AWS Graviton4 vs. AmpereOne 192-Core Benchmarks For Leading AArch64 Server Performance

Written by Michael Larabel in Processors on 9 September 2024 at 11:15 AM EDT. Page 4 of 8. 7 Comments.
Coremark benchmark with settings of CoreMark Size 666, Iterations Per Second. Graviton4 192 vCPUs was the fastest.

The synthetic Coremark CPU benchmark saw around 10% better performance with Graviton4 on the Neoverse-V2 cores compared to AmpereOne.

Algebraic Multi-Grid Benchmark benchmark with settings of . Graviton4 192 vCPUs was the fastest.

When moving to the HPC benchmarks is where the AWS Graviton4 performance began delivering some very significant gains over the AmpereOne top-end SKU.

WRF benchmark with settings of Input: conus 2.5km. Graviton4 192 vCPUs was the fastest.

Graviton4 with the 192 cores via r8g.48xlarge / r8g.metal-48xl instances were commonly twice as fast as the AmpereOne A192-32X server. Some very big wins for Graviton4 for this 192-core AArch64 server showdown. Granted, keeping in mind with Graviton4 it's a dual socket configuration to make 192 cores and there is 12 channel DDR5-5600 memory compared to 8 channel DDR5-5200 memory, but that alone doesn't explain for the huge leads in some of these HPC benchmarks for the AWS Graviton4 processor.

LULESH benchmark with settings of . Graviton4 192 vCPUs was the fastest.

Both configurations as a reminder were using Ubuntu 24.04 LTS with the Linux 6.8 kernel in the 64K page size kernel configuration.

LAMMPS Molecular Dynamics Simulator benchmark with settings of Model: Rhodopsin Protein. Graviton4 192 vCPUs was the fastest.
LAMMPS Molecular Dynamics Simulator benchmark with settings of Model: 20k Atoms. Graviton4 192 vCPUs was the fastest.

It's really a shame though the inability to monitor the CPU/system power consumption for EC2 metal instances for seeing how the overall power consumption and performance-per-Watt compares to AmpereOne.

Related Articles