Message boards :
Number crunching :
Some seconds in cuda crunching saved
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 19 Aug 11 Posts: 38 Credit: 50,018,810 RAC: 0 |
Hy, the system variable CUDA_CACHE_MAXSIZE 4294967296 can enhance cuda crunching times, please test it by yourselfes. Greetings for germany |
|
Send message Joined: 18 Nov 22 Posts: 84 Credit: 653,414,458 RAC: 37,144 |
Didn’t notice any difference really on my TitanV’s but I’ll let it run a bit longer. Does it matter if the env variable is defined as int or str? I did export CUDA_CACHE_MAXSIZE=4294967296 Or maybe it’s more helpful for windows systems than Linux?
|
|
Send message Joined: 19 Aug 11 Posts: 38 Credit: 50,018,810 RAC: 0 |
ok, windows is recommended. |
|
Send message Joined: 19 Aug 11 Posts: 38 Credit: 50,018,810 RAC: 0 |
As i looking on my wu stats, there is no advantage anymore visible. Crule thing. As i am writing the advantage is back, creepy thing. ok before there i haved streamed on twich, maybe thats the fault. |
Joseph StatesonSend message Joined: 18 Nov 08 Posts: 291 Credit: 2,464,216,010 RAC: 9,275 |
Hy, the system variable Milkyway GPU apps are all OpenCL. CUDA is not listed |
|
Send message Joined: 18 Nov 22 Posts: 84 Credit: 653,414,458 RAC: 37,144 |
Hy, the system variable Nvidia GPUs run CUDA. In terms of env variable naming, “CUDA†is synonymous with “NVIDIA†so CUDA environment variables can still be effective. But this one isn’t.
|
|
Send message Joined: 19 Aug 11 Posts: 38 Credit: 50,018,810 RAC: 0 |
Milkyway GPU apps are all OpenCL. CUDA is not listed There are no variables for cache settings for opencl. |
|
Send message Joined: 22 May 11 Posts: 75 Credit: 5,758,579 RAC: 280 |
Like this? |
©2026 Astroinformatics Group