Starting from a local 6c12t machine needing 6 hours running a 48 frame simulation on Maya Bifrost, I decided try speeding up things on Azure.
So far installed and tested Maya Bifrost on 32, 64 and even 120 vCPU instances of D v4, F v2, L v2, HB series. None of these instances does have a dedicated GPU, but I was able to have Maya UI working and Bifrost sim is totally on CPU so that in theory I will have a strong performance boost.
So far I got in the best case a 10% speedup and on average a -50% slowdown. Storage seems not to be an issue (generating only 1MB images a frame as output), cpu is working even if stays almost idle for some time between the various sim computation.
Setting up now on NVv4 GPU powered instances so to check if having an actual dedicated GPU will make any difference, in theory no since once launched the sim is all CPU time and just a few graphics activity for the final image rendering.
Wanted to ask how I can troubleshoot so find the bottleneck, or if there is any other variable I'm not considering (i.e. vm overhead respect a dedicated machine).
