Vps – Hosting with CUDA support

If your system is fully AMD-Vi or VT-d compliant then you should be able to go into the host's advance settings and 'pass through' the device/s to a given VM. This isn't certain to work or be stable but does work for quite a lot of kit, though it stops things like vMotion/FT/HA working properly if that matters to you. I'd say give it a go but if it doesn't work, well it doesn't work ok.

edit - here's grab of where to look;

So it's host/Configuration, Advanced Settings then choose 'Configure Passthrough' - you're on your own from there though as I haven't done this with a GPU.

The strategy is actually fairly simple.

Using qconf -mc you can create a complex resource called gpu (or whatever you wish to name it). The resource definition should look something like:

#name               shortcut   type        relop   requestable consumable default  urgency     
#----------------------------------------------------------------------------------------------
gpu                 gpu        INT         <=      YES         YES        0        0

Then you should edit your exec host definitions with qconf -me to set the number of GPUs on exec hosts that have them:

hostname              node001
load_scaling          NONE
complex_values        gpu=2
user_lists            NONE
xuser_lists           NONE
projects              NONE
xprojects             NONE
usage_scaling         NONE
report_variables      NONE

Now that you've set up your exec hosts, you can request gpu resources when submitting jobs. eg: qsub -l gpu=1 and gridengine will keep track of how many GPUs are available.

If you have more than one job running per node that uses a GPU you may want to place your GPUs in to exclusive mode. You can do this with the nvidia-smi utility.