0

We're using dual cpus, each with ten cores,four GPUs. When two programs run on different GPUs, the speed remains the same, but when programs are added on other GPUs, the speed of the two programs running before is slowed down. I'd like you to give me your advice.

We thought that threads in the Dataloader would interfere with running, so we used "taskset" to limit threads while the program was running, but it didn't work. top

  • presumably there are shared resources, such as memory, bus, etc... – Mitch Wheat Jul 07 '23 at 03:22
  • Check your CPU and GPU usages to see where is the bottleneck. If your GPUs are too fast for your CPU to keep them busy, you'll get lower overall GPU performance. This situation might be alleviated using multithreading on the host if it doesn't already, but you didn't say if it's your own development project or not so might not be an option. – Simon Goater Jul 07 '23 at 11:54
  • I'm very sorry to see your answer just now.We have a complete surplus CPU and memory ,that should not be the bottleneck of the problem.I don't know much about the bus. Do you have any advice – user22188171 Jul 10 '23 at 01:50

0 Answers0