Table of Contents
Introduction
AMD’s newest Threadripper™ processors—the Threadripper™ PRO 9000 WX-series and Threadripper™ 9000 series—support registered DDR5 memory at up to 6400 MT/s (megatransfers per second). This is an improvement from the last-generation 7000 parts, which only supported up to 5200 MT/s. While it is impressive that AMD and motherboard partners were able to support this on existing TRX50 and WRX90 motherboards, it does throw a curveball for those looking to upgrade their Threadripper™ 7000-based computer: do you need to also upgrade your RAM to get the best results? Additionally, the supply of 6400 MT/s RDIMMs has been shaky so far, and a little more expensive, so some users may be inclined to go with a cheaper 5600 MT/s kit.
Image
In this article, we will examine the impact of RAM speeds on the new Threadripper PRO 9000WX processors in content creation applications. To do so, we selected two of these processors, the 9955WX and 9995WX, and tested them with 5200, 5600, and 6400 MT/s registered memory modules running at standard JEDEC settings. At 6400 MT/s, we also tested with both four- and eight-channel memory configurations.
In the future, we would like to investigate this further by looking at more processors, higher-clocked modules, and a wider range of memory channels. However, this investigation should cover the most common scenarios for the Threadripper PRO parts and be illustrative for the Threadripper 9000 CPUs as well.
Test Setup
AMD WRX90 Test Platforms
| CPUs: AMD Ryzen™ Threadripper™ PRO 9995WX AMD Ryzen™ Threadripper™ PRO 9955WX |
| RAM: Micron DDR5-6400 ECC Reg. 64GB (512 GB total) Kingston DDR5-5600 ECC Reg. 32GB (128 GB total) Kingston DDR5-5200 ECC Reg. 32GB (128 GB total) |
| CPU Cooler: Asetek 836S-M1A 360mm |
| Motherboard: ASUS Pro WS WRX90E-SAGE SE BIOS version: 1203 |
| GPU: NVIDIA GeForce RTX™ 5080 Driver 576.80 |
| Storage: Samsung 980 Pro 2TB |
| PSU: Super Flower LEADEX Platinum 1600W |
| OS: Windows 11 Pro 64-bit (26100) |
Benchmark Software
| Adobe Photoshop 26.8 – Puget Bench for Photoshop 1.0.5 |
| Adobe Premiere Pro 25.2.3 – Puget Bench for Premiere Pro 1.1.1 |
| Adobe After Effects 25.3.1 – Puget Bench for After Effects 1.0 |
| DaVinci Resolve 20.0.1.6– Puget Bench for DaVinci Resolve 1.2.0 |
| Blender 4.4.0 |
| V-Ray 6.00.01 |
| Cinebench 2024 |
| Unreal Engine 5.5 |
| Visual Studio 2022 |
| Llama.cpp 5122 Phi-3 Mini Q4 |
Our overall test configuration was relatively simple this time. All the configurations were tested on the same ASUS WRX90E-SAGE SE motherboard with an NVIDIA GeForce RTX™ 5080. In the BIOS, we disabled overclocking features (though they are much more limited to begin with on workstation boards than consumer desktop platforms) and ensured the RAM was running at its JEDEC specifications. We also verified that we were following the recommended module installation order for the reduced-channel testing, though we haven’t seen this make a large difference in the past. One issue we ran into was an incompatibility between the initial release BIOS (version 1106) and the Micron 6400 MT/s RAM we were testing with; updating to the latest 1203 BIOS fixed this.
On the software side, we are using many of our Puget Bench benchmarks, some of which were also used by AMD as part of their official AMD Ryzen Threadripper PRO 9000 WX-Series unveiling. We supplemented them with several other real-world benchmarks, primarily rendering packages like Blender and V-Ray.
Looking for a Content Creation Workstation?
We build computers tailor-made for your workflow.
Don’t know where to start?
We can help!
Get in touch with our technical consultants today.
