Testing Third-Gen Ryzen DDR4 Reminiscence Efficiency and Scaling

When we reviewed Ryzen's latest iteration, we briefly examined how DDR4-3200 CL14 is compared to AMD's DDR4-3600 CL16 memory, which they claimed was an optimal configuration. It turned out that there was little difference between the two, which led us to conclude that both performed optimally. Also, spending more money to get higher clocked DDR4 memory didn't seem like a wise investment.

Over the past few years, we've looked at manually optimizing storage times for Ryzen and found solid performance improvements. So we wanted to check this again. After things calmed down after the start and had more time for other tests, we were asked to measure the storage performance of 3rd generation Ryzen.

We did our best not to miss anything and forced hundreds of benchmark runs to collect our data. We use the Ryzen 9 3900X to work with a variety of memory configurations. Towards the end of this function, however, we will discuss compatibility with cheaper processors.

All testing is done at 1080p, but we considered GPU scaling results with RTX 2080 Ti and Radeon RX 5700 and RX 580. This gives us both CPU-bound and GPU-bound results. We also test presets with maximum and medium quality in four games. At 1440p and 4K this doesn't have to be repeated since we have these RX 580 results. Please also note that at least 120 benchmark runs are required for each game.

Our preferred modules for the compilation consisted of three 16 GB memory kits: G.Skills new TridentZ Neo DDR4-3600 CL16 memory, G.Skills FlareX DDR4-3200 CL14 memory and a dirt-cheap kit from the Team Group, the T-Force Dark DDR4-3000 CL16 memory. The Team Group Kit is available for $ 70, the FlareX material is about twice as expensive at $ 135, and the TridentZ Neo costs $ 170.

We ran the T-Force Dark DDR4-3000 memory in its default configuration, with the XMP profile loaded and nothing else changed. Then we manually set all timings for an optimal Samsung S-Die configuration at 3000 MT / s. The G.Skill FlareX memory was tested in its standard specification with loaded XMP and then the memory speed was reduced to 3000 MT / s, so that we have a CL14 and CL16 comparison between the FlareX and the T-Force memory. Finally, the TridentZ Neo is also in the out-of-the-box specification at 3600 MT / s, an overclocked configuration of 3800 MT / s using the XMP timings and a maximum OC configuration at 3800 MT / s with manual Timings.

Here is a brief overview of the manual timings used for the DDR4-3000 and 3800 configurations. If you want to optimize your own memory, we recommend that you download the Ryzen DRAM Calculator. It's a really cool little tool.

If we look at the memory latency for the various test configurations, we see a fairly large 6% reduction in latency from CL14 DDR4-3000 to 3200 with only a 3% reduction when jumping to DDR4-3600 CL16 and then another 3% DDR4-3800 CL16.

Although the focus of this feature is on gaming performance, you should ask yourself whether these storage speeds and times don't generally affect application performance as much. Although this is a fairly large generalization, any memory-sensitive application will be affected. However, when rendering and coding workloads, you don't see a dramatic difference, as you can see when looking at these corona results.

We see a 6% performance improvement when we switch from DDR4-3000 to DDR4-3800, which theoretically doesn't mean much for a massive 32% increase in memory bandwidth.

Starting with game tests and Assassins Creed Odyssey with the GeForce RTX 2080 Ti at 1080p and activated default for ultra high quality. Here we see some interesting results. First, there is very little difference between DDR4-3200, 3600 and 3800 using the XMP timings. Low latency DDR4-3000 drops off a bit, especially due to the 1% low performance, and we see that performance drops slightly more when using the CL16 timings.

However, by optimizing the DDR4-3000 memory, we can achieve better results than with the CL16 3800 configuration, which is incredible. By manually adjusting the timings, we see an incredible increase from 38% to 1% low performance and an increase of 15% for the average frame rate. That blew us away.

Even better, if we optimize the DDR4-3800 memory, we get a further increase of 8% for the average frame rate and 10% for the 1% low. This means that with a standard setting, the 3900X enabled an average of around 80 fps and, with a little tinkering, reached up to 90 fps.

Interestingly, reducing quality settings reduces margins, which may be explained by lower CPU usage. So we're limited to the GPU, even if that doesn't sound intuitive. However, we believe that this is the case. Regardless of the situation, the optimized DDR4-3800 configuration is now only 6% faster than the XMP version. Optimizing the DDR4-3000 budget memory enables first-class DDR4-3800-like performance.

To get a more complete picture, we also tested it with the mid-range Radeon RX 5700. With the default for ultra quality, we're completely limited to the GPU at 1080p, so memory has almost no impact on performance. I have to fall back on an unrealistic specification like DDR4-2133 to see a drop in performance. Since 3rd generation Ryzen officially supports DDR4-3200, we didn't have to run tests below 3000 because you shouldn't be using slower memory. The medium quality in this configuration leads to deviations in the results, although this is only due to the fact that we are no longer tied to the GPU at these higher frame rates. By optimizing the DDR4-3000 memory, it can again be adapted to the DDR4-3800 XMP configuration as well as the GPU-restricted manual 3800 specification.

