Compatible, tested or recommended julia versions for CPU and GPU? #295
Assignees
Labels
enhancement
New feature or request
Comments
|
Everything looks wonderful with julia 1.8!
|
|
Hi! We greatly appreciate your observations 😃. We have been working in KomaMRI 0.8 which should be compatible with Julia 1.9 and 1.10. It's strange that KomaMRI 0.7.5 isn't working in your environment with Julia 1.9. I just checked in my environment, and my Nvidia GPU is working. Could you share the steps and log messages from your machine? Regarding one of your questions, the default behavior of KomaMRI is to detect any GPU devices and use one of them for simulation; otherwise, the CPU is utilized. This behavior is true for all Julia versions. As you mentioned, indicating passing tests for GPU in the documentation is a great idea. So far, we only test GPU on local machines, and we are actively working on implementing automatic GPU tests (#147), similar to what we already have for CPU tests. Could you please check when the new version of KomaMRI 0.8.0 is released tomorrow, if there are any problems in your specific environment? |
|
Adding a table in the readme/documentation with the compatibility results (of the current Koma version) for CPU and GPU is a great idea. I am thinking of something like this:
And when we use start using Builkite (#147):
Without using Buildkite, separating the results of the CI into different badges per Julia version could be done with #276. Or we could do it manually, but I am not a big fan of that. The only package we really need to test in the GPU is KomaMRICore, I think the best would be to start testing with Buildkite. |
It was julia 1.10.0 that was not working. I did not check 1.9, but guessed it was better to try 1.8 based on #145. It sounds like the 1.9 and cuda issues are now resolved. See below for 1.9 which works fine.
Yes, will do! |
I'm not familiar with the specific tools mentioned, but sounds great, it should help users navigate the ever changing julia, cuda, etc. landscape. Thanks again! |
|
Here is the situation with julia 1.10: julia is hung at that point and will exit to shell with ctr-D system info: |
|
Thank you for your fast response @curtcorum. |
|
The new version 0.8.0 of KomaMRI has been released 😄! If you face any problems, make sure that after running: using Pkg
begin
println("OS $(Base.Sys.MACHINE)") # OS
println("Julia $VERSION") # Julia version
# Koma sub-packages
for (_, pkg) in filter(((_, pkg),) -> occursin("KomaMRI", pkg.name), Pkg.dependencies())
println("$(pkg.name) $(pkg.version)")
end
endyou have the following output: |
|
We added a table to the README with the current compat. I will close the issue, but feel free to re-open it if you feel we are missing something 😄. |
|
FYI all looks great on the main page, very clear what is expected to work with Julia versions. I did an upgrade to KomaMRI latest, and all seems well:
|
|
Julia 1.10 new environment also works:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Feature Request
First, thanks so much for sharing KomaMRI with the community, and for the great introduction at the ISMRM pulseq virtual meeting!
I am new to julia, but familiar with many of the other usual suspects (python, conda, etc.)
As a suggestion: It seems like julia <= 1.8 is currently required for using gpu functions (KomaMRICore v0.7.6)? Also the simulations default to gpu if present. Julia 1.9 works for cpu otherwise?
I went down the dead end with Julia 1.10.0, which is now the default, for a while before figuring it out.
Is it possible to add the gpu and cpu tested julia versions as an explicit dependency to the package (or in documentation or installation notes)?
It does seem like there are some version dependencies in the install files, but nothing warned or failed before runtime with julia 1.10 on ubuntu 18.04 and Juliaup 1.12.5.
thanks again!
Julia 1.9 and gpu compatibility #145
The text was updated successfully, but these errors were encountered: