News AWS Graviton4 CPU benchmarked against AMD and Intel processors — faster than predecessors and more cost-effective

[Admin]

Amazon's Graviton4 processors bring significant performance advantages over its predecessors and can compete against AMD's EPYC 'Genoa' and Intel's Xeon 'Sapphire Rapids' in many workloads.

AWS Graviton4 CPU benchmarked against AMD and Intel processors — faster than predecessors and more cost-effective : Read more

 

[bit_user]

Edit: Oops! I assumed he was talking about today's round of benchmarks! My comments were written against this Phoronix article:

https://www.phoronix.com/review/graviton4-96-core
​

A key detail of these benchmarks is that all instances had SMT disabled. I thought it was impressive to see the (lower-clocked?) Graviton 4 manage a few decisive victories over EPYC (both with the same core count). If you excluded the benchmarks where AVX-512 plays a significant role, I think the results would've tipped even more in Graviton 4's favor.

It's great to see Emerald Rapids and Sierra Forest in the mix, even with the former competing at a core-count disadvantage (and the latter obviously leaning far in the opposite direction, with 144 E-cores).

BTW, the author (Anton) neglected to mention the final Geomean, which went 21.3% in favor of EPYC 9684X. The EPYC 9654 even beat it by 11.5%.


Regarding the Phoronix tests actually discussed in the article, I have only two observations I'd like to note:

1. The Intel CPU (Sapphire Rapids Xeon 8488C) was the only processor with SMT/Hyperthreading enabled, putting it at a substantial disadvantage.
2. The Geomean of EPYC 9R14 was 24.9% greater than that of the Graviton 4, which is a much better win for AMD at 64 cores than the 11.5% they managed with the non-X3D 96-core CPU (above).