GPU Det and Directsum are ridiculously slow

Open jeffry1829 opened this issue 1 year ago • 3 comments

Det uses cusolver ?getrf

Currently not sure whether this only happens to these two methods

Oct 01 '24 08:10 jeffry1829

What did you compare them to, the CPU version? How large is the input tensor? For inspecting the reason, myebe the NVDIA profiler can help.

Oct 02 '24 06:10 IvanaGyro

GPU Det and Directsum are ridiculously slow

Det uses cusolver ?getrf

Currently not sure whether this only happens to these two methods

Are you benchmarking against CPU version? Or old magma version?

Oct 08 '24 12:10 yingjerkao

I believe this issue was due to the fact that our DGX II has been hacked. Should perform the benchmark on some other machines.

Nov 15 '24 01:11 yingjerkao