With an installed Radeon RX 580 or a GPU with roughly the same performance, we are again strongly tied to the GPU. There are slightly more deviations here than with the RX 5700, but ultimately we still consider results that are largely within the error limit. Even if you lower the preset quality by a few steps to medium, we only see a 6% difference between the absolutely fastest and the slowest configuration. When purchasing memory, it is therefore important to take into account the graphics card you are using.

When you look at all the Assassins Creed Odyssey 1080p ultra-quality tests, there are a few important aspects: Yes, faster memory can improve performance, but for the serious gains, you'll need to manually optimize your memory. In CPU-limited scenarios, profits can be massive. Conversely, when the workload is limited to the GPU, the gains are small to none, and while this seems obvious, almost all 3rd generation Ryzen tests we've seen online so far have been done mostly under CPU-limited conditions . As you can see, even with a medium range GPU at 1080p like the Radeon RX 5700, faster memory has little to offer. The same applies to even slower GPUs like the RX 580.

We would even argue that even with an RTX 2080 Ti, you're very likely bound to a GPU, at least if you're using a modern processor with six or more cores. If we increase the resolution to just 1440p, it reduces the 2080 Ti to an average of around 70 fps with a 1% low value of around 50 fps, which is very similar to that of the RX 5700 at 1080p and means GPU-bound performance faster memory only a very small chance of leaving traces.

When using medium quality settings, we find that even at an average of 70 fps you are much more tied to the GPU than to the CPU. Only when we start to push over 80 fps will the game be a bit more CPU-bound. For those wondering, the RTX 2080 Ti is only 106 fps on average at 1440p with a 1% low of 67 fps, similar to what we see on the RX 5700 at 1080p. This means that in addition to the official DDR4-3200 AMD specification, you can increase performance with faster memory by about 10%.

Next we tested all of the memory configurations in Far Cry New Dawn and this time we only see a 4% increase over the DDR4-3200 configuration when the DDR4-3800 memory is manually set. Interestingly enough, there is quite a bit of a drop in DDR4-3000 memory, and even manually adjusting the timings doesn't help make up for the higher-frequency kits. We know that Far Cry New Dawn is sensitive to memory bandwidth, so this is likely to be a problem for 3000 MT / s memory.

We see something similar with the use of the Radeon RX 5700, although the manually tuned DDR4-3800 memory interestingly offers a nice little performance increase and is therefore 7% faster than the DDR4-3200 memory. The GPU of the RX 580 is completely limited to around 80 fps. So in Far Cry New Dawn you have to push the highest quality preset over 100 fps to use the faster memory.

Reducing the quality preset by two levels to the normal value does not change anything. The average frame rate of the RX 580 is only increased by 10 fps, which is why we are still very limited to the GPU.

If we move to Rainbow Six Siege, we have a mostly GPU-bound competition shooter here. When using an RX 5700 or equivalent GPU with medium range, the performance does not change when using the ultra quality settings even at 1080p. Of course, this also applies to slower GPUs like the RX 580.

Even with the RTX 2080 Ti, we only see a 4% increase in performance, which ranges from the DDR4-3200 specification to the manually tuned 3800 memory. By reducing the quality settings for higher frame rates, the RX 5700 continues to offer a strongly GPU-bound scenario. With the RTX 2080 Ti we still only see a 4% increase from DDR4-3200 to manually set 3800 memory.

We recently tested World War Z and at 1080p with the ultra settings, the RX 580 achieved an average of slightly more than 140 fps and still we were still heavily tied to the GPU despite the cheap DDR4-3000 memory. We see a little difference from the RX 5700, but still the manually tuned DDR4-3800 memory was only 7% faster than the budget 3000 memory. So that's pretty weak, although we see a 15% increase for the 1%. bad performance. Compared to 3200 memory, the fastest configuration offered only a 9% performance increase.

When using the medium quality preset, we see a large increase in 1% low performance when using manually tuned memory, namely the DDR4-3800 material. With the RTX 2080 Ti, we see an increase of 18% for the manually tuned DDR4-3800 over the CL14 DDR4-3200 memory with low latency. A nice boost, although again you can expect these gains to largely disappear at 1440p even with an RTX 2080 Ti.

Finally, we suspect that a clock frequency of 1900 MHz Infinity Fabric will be a bit too high for most 3rd generation Ryzen processors. For example, our 3900X made it comfortable, but it was a bit sketchy with the 3700X and then the 3600X and 3600 CPUs would not go beyond 1800 MHz. The vanilla R5 3600 even had to be tinkered a little to become stable.

For this reason, we believe that DDR4-3600 is the sweet spot for the X models. All 3rd generation high-end Ryzen processors should handle this frequency. But for the cheaper models, DDR4-3200 to 3400 is a safer choice, and as we've found, even 3000 is fine, especially if you can conveniently set the sub-timings.

Keep in mind that in most cases you will be tied to the GPU anyway while gaming, as these 3rd generation Ryzen processors are very fast even with loose DDR4 memory.

The bottom line is that you can get a cheap 16GB Samsung S-Die kit for $ 70 and get the maximum gaming performance even with a 3900X + RTX 2080 Ti configuration. Ryzen doesn't need premium storage to get the best possible performance. For those who buy a Ryzen 5 model, we strongly recommend not spending any money on expensive CPU-bound memory.

Purchasing links:

Leave a Reply

Your email address will not be published. Required fields are marked *