You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
v0.21 has gpu-enabled versions of synthstrip, synthseg, drbuddi and eddy included. Previously only eddy had gpu capabilities and these had to be accessed by a special eddy config json file.
I propose adding a --prefer-gpu flag that will switch each of these to their gpu version. There will still have to be an eddy config json file for specifying all the other options, but the "use_cuda" option will be taken out of the json file and specified on the commandline.
The text was updated successfully, but these errors were encountered:
Or maybe it would be better to have a --gpu [require|try|none] where require mode the will raise an error if the gpu version fails, try will try to use the gpu and fall back to the cpu version if the gpu fails and none won't attempt to use a gpu at all?
I'm just starting to run gpu jobs on an HPC and it looks like they randomly fail a lot
v0.21 has gpu-enabled versions of synthstrip, synthseg, drbuddi and eddy included. Previously only eddy had gpu capabilities and these had to be accessed by a special eddy config json file.
I propose adding a
--prefer-gpu
flag that will switch each of these to their gpu version. There will still have to be an eddy config json file for specifying all the other options, but the "use_cuda" option will be taken out of the json file and specified on the commandline.The text was updated successfully, but these errors were encountered: