Impact of DDR5 memory characteristics on Ryzen 7 7800X3D performance

This short study focuses on comparing the performance of two configurations using different types of DDR5 memory on the Ryzen 7 7800X3D processor. One configuration uses DDR5-5200 with timings of 40-40-40-76, and the other uses DDR5-6000 with timings of 36-36-36-76. The differences between them are throughput and latency. The first configuration is considered less productive and more affordable, while the second is more productive, but has a price around or above average.

Using a pre-tested Ryzen 7 7800X3D processor, we replaced only the memory and ran the same tests as in regular processor reviews. The graphs of the results of the Ryzen 7 7800X3D with two different memory modules will also show the results of the top gaming model Ryzen 9 7950X3D. This will allow you to evaluate how fast memory can bring the performance of the Socket AM5 platform to its maximum. This pair of processors will also be interesting from the point of view of the difference in the number of cores by a factor of two. Some benchmarks for the 7950X3D may use memory to achieve maximum performance potential.

We also compared the results of two 7800X3D variants with a similar CPU model without additional L3 cache — the Ryzen 7 7700X.

The choice of the previous generation AMD video card was made more than six months ago. At that time, there were no new video cards of the Radeon RX 7900 and GeForce RTX 40 series. The Radeon RX 6800 XT delivers ample performance at lower resolutions and delivers faster rendering speeds under CPU pressure than competing Nvidia graphics cards from the same era, which require more GPU processing time.

Memory and caching performance

Although the efficiency of the DDR5 controller in modern AMD processors is somewhat lower than that of Intel solutions, the bandwidth of this type of memory is still noticeably higher than that of DDR4. And since the real memory bandwidth for the Ryzen 7000 turned out to be lower than expected, it is quite possible that increasing it by increasing the DDR5 frequency from 5200 to 6000 (+15% in theory) will bring some benefit both in specialized synthetic memory tests and in real ones tasks. The AIDA64 package showed the following:

It can be seen that the difference in memory timings led to a decrease in the access delay to RAM by about 15%, so for simplicity we can conditionally assume that both in terms of bandwidth and delays, the maximum difference in real tasks can reach this value. As for the throughput indicators, the write speed actually turned out to be +15% of the bandwidth in this synthetic test, but the read speed is already only 2.5%, so it’s far from a fact that in software and games it will be so big growth. Let's look at the PSP indicators in a more convenient form:

The use of more powerful DDR5 memory in the system with the Ryzen 7 7800X3D led to an increase in the write speed of information to 81 GB/s, which is much more than 69 GB/s for the 16-core top-end Ryzen 9 7950X3D, and this is explained precisely by the increased operating frequency of DDR5 -memory. But the increase in memory reading speed did not lead to a strong increase — this figure for Ryzen 9 remained even slightly higher — most likely this is due to the internal organization of AMD processors, since the Infinity Fabric bandwidth is limited to 64 GB/s.

Obviously, access latencies to the cache memory, as well as its throughput, did not change in any way due to the acceleration of DDR5 memory. But purely to check and confirm what has been written, let’s also consider the results of the bandwidth test of all levels of cache memory from the AIDA64 package.

Actually, this is how it all turned out — the bandwidth of all levels of cache memory in the low-end model is clearly lower than that of the top-end model — almost exactly half, and this is explained by the smaller number of computing cores, since the total memory bandwidth for all cache blocks is shown. And there is no difference between the Ryzen 7 7800X3D with two memory configurations, as it should be.

Purely synthetic benchmarks from packages like Sandra and AIDA64 can be interesting for assessing low-level performance in specialized tasks, although they claim some versatility. The first group of tests shows relative performance in different tasks and a certain CPU Overall score calculated from all the results.

And this is where things start to get interesting. In this set of tests, the Ryzen 7 7800X3D with different memory performs slightly differently. Yes, on average the difference is small — only about 2%, but in the cryptography subtest the result was already 4% higher, and in scientific calculations there was a whole 10% increase! Of course, this did not allow us to catch up with the 7950X3D, but the increase is quite noticeable. As we mentioned earlier, these subtests are quite demanding on the speed of the memory subsystem and receive decent gains from memory and cache acceleration. Let's see what happened in the remaining subtests from the same package:

The remaining subtests show computing performance when processing media data, and in them, memory accelerated in all respects did not give any increase at all, unfortunately. We previously noticed a lack of acceleration from improved cache performance in these subtests, and the lack of focus on memory subsystem performance has been confirmed today. Let's also consider tests from another universal package.

