[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] cron and specific slots
- Date: Tue, 2 Nov 2010 15:37:33 -0400
- From: Michael Di Domenico <mdidomenico4@xxxxxxxxx>
- Subject: Re: [Condor-users] cron and specific slots
On Tue, Nov 2, 2010 at 2:50 PM, Burnett, Ben <ben.burnett@xxxxxxxx> wrote:
> Nice. Thanks :)
> Are your GPU jobs CPU/IO/etc. intensive? I ask because It's not entirely clear to me on our grid if they are, so I'm unsure if having an existing slot tied to a GPU is a waste of a slot (i.e. that it could otherwise be doing other work), or if I should continue with the idea of having GPU slots that are somehow independent of the machine slots. I suppose it depends on the nature of the job, and how much of the work is offloaded to the GPU.
We've generally found that in order to effectively drive GPU you need
one full CPU as well. Obviously this is completely code dependent.
Our typical use case
CPU - stage data
CPU - pre process
CPU - load to GPU
GPU - process data
- CPU - stage some new data
- CPU - pre process
CPU - unload data from GPU
In this instance, it's very handy to have the slot CPU along with the
GPU, this way we can load and stage the data while the GPU is doing
the calculations. If we time it right, we're making sure the GPU has
a constant feed of data.
But obviously there are some use cases where the CPU is doing next to
nothing just waiting for the GPU to finish. We have those as well,
but we're okay with the CPU waste in that instance.