Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add controllable delay between starts of compute blocks #16277

Open
ttmtrajkovic opened this issue Dec 23, 2024 · 1 comment
Open

Add controllable delay between starts of compute blocks #16277

ttmtrajkovic opened this issue Dec 23, 2024 · 1 comment
Assignees

Comments

@ttmtrajkovic
Copy link
Contributor

Implement a controllable delay between next start of compute block.

llk_wait_tiles() - start of workload
llk_push_tiles() - end of workload
--> measure time between push_tiles() and next wait_tiles() and make sure it’s always at least some amount of time

The value to sweep would be MINIMAL_WAIT_TIME and if between push_tiles() and wait_tiles() is less than MINIMAL_WAIT_TIME then push_tiles() should have an extra delay to start the matmul

there should be a check on both ends:
the time between end and start of next block of data should be either less than N or greater than K:

  • N = 200
  • K = 2000
    (initial value of the numbers)

Test this further with board team to observe how input current requirements change.

@ttmtrajkovic
Copy link
Contributor Author

Added support to rd/didt_delay

Could be invoked with:
TT_ENABLE_WORKLOAD_DELAY=25000 pytest tests/didt/test_lm_head_matmul.py::test_lm_head_matmul -k "1chips" --iterations 1000000

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

2 participants