Is it possible to limit the number of SMs that a program is using? Assume the device has 15 SMs an I want to use only 8 of them? Is there any option in the nvcc or nvidia commands?
there are no options in any NVIDIA tools, nor anything provided by the CUDA runtime to easily enable this.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Specify number of cores to use | 5 | 5208 | November 4, 2010 | |
| Opencl/CUDA processes that keep gpu percentage busy | 3 | 499 | March 21, 2019 | |
| How to control the SMs number during kernel execution? | 2 | 1027 | January 14, 2010 | |
| Can we restrict number of SM for a kernel in cuda | 2 | 1510 | May 21, 2024 | |
| how to limit the number sm in my program? | 4 | 1661 | March 6, 2019 | |
| How to limit the number of SMs used by a program? | 6 | 406 | March 17, 2025 | |
| Is there any way to disable part of the streaming processors for the CUDA Computing | 4 | 883 | October 21, 2013 | |
| Disable SMs in RTX 3090 | 1 | 514 | September 13, 2022 | |
| Any possibility to disable some SMX in a GPU? | 3 | 1047 | December 27, 2017 | |
| Some questions on selecting a certain number of SM to simulate. | 2 | 1105 | December 4, 2012 |