Skip to content

Conversation

@dagelf
Copy link

@dagelf dagelf commented Sep 16, 2024

2024-09-16-182222_902x862_scrot

From a casual look it looks like

( kill -STOP -1; CUDA_VISIBLE_DEVICES="0" timeout 3600s python mamf-finder.py ; kill -CONT -1)

Might yield substantially different results...

Either that, or the CUDA scheduler is interfering with the timing, or the timing isn't done properly...

@dagelf
Copy link
Author

dagelf commented Sep 16, 2024

PS Don't run that kill command if you want any kind of interaction with the machine running this, it will even suspend SSH.

@mag-
Copy link
Owner

mag- commented Sep 18, 2024

Hi,
Sorry for late reply I didn't get the notification

The speed of matrix multiplication on GPUs varies depending on the dimensions of the matrices involved. CUDA and other GPU programming libraries use this information to select appropriate kernels and implement different computational strategies.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants