ECC DDR5-4800 vs. DDR5-5200 Memory Performance For AMD Ryzen Zen 4

Written by Michael Larabel in Memory on 6 May 2024 at 02:30 PM EDT. Page 4 of 4. 7 Comments.

TensorFlow benchmark with settings of Device: CPU, Batch Size: 1, Model: AlexNet. 2 x 32GB DDR5-5200 was the fastest.

Blender benchmark with settings of Blend File: Junkshop, Compute: CPU-Only. 2 x 32GB DDR5-4800 was the fastest.

GROMACS benchmark with settings of Implementation: MPI CPU, Input: water_GMX50_bare. 2 x 32GB DDR5-5200 was the fastest.

Llama.cpp benchmark with settings of Model: llama-2-7b.Q4_0.gguf. 2 x 32GB DDR5-5200 was the fastest.

Llama.cpp benchmark with settings of Model: llama-2-70b-chat.Q5_0.gguf. 2 x 32GB DDR5-5200 was the fastest.

Llamafile benchmark with settings of Test: llava-v1.5-7b-q4, Acceleration: CPU. 2 x 32GB DDR5-5200 was the fastest.

Those wishing to see all 120 benchmarks I ran in full for this memory performance comparison can find them via this result page.

Geometric Mean Of All Test Results benchmark with settings of Result Composite, AMD Ryzen 9 7900 DDR5-5200 ECC Memory. 2 x 32GB DDR5-5200 was the fastest.

On average across the 120 benchmarks the DDR5-5200 vs. DDR5-4800 performance advantage was around 2% overall. However, in a number of real-world workloads like OpenFOAM CFD, Incompact3D, Llamafile, Llama.cpp, and others the improvement with these faster ECC UDIMMs on the AMD Ryzen Zen 4 server was in the 4~6% range. Long story short if running memory-intensive workloads on AMD Ryzen 7000 series budget servers, the faster memory can be worthwhile assuming you aren't challenged by any availability issues in finding the faster memory modules. As of writing the KSM56E46BD8KM-32HA DDR5-5600 DIMMs I've been testing with can be found in-stock at Amazon.com (affiliate link) for around $120 USD per 32GB DIMM.

7 Comments

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.

Page: 1 2 3 4