TOOLS: use separate cuda alloc in perftest

openucx / ucc

Unified Collective Communication Library

https://openucx.github.io/ucc/

BSD 3-Clause "New" or "Revised" License

196 stars 96 forks source link

Closed Sergei-Lebedev closed 9 months ago

Sergei-Lebedev commented 11 months ago

Add CUDA memory allocation function to ucc_perftest

Result of perftest might be biased due to use of internal memory pool in UCC.

Sergei-Lebedev commented 9 months ago

Copyrights needs to be updated.

Can you explain why it is different for host mem type ?

It's not different for host mem type, but host memory is less sensitive to memory pools. Update PR with host and cuda managed memory