ROCm
Where can i find the meaning of some metrics from omniperf?
Describe your question
Just wondering if there are some explanations on what some mertics means (like nvidia compute). For example, i'd like to know MFMA, VALU and so on, while there're no documents i could refer to. I appreciate it if you could provide some documents in the README.md.
Also, i wanna know if there're some metrics we can regard as the counterparts of Compute Throughput/ Memory Throughput in nsight compute.
Additional context Add any other context or screenshots about the question here.
@xinji1 Excuse me, do you solve this problem or find some documents? I am also very confused about the meaning of the metrics.
Just wondering if there are some explanations on what some mertics means (like nvidia compute). For example, i'd like to know MFMA, VALU and so on, while there're no documents i could refer to. I appreciate it if you could provide some documents in the README.md.
@Xi-Meng @xinji1 please take a look at our new "Performance Model" section in the Omniperf documentation: https://rocm.github.io/omniperf/performance_model.html#
We hope to eventually embed these definitions directly within Omniperf for even easier access (#223). If you have additional questions after reading please follow up and we'd be happy to help.
Also, i wanna know if there're some metrics we can regard as the counterparts of Compute Throughput/ Memory Throughput in nsight compute.
@skyreflectedinmirrors could you please help with this request.
Hi @xinji1. Internal ticket has been created to assist with your request. Thanks!
Hi @xinji1, the system speed of light section in the documentation mentioned above (https://rocm.github.io/omniperf/performance_model.html#system-speed-of-light) contains all the metrics that we currently support. There is no direct equivalent to Compute Throughput/ Memory Throughput, but we do have all the bandwidth and utilization metrics separately. Hope this clears things up.
