Message boards :
Number crunching :
Fine tuning dnet1.03 cuda31
Message board moderation
Author | Message |
---|---|
Send message Joined: 8 Nov 19 Posts: 4 Credit: 76,290,155 RAC: 0 |
I was wondering if you guys could adjust the CPU threshold on this project by a bit? For faster systems (limited to 4Ghz CPU, and RTX 2080Ti), a single WU is too slow for a GPU. Yet running 2 WUs in 1 GPU works, but it requires 2x 0.97% of CPU. The GPU can run 3 WUs together as optimal setting (GPU = 0.33), but I don't have enough CPU cores, to run this setting. 3 WUs per GPU, 3 GPUs, means I need 9 CPU cores, and I only have 6 (full cores non hyperthreaded) at 4Ghz, which are also divided between other projects. I'm currently running 2 WUs per GPU, resulting in 6x GPU WUs, and each CPU core running at 97%. Not enough to squeeze a CPU WU in there. At best, currently, I can limit the amount of instances MooWrap can run, to 4WUs. But this is far from optimal, knowing my system could run 9, if only it used ~66% CPU instead of ~97%CPU. |
Send message Joined: 8 Nov 19 Posts: 4 Credit: 76,290,155 RAC: 0 |
I would be interested in your settings, you can present the files here. What settings are you referring to? All the parameters (CPU and GPU) in App_config.xml are mentioned above. |