Łukasz Plewa comments

Repositories
Issues
Comments

Results 43 comments of


                                            Łukasz Plewa

umfMemspaceHighestBandwidthGet does not use HBM on SapphireRapids+HBM

In UMF we do exactly what you did with numactl—we generate a HighestBandwidth namespace, which is really just a set of NUMA nodes (for each CPU we add the node...

umfMemspaceHighestBandwidthGet does not use HBM on SapphireRapids+HBM

Hey, I was able to reproduce the issue on our side. There is no difference between the UMF and malloc versions in terms of which NUMA nodes the memory is...

umfMemspaceHighestBandwidthGet does not use HBM on SapphireRapids+HBM

> [@lplewa](https://github.com/lplewa) great work! [@breiters](https://github.com/breiters) perhaps you have an idea why the order of the benchmarks matters? Short sleep fixes performance, so probably kernel is still doing something after program...