AmpereOne Performance Scaling From 32 To 192 Cores, Core-For-Core Benchmarks Against Ampere Altra Max
Written by Michael Larabel in Processors on 29 August 2024 at 10:25 AM EDT. Page 2 of 8. 21 Comments.
Of course, not all workloads scale effectively up to 192 threads in which case for some CI/CD type build farms and the like it can be more beneficial running multiple VMs/containers concurrently for building multiple codebases/builds at a time.
👁 WRF benchmark with settings of Input: conus 2.5km. AmpereOne A192-32X @ 192 Cores was the fastest.
Some of the MPI/OpenMP workloads able to scale well could benefit from greater memory bandwidth. Currently AmpereOne allows up to eight channels of DDR5-5200 memory. But with AmpereOne M set to stat shipping in Q4, there will be up to 12 channel DDR5 memory for AmpereOne M. Next year with AmpereOne MX is up to 256 cores with 12 channel DDR5 memory. This will be very important for some HPC workloads where they can be memory bandwidth starved near 192 cores with only eight channels of DDR5-5200 memory.