Media and Entertainment
In media and entertainment workflows—applications like Adobe Photoshop and Premiere Pro—we have historically seen little performance improvement from RAM speed. This trend continued. In Photoshop, the overall spread was about 3%. We saw more difference when dropping memory channels, but it was still minor enough not to make a real difference. Premiere Pro was essentially the same, although the memory channel reduction was even more apparent here, with the 4-channel configuration on the Threadripper PRO 9995WX 8% slower than with the full 8 channels. These results held true in After Effects and DaVinci Resolve as well.
If you primarily work with standard media and entertainment workflows, the performance gains from higher-clocked RAM are minor. With overclocked memory, where frequencies would be even higher and timings tighter, there may be more of a benefit—but we don’t expect it to be huge, and it could come at the cost of reduced stability. However, we would caution against leaving memory channels unpopulated.
Rendering
In V-Ray and Blender, performance impacts from memory (both frequency and channels) were less than 1%. Cinebench was more sensitive, but the spread was still only about 3%—save for the 9995WX 4-channel configuration, which was slightly lower. We wouldn’t normally expect much impact from memory speed on rendering, and didn’t see any here.
Unreal Engine & AI
Unlike our previous categories, our various Unreal Engine tests have often shown scaling with memory bandwidth and frequency. In this round of testing, the Unreal Engine compilation test using Visual Studio found that the 6400 MT/s 8-channel memory for the Threadripper PRO 9995WX was 10% faster than the 5200 and 5600 MT/s configurations. Additionally, it was 20% faster than the 4-channel configuration. Interestingly, though, we didn’t see this same performance impact on the 9955WX. We suspect that the 9955WX does not have sufficient cores to make use of the additional memory bandwidth in this workload.
Shader compilation performance was similar. The 9995WX was 3% faster with 6400MT/s than with 5600 MT/s, and 7% than with 5200 MT/s. Reducing the number of channels was also punishing, with the 8-channel configuration being 82% faster than the 4-channel configuration. However, much like in the UE compilation test, the 9955WX saw little difference in time to completion based on memory. All the 8-channel configurations were essentially identical, although reducing to 4-channel did reduce performance by about 6%.
Our Light Baking benchmarks showed no difference in time to completion based on the memory configuration in these tests.
Lastly, we did see some differentiation in the Llama LLM benchmark using Phi Mini Q4. Memory speed made little difference, but for the 96-core part, reducing from eight to four channels dropped performance by about 14%.
What Impact Does Memory Speed Have On Threadripper PRO 9000?
We found that in most applications, increasing memory speed had little performance improvement for either the Threadripper PRO 9995WX or 9955WX. This was surprising given that we saw some scaling with frequency on desktop CPUs the last time we investigated this. The more complex memory subsystem of the Threadripper PRO likely reduces the overall impact. However, we did find that even in applications where memory frequency didn’t have much effect, the number of channels sometimes did. Specifically, for the 96-core 9995WX, we saw a consistent reduction in performance for 4-channel memory vs 8-channel memory at the same frequency.
There were some exceptions as well. In our Unreal Engine code and shader compilation benchmarks, performance scaled with memory frequency on the higher core-count part. This processor appears to be memory-constrained on these workloads. Other memory-intensive workloads in coding, HPC, and AI will likely see similar uplifts. However, this appeared to be irrelevant on the 16-core part, which seemed saturated even at four channels. Although we would have to investigate to be sure, we suspect that the 32-core and 64-core parts will see some, but more minor, improvements with RAM bandwidth in these workflows.
If you need a powerful workstation to tackle the applications we’ve tested, the Puget Systems workstations on our solutions page are tailored to excel in various software packages. If you prefer to take a more hands-on approach, our custom configuration page helps you to configure a workstation that matches your exact needs. Otherwise, if you would like more guidance in configuring a workstation that aligns with your unique workflow, our knowledgeable technology consultants are here to lend their expertise.
Looking for a content creation workstation?
We build computers tailor-made for your workflow.
Don’t know where to start?
We can help!
Get in touch with one of our technical consultants today.