machinelearning icon indicating copy to clipboard operation
machinelearning copied to clipboard

Published 20 hours ago • verified publisher icon dotnet

Poor reasoning performance

Open shiqiangz-git opened this issue 3 years ago • 4 comments

Gpu concurrent reasoning is too slow, CPU reasoning occupancy is too high, resNet50 image classification

shiqiangz-git avatar Jan 20 '22 06:01 shiqiangz-git

With gpu 1080ti

shiqiangz-git avatar Jan 20 '22 06:01 shiqiangz-git

Can you share any results/numbers you are seeing along with what you would expect to be seeing? And share a sample project that repros the situation.

michaelgsharp avatar Jan 27 '22 19:01 michaelgsharp

Can you share any results/numbers you are seeing along with what you would expect to be seeing? And share a sample project that repros the situa

Can you share any results/numbers you are seeing along with what you would expect to be seeing? And share a sample project that repros the situation.

Sorry, I can't share my project. I just import about 670,000 pictures, then train them with resnet50, and then build webapi with pudectpool. If I use CPU, the CPU utilization rate slowly rises, and the gpu speed is not ideal. Can you tell me the reason? The performance has not reached an acceptable range, and the CPU utilization rate of a few pictures is 100.

shiqiangz-git avatar Jan 28 '22 15:01 shiqiangz-git

6,7w pictures

shiqiangz-git avatar Jan 28 '22 15:01 shiqiangz-git