Hi, i’m curious whether the significant CUDA kernel launch overhead on Windows Vista / Windows 7 (see The Official NVIDIA Forums | NVIDIA) affects also Quadro cards, or only Geforce cards. Could be likely a downside for Quadro’s, as they are marketed by NVIDIA also as ‘computational’ device (many programs in DCC, post production - where Quadro is de-facto standard - now use CUDA acceleration). And a lot of basic GPU functions (e.g. stuff from CUDPP, Thrust like reduction, compaction etc.) has lot of kernel calls. Furthermore I’m interested, whether this WDDM kernel launch overhead is going to decrease in Windows 8 - some preliminary information by NVIDIA guys maybe ?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Can too many kernel calls affect the performance ? | 4 | 983 | June 10, 2010 | |
| Very slow kernel launches | 8 | 7724 | March 28, 2015 | |
| Concurrent Kernel Launching to Hide Kernel Launching Overhead (Not only kernel execution)) | 0 | 402 | April 9, 2020 | |
| fundamental cuda kernel launch questions | 2 | 16492 | July 31, 2008 | |
| cost of launching a new grid | 3 | 701 | June 24, 2019 | |
| Dispatch Kernel Overhead (OpenCL) | 6 | 3583 | March 28, 2017 | |
| CUDA slower in Windows 7 than in Windows XP same computer, two OSs, different run times | 21 | 18960 | November 11, 2009 | |
| kernel launch time expensive? | 2 | 1597 | July 28, 2009 | |
| Kernel Overhead/Profiler Accuracy | 4 | 6393 | May 25, 2008 | |
| Kernel launch overhead for GTX 680 | 2 | 1827 | May 19, 2012 |