AWS Graviton4 Benchmarks Prove To Deliver The Best ARM Cloud Server Performance
First up was the miniFE finite element mini HPC benchmark. The generational gain from Graviton3 with r7g.16xlarge to Graviton4 with r8g.16xlarge was enormous. Going from Neoverse-V1 to Neoverse-V2 and additional memory bandwidth were paying off big time for this benchmark. In this case it even led over the AMD EPYC 9R14 Genoa competition on AWS... Granted, AMD very soon will be introducing their new 5th Gen EPYC "Turin" processors that should better compete with Graviton4 here. Similarly, the current Intel Xeon R7i instance is based on Sapphire Rapids while Emerald Rapids is available in the marketplace but not currently on AWS while more interestingly will be the upcoming Xeon 6 Granite Rapids processors. In any event right off the bat the huge jump for Graviton4 was very surprising in yet another huge generational gain for Graviton.
Another HPC benchmark showing Graviton4 delivering huge generational gains and even now coming out ahead of 4th Gen AMD EPYC at 64 vCPUs is Incompact3D/
Not only did the Graviton4 outperform the EPYC instance in raw performance but also was yielding much greater value in performance per dollar on AWS.
OpenFOAM for computational fluid dynamics was also yielding huge gains with Graviton4. Graviton3 was performing just under the Intel Xeon 8488C instance at the same 64 vCPU count while now with Graviton4 it leaps ahead of Intel Xeon and delivers similar performance to the EPYC 9R14 instance.
The R8g instance was in turn yielding the best performance per dollar now of the tested AWS cloud instances.
Even with larger mesh sizes, the Graviton4 instance continued competing with the AMD EPYC instance. The gains from Graviton2 to Graviton3 and now Graviton4 are very impressive and will be interesting to see how long AWS and Arm can keep up the generational improvements.
There were healthy gains for Graviton4 with GROMACS but AMD EPYC and Intel Xeon were the fastest.