Differences

This shows you the differences between two versions of the page.

--- cluster:179 [2019/06/27 13:54]
hmeij07 [GPU Allocation Problems]
+++ cluster:179 [2019/06/27 14:11]
hmeij07 [Solutions Problems]
@@ Line 20: / Line 20: @@
 ===== Solutions Problems =====
-Yea.
+Somehow, using static or dynamic resource, the physical GPU device ID allocated needs to be "remembered". One way might for the wrappers remove the allocated GPU device ID from a config file when CUDA_VISIBLE_DEVICES is set.  The "elim" can now report idle GPUs of non-allocated devices.  The schedulers' ''post_exec'' script could add that device ID back into the config file.
+Another solution would be to break the cycling CPU/GPU job into multiple jobs. The first CPU cycle can not start until first GPU cyle is done etc. That can be done with the //done(job_ID|"jobname")// or //ended(job_ID|"jobname")// parameters of ''bsub''.
 \\
 **[[cluster:0|Back]]**

DokuWiki