Polish kit at lower frequencies was surprisingly slow. The memory of Apacer performed well, its three blue bars are evenly distributed in the diagram. Kingmax at 2133 MHz, outperformed Geil by two points with almost the same delays.

SiSoftware Sandra Cryptography & Encryption

Let's take a look at the Pivot Chart. It can be seen that the results for almost all participants are similar, it looks like an array of columns of almost the same length.

And again we see a clear ranking of memory kits. High-speed modules occupy the best lines; with decreasing frequency, they also show a uniform decrease in performance. Apacer, Goodram, Kingmax are an exception to the general rule. The first one pulled out far ahead (the test was carried out 4 times with the same result).

Goodram lost to all competitors (although, as we found out, judging by the latency, it is very good), and Kingmax DDR3-2133 never woke up and remained in the middle. Let's see if the same pattern repeats in the next SiSoftware Sandra benchmark?

Yes, it turned out almost like a carbon copy. The nature of the task being performed has changed, but again we see a dense block of the most productive participants and several lagging sets.

PC Mark 7

Let's see if PC Mark's comprehensive benchmark suite has felt the change in memory performance.

If you look at the distribution of results, you can see that PC Mark also prefers high-speed 2400 MHz memory, it is also sensitive to short timings, precisely for the latter reason Kingmax DDR-2133 9-10-10-24 bypasses Corsair DDR3-2133 9-11- 10-30. But as with all other tests, short latencies do not help Goodram DDR3-2133 9-10-10-26 to emerge as a test winner. From the diagram, we can conclude that the memory frequency and its latency have the most direct impact on the speed of content processing. Let it be five percent, but a computer with high-frequency memory will be faster. And this factor, among other things, will help save the precious working time of professionals.

The next test is also very indicative; it is a test of the speed of Internet tasks.

A different color distribution in the diagram is immediately noticeable, of the recognized leaders, only Corsair remained on the podium, and the prize lines were occupied by former outsiders. Anyway, all memory with a frequency higher than 2133 MHz is "rejected". There are no objective reasons for such a finale, rather it looks like a random result, so we can talk about the absence of the influence of the frequency and timings of RAM on the computer's performance in typical Internet tasks.

It makes no sense to give one of the diagrams. According to the results of the graphics test, we could not find a direct influence of the RAM frequency on the FPS, which is not surprising. Typical values \u200b\u200bin Graphic Test 4 with the "Performance" presets are 42 - 44 fps, and we didn't manage to achieve an approximately equal result even in ideal conditions, so we made a note to ourselves and excluded the results of this test from the final.

But there is definitely a direct dependence of the speed of calculating physical models on the parameters of RAM. Moreover, the physics test in 3D Mark 11 is a great option to test your computer for stability.

Very, very revealing result. A noticeable increase in the results with high-speed memory, if you look closely, you can see that memory kits with the same clock frequency gain approximately the same number of points (the results are located in blocks), timings in this test almost do not play an important role, but in fairness I must say that the effect of delays is all -and there is. It turned out that in the physics test, there is practically no difference between the memory with a frequency of 1600 MHz and 2133 MHz, if the first has the right timings. So gamers have a lot to think about.

DDR3 Memory Tests | conclusions

Today's testing of DDR3 memory, carried out in the THG laboratory, turned out to be very voluminous, but it allowed not only to answer the question "whose memory is better", but also to reveal unobvious relationships between the characteristics of memory and its performance. But it is better to first summarize the impressions of all the test participants one by one.

\u003e Goodram Pro , in general, should have been out of competition, the reason is the volume of 2 GB per bar, and not 4, as in competitors. Moreover, it was not possible to overclock it to 2400 MHz. Among the positive characteristics, you can include work with a supply voltage of 1.5 - 1.6 V, the possibility of overclocking by reducing delays. These modules performed well at clock speeds of 1333 MHz, 1600 MHz and 1867 MHz. At 1867 MHz, we managed to limit ourselves to 8-9-9-24 timings, which is very good. We can recommend this memory to owners of AMD platforms. Goodram is efficient at 2133 MHz, but its results are rather low, so it is worth choosing other modules to equip modern Intel platforms.

Apacer armor series can be considered a rather strange debut of the company on the Russian market. The advantage of this kit should have been a low price, but we still don't know it. But the shortcomings have already been discovered: a poor quality thermal interface, with poor adhesion, average performance, poor overclocking capability. However, everything is said above, there is information in the diagrams. For final conclusions, there is not enough information on the cost of the kit Apacer armor series , but it is clear that for computers "for every day", such as office or multimedia, it is enough, and it should not fail.

Corsair vengeance - the clear leader of today's test, won us over not only with its excellent, and most importantly, simple overclocking, but also very high results in tests both at 2133 MHz and at non-standard 2400 MHz. If you spend a certain amount of time and properly align the sub-timings to optimal values, you can get even more pleasant results. The disadvantages of the Corsair kit can only be high radiators, the need for which is not obvious. This fast memory will appeal primarily to gamers and overclockers, especially if their platform allows 2400 MHz memory.

