Benchmarking Google's New "C2D" Compute-Optimized AMD Zen 3 Instances - N2D vs. C2 vs. C2D Performance
First up was HPCG where c2d-standard-8 came out 22% faster than the prior-generation AMD Zen 2 powered n2d-standard-8 instance type. The c2d-standard-8 was also 14% faster than the Intel Xeon powered C2 instance type, which to reiterate is based on Cascade Lake with no C-series / Compute Optimized Ice Lake instance type yet.
On a performance-per-dollar basis the C2D value was the leader as well.
NAMD was the only workload tested where the Intel C2 instance of the same size did deliver a small lead over the EPYC Zen 3 compute instance. In any event, the generational uplift was very good for AMD Zen 2 to Zen 3 with 8 vCPUs and C2D delivered the best value.
Google was promoting big OpenFOAM computational fluid dynamics uplift with C2D and I can indeed confirm that.... OpenFOAM is not only significantly faster with C2D over N2D, similar to Google's findings, but it's also significantly better than the C2 instance too.
The C2D instance comes at excellent value for this HPC workload.
ACES DGEMM was another benchmark seeing huge uplift with the Zen 3 compute-optimized instance.