The Performance Impact Of Genoa-X's 3D V-Cache With The AMD EPYC 9684X

Written by Michael Larabel in Processors on 24 July 2023 at 01:30 PM EDT. Page 4 of 5. 30 Comments.
LeelaChessZero benchmark with settings of Backend: BLAS. Default was the fastest.
LeelaChessZero benchmark with settings of Backend: Eigen. Default was the fastest.
Xmrig benchmark with settings of Variant: Monero, Hash Count: 1M. Default was the fastest.
ASKAP benchmark with settings of Test: tConvolve MT, Gridding. Default was the fastest.
Google Draco benchmark with settings of Model: Church Facade. Default was the fastest.

Other workloads like the LeelaChessZero AI chess engine and Xmrig CPU miner were among other workloads showing sizable uplift from having the AMD 3D V-Cache. Or even workloads like Google's Draco 3D geometric mesh library.

Xcompact3d Incompact3d benchmark with settings of Input: input.i3d 193 Cells Per Direction. Default was the fastest.
Xcompact3d Incompact3d benchmark with settings of Input: input.i3d 193 Cells Per Direction. Default was the fastest.

As with the prior Milan-X testing, Genoa-X provided big advantages for the Xcompact3D/Incompact3D workload.

GROMACS benchmark with settings of Implementation: MPI CPU, Input: water_GMX50_bare. Default was the fastest.
Embree benchmark with settings of Binary: Pathtracer ISPC, Model: Crown. Default was the fastest.
Embree benchmark with settings of Binary: Pathtracer ISPC, Model: Asian Dragon. Default was the fastest.
Palabos benchmark with settings of Grid Size: 500. Default was the fastest.
NAMD benchmark with settings of ATPase Simulation, 327,506 Atoms. Default was the fastest.
PyHPC Benchmarks benchmark with settings of Device: CPU, Backend: Numpy, Project Size: 4194304, Benchmark: Isoneutral Mixing. Default was the fastest.

A wide range of workloads were intentionally tested for seeing what other areas are benefiting or not from the 3D V-Cache.

Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. Default was the fastest.

When compiling an "allmodconfig" Linux kernel build around ten seconds were shaved off the kernel build time with 3D V-Cache.

Timed Godot Game Engine Compilation benchmark with settings of Time To Compile. Default was the fastest.

Or a few seconds shaved off the Godot game engine build time but ultimately the core count is more important here than the cache size.


Related Articles