It has already been reported a few times on these forums that putting the GPU to sleep will affect cuda processing speed immediately after the wakeup. That is known behavior. If you’d like attention on your specific case, I’d suggest filing a bug. I’m not saying its a defect or that anything can be done about it.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Why kernel calculate speed got slower after waiting for a while? | 9 | 1779 | July 19, 2022 | |
| Cuda slow performance after process sleep/wait | 1 | 1252 | June 14, 2022 | |
| GPU and CPU don't run in (pure) parallel ? | 24 | 20173 | May 4, 2007 | |
| cudaSynchronizeDevice() returns error code 6 | 7 | 8611 | June 16, 2011 | |
| CUDA very slow performance | 21 | 16813 | March 6, 2020 | |
| Needing expert advice.. | 4 | 1272 | July 21, 2014 | |
| processing time check | 5 | 552 | November 16, 2010 | |
| Language confusion with multi-gpu | 11 | 19297 | October 30, 2007 | |
| Time consuming comparison between 820M GPU combination code and pure C++ on CPU | 2 | 497 | May 22, 2019 | |
| Error 719 (failure to launch) for JCUDA and PyCUDA; How to run GPU consecutive times for 'large' data blocks | 0 | 2336 | December 13, 2016 |