These are also purely synthetic tests that show performance in tasks with a certain specialization. For example, CPU Queen uses integer operations when solving a classic chess problem, and AES uses the encryption speed using the cryptographic algorithm of the same name:

In the first couple of synthetic tests, improving RAM performance did nothing to help the Ryzen 7 7800X3D processor speed up, and we again note that these conditions did not show a noticeable relationship with the speed of the memory subsystem in previous tests — for example, adding cache to X3D processors did not lead to acceleration of similar computing tasks. The more powerful DDR5 memory didn't help either.

The first two subtests in the diagram above use integer operations for image calculations and data compression, while SHA3 is another cryptographic algorithm. And when the Ryzen 7 7800X3D received DDR5 memory with the best characteristics, it was able to noticeably accelerate only in the first subtest — the increase was a significant 11%. But in the second and third subtests there is no difference at all. It is interesting that an advantage from an increased cache memory is observed only in the first of three subtests — the memory subsystem also includes cache memory, and if any of the components is accelerated, then the overall result improves.

Well, the most numerous set of subtests from AIDA64 includes performance tests of floating point operations, including instructions for all SSE and AVX/AVX2 variants. AMD processors with additional cache did not perform better in these tests compared to their conventional counterparts, so we were not surprised that the Ryzen 7 7800X3D with 15% faster DDR5 memory also showed no gains at all.

Another synthetic test that we decided to include in this section is that in terms of the load on the cores it is closest to the rendering tests, and it is also very convenient to compare single-threaded and multi-threaded processor performance. In the case of Ryzen 7000 processors, a variant of the AVX-512 test was used, which allowed for a slight increase in performance compared to other CPUs.

Since this test loads the processing capabilities of the CPU with work, and in previous articles we did not note acceleration from the increased cache memory, we were not surprised by the complete lack of acceleration from the improved characteristics of more powerful DDR5 memory modules. The difference in performance does not exceed the measurement error and is not even in favor of the configuration with accelerated memory. We suspect that even with a multi-threaded load, absolutely nothing will change.

This is true, and in multi-threaded performance the Ryzen 7 7800X3D with 15% faster memory did not improve its performance. So far, everything is confirmed by previous studies — the acceleration is mainly only where there was a benefit from adding an additional crystal with a third-level cache to Ryzen processors, but where the 7800X3D was even inferior to the similar 7700X without an additional crystal with a cache, and from improved DDR5 -there is no increase in memory.

Let's move on to not entirely synthetic tests that measure system performance in several types of application tasks, at the same time displaying an average value showing overall performance, like the PCMark 10 package. This approach also has its advantages (ease of evaluation using a single value for an entire software area) and disadvantages (they try to cover too much and do it imperfectly), but most often processors are still tested in it.

We have long understood from previous materials that most PCMark subtests do not use a large number of threads, do not load the memory subsystem too much, and overall performance depends little on the number of cores. Most often, the overall speed here depends on the CPU frequency, and even then all modern high-performance models hit a certain limit, and even in the gaming subtest it can be difficult to find the difference. Not surprisingly, two Ryzen 7 7800X3D configurations with different memory showed almost identical results.

Another general performance test we looked at was the 3DMark CPU Profile, which relates more to game rendering speed. In this subtest, the Ryzen 7 7800X3D is inferior to the “cacheless” 7700X processor, so we did not expect any acceleration from installing faster DDR5 memory. And so it happened: neither the additional cache memory helped, nor the improved frequency and RAM latency were able to increase performance. Purely number crushers are important here; everything else does not provide an advantage.

Let's look at a few more processor tests from the 3DMark suite — most often these are physical calculations that can use multi-threading, but with varying degrees of efficiency. And in them we saw a very small increase of 3%-4% in a couple of tests — the second and third. By the way, the additional cache in the 7800X3D showed a strong result compared to the 7700X only in the first of them — Time Spy, but not in Night Raid. However, a difference of only a few percent does not allow us to draw far-reaching conclusions.

The last test in this section is the JetStream 2.0 browser benchmark, which measures the performance of JavaScript and WebAssembly code. For tests, we used an updated version of Microsoft Edge on the Chromium engine. Judging by past research, this test also does not make much use of either multithreading or the capabilities of the memory subsystem; everything in it depends on the clock frequency of the CPU itself. As a result, the Ryzen 7 7800X3D in both configurations is slightly behind the 7700X, which also confirms our conclusions — neither additional cache memory nor faster DDR5 memory will bring you anything in such tasks.

Rendering tests are some of the most difficult for modern processors due to the multi-threaded nature of the ray tracing load — modern processors try to maintain the highest possible frequency, which can consume a lot of power and get very hot. But from the point of view of today’s material, they will turn out to be useless; most likely, memory does not work very actively during rendering. While cache memory sometimes affects overall performance, bandwidth and RAM latency have a very slight effect on the results. Let's take a look at the Cinebench benchmark, beloved by AMD and Intel, which is currently the penultimate version.

Everything is as we thought: judging by the results of the Ryzen 7 7800X3D in two configurations, as well as against the 7700X, in such tasks there is no point in either additional cache or accelerated RAM. A purely numerical problem receives an increase only from the number of computing cores and their clock frequency; everything else does not limit performance in any way. We don’t think that anything will change in the other tests in the section, but we are obliged to check them.

Test scenes in Blender show slightly different results, but in general everything is the same there — the advantage of the 7700X over the 7800X3D remains, the additional third-level cache does not help rendering, and the differences between the two configurations with DDR5 memory are different (by as much as 15 % in theory, by the way) the characteristics are not visible.

Another rendering test is Corona, which measures the time it takes to render one frame. It would seem that there is finally a difference of a whole second in 3D rendering acceleration for the Ryzen 7 7800X3D with DDR5-6000 — but this is only a little more than 1% increase, which can not even be considered a difference, because it may also be a measurement error — even more so Moreover, the test gives results in whole seconds. But since in this test the additional L3 cache allowed the 7800X3D to slightly improve performance compared to the 7700X, we will assume that all this is not accidental.

Well, the latest 3D rendering benchmark is VRay, it measures the speed of image rendering for three scenes at once. Its results repeat everything that we saw in previous tests of the section, but although the Ryzen 7 7800X3D was practically not inferior to the 7700X model, that is, there was some acceleration from the additional cache memory, there was no gain from installing DDR5-6000 instead of DDR5-5200. It is clear that the top AMD solution is exactly twice as productive here — because it has twice as many computing cores, and memory bandwidth does not limit their capabilities in any way.

The next test section examines several programs for processing media data — photos and videos. These are already quite practical tasks, like exporting hundreds of high-resolution images in RAW format of about 3 GB in Adobe Lightroom Classic — most serious photographers do such tasks on a regular basis.

In the photo processing test, the junior AMD processor with an additional cache was inferior to its cacheless counterpart, the 7700X, quite a lot, so specifically in Lightroom, the large L3 cache did not help at all. And the faster DDR5 memory, both in terms of memory bandwidth and latency, was also unable to speed up the process of processing photos in this software. So users of applications for creating digital content don’t really need additional cache, just high-performance memory. The main thing here is the computing capabilities of the cores and software optimization in specific software.

But in the video editor of the same company it turned out a little differently. We tested the rendering of a not too complex project in Full HD and 4K formats — many people face this task when preparing an edited video for streaming video services, so the situation is also vital. The Ryzen 7 7800X3D is only slightly slower than the 7700X without cache in this test, and given the 7800X3D's lower clock speed, it's more likely to help in this task. And the memory accelerated by 15% in this Adobe package gave, albeit a small, but still tangible increase — plus 2% -4%.

The next Handbrake test is a package for converting video data to other formats. We used the input video in H.264 format and transcoded it into H.265 format — also a fairly common task that modern users have to solve. Unfortunately, the additional cache of the Ryzen 7 7800X3D did not allow it to show decent results compared to the 7700X (the clock speed of a crystal with an additional cache is lower than that of a regular one), and more powerful memory did not help improve the overall data processing time.

The second video transcoding test is SVT-AV1, but this time it encodes video data into the AV1 format, a relatively new open standard. And this time we see exactly the same outcome — nothing depends on memory bandwidth and latency, as well as on additional cache memory in the Ryzen 7 7800X3D, which is about the same behind the 7700X as in the previous test.

