AWS Graviton4 96-Core Performance vs. AMD EPYC & Intel Xeon CPU Benchmarks
First up is looking at the code compilation performance across these tested processors. The Graviton4 96 core metal instance proved to be much, much faster than the Ampere Altra Max 128-core flagship but behind the Xeon and EPYC processors of varying core counts. It's too bad there wasn't the ability to monitor the CPU power consumption of the Graviton4 metal instance for where it's an even more interesting story with Intel Xeon 6 Sierra Forest and the like. Graviton4 being much faster than Ampere Altra Max isn't really surprising considering Graviton4 uses Neoverse-V2 cores compared to Ampere Altra (Max) on Neoverse-N1 as well as making use of DDR4 system memory, etc. It will be more interesting to see how Graviton4 competes with AmpereOne once finally getting hands on with AmpereOne and/or finding GA availability from cloud providers like Oracle.
Between x86_64 and AArch64 there can be some code path differences during compilation, but in any event when looking at the LLVM build time it was fascinating to see the Graviton4 performing so well.
For the SHA256 performance due to utilizing all available CPU threads, in this basic benchmark the Ampere Altra Max at 128 cores delivered similar performance to Graviton4 at 96 cores.
The Graviton4 performance came on strong with the ClickHouse database benchmark. The Graviton4 96 core instance was only a short distance behind the 96 core AMD EPYC Genoa(X) competition. Granted, AMD EPYC Turin is due out in the coming months and over on the Intel side the Xeon 6 Granite Rapids launch is also coming up.