Hi, I’m trying to figure out exactly what the power consumption metric tells me in the profiler. I’m trying to find out how much power my application would consume if it was running continuously. It consists of about 8 kernels, so there’s a few pauses in between kernel launches. When I click the device in the visual profiler it says 40W average. However, if I run the application continuously with nvprof, it will slowly climb to an average of 70W. The visual profiler said it was running it 30 times, so I’m not sure which is correct. Also, how often are clocks turned off/slowed down in response to no activity on the SMs?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| power profiling with nvprof and averages | 1 | 2608 | May 13, 2016 | |
| Timing CUDA program CUDA Visual Profiler vs. clock() | 2 | 1529 | June 17, 2009 | |
| How to explain the performance difference? | 7 | 3569 | March 26, 2008 | |
| visual studio performance profiler on CUDA code | 1 | 6957 | March 20, 2008 | |
| How to Understand output of nvprof? | 1 | 2549 | June 3, 2015 | |
| Difference between time measured and time reported by profiler | 0 | 970 | January 19, 2009 | |
| Compute Visual Profiler # runs Compute Visual Profiler on CUDA | 2 | 1925 | March 7, 2011 | |
| Question about NVIDIA CUDA Visual Profiler Version 2.2 | 0 | 2933 | November 13, 2009 | |
| Profiler speeding up my kernels? Nvidia employees please read Weird timing behavior during profiler | 6 | 5883 | November 9, 2009 | |
| Updated beta visual profiler v0.2 | 0 | 2914 | April 23, 2008 |