Ampere Altra Max Continues To Deliver Competitive Power Efficiency To AMD EPYC & Intel Xeon

Written by Michael Larabel in Processors on 11 December 2023 at 01:42 PM EST. Page 2 of 6. 81 Comments.
High Performance Conjugate Gradient benchmark with settings of X Y Z: 144 144 144, RT: 60. EPYC 9654 was the fastest.

In raw performance the Ampere Altra Max 128-core tended to trail these current AMD EPYC and Intel Xeon processors, to not much surprise. After all, Ampere Altra is now three years old and relying on DDR4-3200 and other dated capabilities compared to the newest AMD EPYC and Intel Xeon Scalable platforms with DDR5 and other leading features. AmpereOne is where the raw performance should be quite interesting when we finally manage to see hardware in the lab.

High Performance Conjugate Gradient benchmark with settings of X Y Z: 144 144 144, RT: 60. EPYC 9654 was the fastest.
High Performance Conjugate Gradient benchmark with settings of X Y Z: 144 144 144, RT: 60. EPYC 9654 was the fastest.

But when it came to the performance per Watt was still very interesting. The Ampere Altra Max M128-30 that I received in mid-2021 was capable of delivering better power efficiency in some workloads than the latest AMD Genoa(X) and Bergamo processors and Intel Sapphire Rapids.

NAS Parallel Benchmarks benchmark with settings of Test / Class: CG.C. EPYC 9684X was the fastest.
NAS Parallel Benchmarks benchmark with settings of Test / Class: SP.C. EPYC 9684X was the fastest.
miniFE benchmark with settings of Problem Size: Small. EPYC 9684X was the fastest.
miniFE benchmark with settings of Problem Size: Small. EPYC 9684X was the fastest.
miniFE benchmark with settings of Problem Size: Small. EPYC 9684X was the fastest.

In some workloads though they are just not well optimized yet for Ampere processors or more broadly AArch64.

miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. EPYC 9754 was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. EPYC 9754 was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. EPYC 9754 was the fastest.

The aging Ampere Altra Max processor still beat out the Intel Xeon Platinum 8490H for performance-per-Watt in the miniBUDE HPC benchmark.


Related Articles