-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
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
Set OPENBLAS_NUM_THREADS=1
on local Distributed workers
#47803
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We could also check if the environment variable is already defined by the user, and if it is, we don't override it?
Can we backport to 1.8 and/or 1.9? |
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS.
471bbc0
to
29faa50
Compare
For context, we hope that this will fix the Windows OOMs that we are seeing on CI. |
Does this also work for Pkg.add? I'm not sure how it forks Julia processes, or if this trick is already used there. If it uses Distributed, then this would automatically cover it? I tried locking where add is implemented, I think at least this file is related: |
Do you mean for the precompilation step that runs at the end of |
I can confirm that this reduces the commit charge significantly! Compare the commit values versus the working set values: Compare against an equivalent run on Let's let this ride for a bit on |
Now that JuliaLang/julia#47803 has been merged, we are using [significantly less commit charge](JuliaLang/julia#47803 (comment)), and should be capable of running with only 3GB/core allocated, rather than the currently rather wasteful 4GB/core.
Maybe fixes #45434 |
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit a8b3994)
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit a8b3994)
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit a8b3994)
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit a8b3994)
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit a8b3994)
Now that JuliaLang/julia#47803 has been merged, we are using [significantly less commit charge](JuliaLang/julia#47803 (comment)), and should be capable of running with only 3GB/core allocated, rather than the currently rather wasteful 4GB/core.
…julia#47803) This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS. (cherry picked from commit f77b8a6) (cherry picked from commit 72cb2fd)
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its `__init__()` method when we're not trying to enable threaded BLAS.
This should prevent LinearAlgebra from trying to increase our OpenBLAS thread count in its
__init__()
method when we're not trying to enable threaded BLAS.