The last test in this section is Topaz Video Enhance AI. The application is used to improve video quality using artificial intelligence capabilities. This very compute-heavy task uses Artemis High Quality upscaling from Full HD to 4K, and we had little hope that better-spec DDR5 memory might speed things up, but alas, there's no difference between the two sets of modules. Once again we did not find any characteristics differing by 15%. And this despite the fact that in this test the eight-core processor without additional cache was almost on the same level as the 7800X3D — that is, the L3 cache still helped a little.

Another important area of processor performance testing is cryptographic tasks. Modern CPUs can encrypt large amounts of information literally on the fly, and some even support special instructions for common algorithms such as AES. The first test is John The Ripper, a free software for recovering passwords using hashes that can take advantage of all the capabilities of modern processors.

Even in previous tests, we realized that everything here is decided by the number of computing cores and a powerful architecture with a maximum clock frequency, and the cache and memory are not particularly important. Actually, this is why an eight-core processor with an additional cache is much inferior to an analogue without it, and a 16-core processor is almost twice as fast. Alas, no difference was found between the two configurations with different DDR5 memory in this test either.

VeraCrypt is on-the-fly encryption software that uses several different data encryption algorithms and can take advantage of hardware-accelerated encryption on the CPU. In the tests we used a 1 GB buffer and… again we got the expected result — the more productive DDR5 memory did not give any advantage to the test system in this software. There is simply no difference between the two configurations, and the similar 7800X3D processor is significantly behind the cacheless Ryzen 7 7700X — the cache doesn’t help here either.

The latest cryptographic test is a CPU mining program that also uses cryptographic calculations, it is very well optimized for execution on modern CPUs. For tests, we chose the x25x algorithm, used in some cryptocurrencies, and for comparison we took the best result from several optimized miner options using instruction sets: SSE2, AVX2, AVX-512, as well as hardware support for AES and SHA.

And here we have already seen, albeit a very small, but still a clear increase from installing faster memory — DDR5-6000 turned out to be up to 2% -4% faster than DDR5-5200. At the same time, in the SSE2 subtest there is no practical difference between the 7800X3D and 7700X, but in the more productive versions of the test it appeared, although not too significant.

Compressing and decompressing data in archives is known to most users, as are the most prominent representatives of advanced modern archivers, one of which has been WinRAR for many years. We used the benchmark built into the archiver — it measures the maximum data compression speed.

Perhaps, we pinned a decent portion of our hopes on this section in the hope that the benefits of quick memory would emerge. Moreover, AMD processor models with additional cache received a significant increase in the built-in WinRAR benchmark, and the same 7800X3D was very significantly ahead of the 7700X model without an additional cache memory chip. The results of this test are greatly influenced by the memory subsystem as a whole, so we were not surprised by the increase from installing more powerful DDR5 memory — even if it is not 15%, but “only” 5%, but this difference is noticeably higher than the average.

The second archiver, 7-zip, may be somewhat less popular, but it is interesting because it supports a more efficient and demanding compression method. The advantage of the 7800X3D over the 7700X is only in data compression, and it is small. And we were even a little surprised that the 7800X3D configuration with DDR5-6000 brought a 6% increase compared to the lower-performance memory that we usually use in tests. So there is an advantage in this archiver, but only when compressing data, but when decompressing, the memory characteristics again turned out to be unimportant.

The section is not the most voluminous — we included Y-Cruncher, a program for calculating pi, as a conditional mathematical problem. Of particular interest to us is this program’s support for the AVX-512 instruction set, as well as the optimization of this software specifically for Zen 4 in the latest version, which we used. Let's check how the authors did it:

We tested calculating a billion digits of Pi in single-threaded and multi-threaded modes with DDR5-6000, but it didn't make much sense — the Ryzen 7 7800X3D in two configurations showed very similar results, and memory with improved performance characteristics did not give anything at all. It is also interesting that a 16-core processor with a cache is not so far ahead of all 8-core processors and is not very efficient, but this issue is no longer relevant to today’s material.

The built-in benchmark in Matlab can hardly be considered an indicative test, since it is quite outdated and runs too quickly on modern CPUs, and its results fluctuate greatly from one run to another, so they are almost meaningless. Moreover, the difference between the two sets of DDR5 memory would not exceed a few percent in any case. This time we were simply convinced again that the results simply do not allow us to draw meaningful conclusions and better analyze the scientific calculations section of the 2020 test methodology, which also includes a longer and more revealing test in Matlab.

As for processor performance in games, you can find a large study on this topic at the link in the sidebar to the right of this text. In it we examined a large set of CPUs of different levels in terms of performance and price, ranging from old six-core processors to the most modern top-end processors. But in general, we note that in modern games, most often there is not much difference between a 6-core and a 16-core processor with the same frequency, because even six fast cores are still quite sufficient for most games, and the most important characteristic of a CPU for games remains performance at tact.

A large amount of fast cache memory also has an effect, but not all games benefit from the increased amount of L3 cache. In the best case, the benefit from the additional cache was in about half of the projects, and in the remaining games we observed either equality or a slight advantage with the higher-frequency “cacheless” Ryzen 7000 processors. Detailed tests are available at the link, but today we are looking only at the average data for our a test set of a dozen games of various genres and pay full attention specifically to the difference between two Ryzen 7 7800X3D configurations with two different sets of DDR5 memory.

It is immediately clear that in Full HD resolution and with average graphics settings, even relatively inexpensive and slow processors like the Core i5-13600K show, although lower performance compared to top-end CPUs, but the average difference between them is small — less than 10%, and with a frame rate of more than 200 FPS on average, this difference is not very important, in fact.

But today we're mainly interested in a couple of 7800X3D configurations and whether the boost from higher-performance DDR5 memory can help it close the gap to the 16-core flagship. So, the average difference in average FPS between DDR5-6000 CL36 and DDR5-5200 CL40 turned out to be… less than 1%! Well, according to the minimum indicator, more, but not much. That is, technically, such a system is closer to the Ryzen 9 7950X3D, since even with slower memory they are separated by only 3%, but specifically the more powerful memory gave only 1%.

It is important to note that the speed increase from accelerated DDR5 memory in the games we tested varied. Most often it was not there at all, but there were games that won more than others, these are Watch Dogs: Legion and Hitman 3 — in them, installing faster memory gave a plus of 3%-4%, which is already decent. It is also interesting that these games are among those that also received the greatest boost from the increased L3 cache in the Ryzen 7000X3D, so our thought is confirmed — most often the software that benefits from a larger L3 cache also It also gives better performance when installing high-speed memory.

We present the results in a resolution of 2560x1440 with maximum rendering quality only nominally, since it is obvious that there will be nothing left of the 1% difference in the previous tablet. In more complex conditions, almost all CPUs are conditionally equal in terms of gaming performance, and accelerated memory gives practically nothing on top, since everything is limited by the performance of the video card in any case. For games in such realistic testing conditions, processors of the Ryzen 5 and Core i5 level are quite sufficient, since the difference between the weakest and fastest CPU in this mode will be only 2%-3%, and you will not notice this difference without special tools.

«When installing higher-performance DDR5 memory, especially in a Ryzen 7000-based system, there are some considerations. Memory speed is an important aspect because it provides data flow for computing devices. However, increasing memory speed does not always lead to significant performance gains. In modern systems, especially when using dual-channel DDR5 mode, the memory bandwidth is already so high that it is often redundant.

In the context of Ryzen 7000 processors, the key is to have the memory controller work in 1:1 synchronization with the memory itself. DDR5-6000 memory is ideal in this regard, while frequencies above 6000-6400 MHz can lead to worse latency and loss of performance. However, adjusting the minimum timings for the Ryzen 7000 can have a significant impact on performance, especially with regard to the main timings.

Analysis of the test results showed that installing more powerful DDR5-6000 CL36 memory instead of DDR5-5200 CL40 gives only a small performance increase, on average 1%-2%. The difference in performance varies depending on the type of task: from zero increase in rendering to 5%-6% in information compression, and in some synthetic tests reaches 10%. In games, the difference is most often insignificant, and a realistic game mode shows only units or fractions of a percent in the change in performance. Using faster DDR5 memory makes practical sense only in rare tasks with low load on the video card.

Given the significant increase in the cost of higher-performance memory compared to base modules that operate at close to standard frequencies, it is not always justified to invest in high-performance modules. We also note that the addition of on-chip cache in Ryzen 7000 processors partially compensates for the relatively low efficiency of the on-chip DDR5 controller, which may have an impact on overall system performance.»