ADATA XPG Xtreme Series can be called the discovery of today's test. The ratings are neither successful nor advanced, but in fact this " simple memory"It turned out to be overclocking, easily taking 2400 MHz, while showing the best results in almost all tests. We could not identify any drawbacks. The peculiarity of ADATA bars is that they require control of the Command Rate parameter in BIOS, reacting very negatively to" one ". Better put "2." right after the first turn on. The DDR3 kit from ADATA is well suited for compact computers, this memory has very compact heatsinks.

Geil evo corsa - another today's leader on a par with "Corsair". At a lower cost, this memory has a higher nominal clock frequency, excellent characteristics, it is suitable for serious overclocking by reducing timings. Of the shortcomings, it is not worth highlighting high productivity at 1600 MHz, so this memory is more likely for adherents of Intel platforms. To get good results, you don't have to spend hours adjusting the settings, the XMP profile is written quite correctly. The disadvantage is the same as that of the Corsair - a tall radiator. This will not prevent the memory from getting our recommendation, but it will block its way into HTPC and embedded systems.

Kingmax Nano Gaming RAM - innovative memory - turned out to be a very capricious person. Firstly, its "native" frequency is 2133 MHz, not 2400 MHz. There is no doubt that after a long "polishing" of the timings it will be possible to make it work at 2400 MHz, but the parameters recommended by Kingmax are not optimal. This DDR3 kit can be recommended for trained users who are able to independently conduct a set of tests and adjust parameters for maximum performance. The greatest bewilderment is caused by the supply voltage at a frequency of 2400 MHz - 1.8 V or does it still need 1.9 V? Other participants were satisfied with 1.65 volts, while Corsair and Goodram were content with 1.5 volts.

Does it make sense to add high-speed memory to your computer? Of course, yes, when it comes to assembling a powerful system unit. The difference in memory test results at 1333 MHz and 2133 MHz is very noticeable. But even in the event that the maximum frequency of the memory controller in your computer does not exceed, say, 1600 MHz, with good memory, fine-tuning it, you can achieve good results in this case, which is clearly confirmed by the test diagrams.

The question of comparing DDR4-2133 vs DDR4-2400 has been haunting us since winter, when Intel Kaby Lake processors with DDR4-2400 support appeared, and many at least had a desire to overclock the existing DDR4-2133 modules or buy the most budgetary DDR4-2400 for this. This does not mean that higher frequencies are not supported, just anything higher is already overclocking, which basically requires motherboards based on Intel Z170 or Z270 chipsets, as well as processors with an unlocked multiplier.

If you have an assembly on Intel H110, B150 or H170, for which you bought a Kaby Lake processor, then after bIOS updates it must support DDR4-2400 memory, because the RAM controller is located directly in the processor itself. Therefore, if you have a motherboard from a reputable manufacturer, then you shouldn't have any problems after flashing: just go into the BIOS and select the desired memory frequency, as we did on the ASRock H110M-HDS motherboard. The selection of timings can be done independently or you can trust the system.

And to assess the usefulness of such an action, we decided to conduct four short test sessions. The first two will be on a budget system that includes an Intel Pentium G4560 processor, 8 GB of RAM and an integrated Intel HD Graphics 610 graphics core or a SAPPHIRE NITRO + Radeon RX 470 4GB graphics card. The other two will use a more powerful configuration based on the Core i7-6700K and 16 GB of RAM. In this case, either the integrated Intel HD Graphics 530 core or the Colorful GTX 1060 SI-6G video card will be responsible for graphics processing.

Building on Intel Pentium:

• Intel Pentium G4560
• ASRock H110M-HDS
• SAPPHIRE NITRO + Radeon RX 470 4GB
• 2x4GB DDR4-2133 Transcend TS512MLH64V1H
• WD Blue Desktop 2 TB (WD20EZRZ)
• AeroCool VX500 500W
• GameMax MT521-NP
• AVerMedia Live Gamer HD
• AVerMedia Live Gamer Portable 2

System on Intel Core i7:

• Intel Core i7-6700K (OC 4.5 GHz)
• Thermaltake Water 3.0 Riing RGB 240
• ASUS MAXIMUS VIII RANGER
• 2 x 8 GB DDR4-3200 G.SKILL Trident Z
• Colorful GTX 1060 SI-6G
• SSHD Seagate ST2000DX001 2TB
• HDD WD WD1000DHTZ 1TB
• Seasonic Snow Silent 1050 1050W
• Thermaltake Core P3
• ASUS VH228H
• AVerMedia Live Gamer HD

So let's start with the Intel Pentium G4560 and integrated graphics. DiRT Rally in HD resolution with a very low graphics preset provides very comfortable performance in both cases. However, when moving from DDR4-2133 to 2400, the average FPS rises from 95 to 98 frames per second, and the minimum - from 75 to 77. The difference is 3%.

Rainbow six siege is already heavier, so even with a low profile in HD resolution, the difference in absolute performance is almost not felt, because it is less than 1 FPS. If we go to relative units, then the system with faster memory wins 0.4% in average and loses 6% in minimum.

For start Far cry primal I also used low graphics preset and HD resolution. In both cases, we got a slideshow, but already with a margin of 1 FPS in favor of DDR4-2400 memory: 13 versus 12 frames / s in the average and 12 versus 11 in the minimum. In relative terms, this is about 8-9% of the bonus.

Now let's move on to the second session, where we added an RX 470 series video card to the Intel Pentium G4560. And the memory overclocked to 2400 MHz was done in two ways: with automatic timing selection, which in our case corresponds to the CL17 scheme, or with the CL15 scheme manually setting. Let's see if there is any benefit to this in Full HD resolution.

The first surprise awaited us in DiRT Rally with a very high graphics preset. According to the test results, a system with DDR4-2133 memory unexpectedly took the lead with an average frequency of 101 FPS and drawdowns of up to 78. In second place was the configuration with 2400 memory and CL15 latencies, which produced an average of 99 frames / s with drawdowns of up to 68. Results of the third configurations were 98 and 67 FPS, respectively. The tests were run several times, but the trend continued.

High profile graphics settings in Rise of the Tomb Raider leads to a change in the leader in terms of the overall average, although the difference between the first and third places does not exceed 1 FPS. Nevertheless, this test turned out to be behind DDR4-2400 memory with CL17 timings, the second result was shown by DDR4-2133, and the third - by DDR4-2400 CL15. At the minimum, the practice corresponds to the theory: that is, more high frequency and low latencies provide better performance.

Benchmark Far cry primal at ultra settings, I agree with DiRT Rally that the combination with DDR4-2133 memory should be in the lead with an average result of 54 frames / s and drawdowns up to 43. But the timing scheme does not play a special role for it, so both competing systems with DDR4 -2400 showed the same results: 49 on average with drawdowns up to 38 FPS.

The last in this session will be Ghost recon wildlands with high graphics preset. On average, the best 60 FPS result was obtained by a system with DDR4-2400 and CL15 latencies. In second place with 59 fps was the bundle with DDR4-2133, and the third result was behind DDR4-2400 CL17. In terms of the minimum indicator, the opposite is true: the best result is for DDR4-2400 CL17, the worst is for DDR4-2400 CL15.

In the third part, we switched to Intel Core i7-6700K and limited ourselves to the integrated Intel HD Graphics 530 adapter. Therefore, the tests were passed in HD resolution.

DiRT Rally at medium settings, it gave a clear preference for faster memory: 44 versus 41 FPS in the average and 38 versus 35 in the minimum. The difference tended to be 8%.

Rainbow six siege at a low preset, I agree with this state of affairs, but already more modestly estimates the advantage of DDR4-2400 memory: the average growth rate was about 4%, and the minimum - about 2%.

But For Honor with a low profile, I could not determine a clear winner: in terms of the average, the system with faster memory took the lead, and the configuration with DDR4-2133 left the lead in terms of the minimum. Although in both cases the difference did not exceed 1 FPS.

And the final test session was held using a bundle of an Intel Core i7-6700K processor and a GeForce GTX 1060 video card, so the resolution was increased to Full HD.

Rise of the Tomb Raider at a very high preset gives minimal preference to a system with faster memory on average. At the minimum frame rate, the difference between DDR4-2400 and 2133 already reaches 9 frames / s or 34%.

But Far cry primal at ultra settings I decided not to take sides, giving both systems the same results: 67 FPS with drawdowns up to 53. Only the percentage of memory used and the maximum FPS indicate a slight advantage of DDR4-2400.

And in Ghost recon wildlands with a very high profile, we again see a minimal advantage of faster memory: the difference in average and minimum speed does not exceed 1 FPS. But at the same time, the average load on the processor in the case of DDR4-2400 decreased by 4%.

Outcome

As a result, we see that in terms of gaming performance, overclocking memory from DDR4-2133 to 2400 on Intel platforms does not always give a tangible result. In a budget system, you can count on a positive effect only when using integrated graphics. Yes, the gain can reach 8-9%, but the HD Graphics 610 doesn't play much with demanding projects. And if you add a discrete video card to such a system, then a higher result in some tests can be obtained with DDR4-2133 memory, not 2400. With lower timings, you can also not always count on a noticeable bonus.

In a productive configuration based on Intel Core i7, practical tests are more in line with theory: in almost all cases, a bundle with faster memory looks better, but the difference is still small. With integrated graphics, we got a bonus of 3-8%, but for this we have to lower the graphics settings and switch to HD resolution. With a discrete graphics card, the gain is mostly at the level of measurement error, if you do not take into account the result in Rise of the Tomb Raider.

Article read 21827 times

Subscribe to our channels