What you could do is creating a queue for each GPU you have on a host and assign them a queue exclusive GPU complex. The amount of GPU queues are limiting then the amount of GPU jobs. Then the total amount of cpu cores must be limited differently by a RQS on a per host basis.
Daniel Am 23.08.2012 um 12:19 schrieb Reuti: > Am 23.08.2012 um 09:32 schrieb William Hay: > >> <snip> >> You have consumable set to YES which means the request is multiplied >> by the number of slots you request 6>4 so you can't allocate them. >> Assuming you have a recent version of Grid Engine you can set the >> consumable setting to JOB. This will mean that the gpu resource is >> allocated for the MASTER slot of the job. Unfortunately no version of >> grid engine (AFAIK) supports per-host resource consumption but as you >> are using a $pe_slots PE it shouldn't matter. > > Correct. IIRC it was in the original proposal for the new handling of > consumables but never implemented. > > -- Reuti > _______________________________________________ > users mailing list > users@gridengine.org > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
