[GE users] Graphic cards processors

Joe Landman landman at scalableinformatics.com
Wed Jul 9 17:11:22 BST 2008


    [ The following text is in the "ISO-8859-2" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Jacek Strzelczyk wrote:
> Joe Landman wrote:
>> Jacek Strzelczyk wrote:
>>> Hello everyone,
>>>
>>> If there is a possibility in the Grid Engine (or commercial SGE) to 
>>> add a graphic cards processors as a computation resource? I would 
>>> like to make some calculations on those processors. Can I make them 
>>> available in some way through SGE?
>>
>> Yes.  Create a consumable resource per node.  Create a queue that uses 
>> this consumable resource.  Launch jobs into that.
>>
>> Note:  you will need to write in Cuda or similar (we are doing this 
>> for customers).
>>
>>
> 
> When I searched for "consumable resource" I found on the SGE page:
> 
> "Consumable resources provide an efficient way to manage limited 
> resources such as available memory, free space on a file system, network 
> bandwidth, or floating software licenses."
> 
> Ok, but how to tell the SGE that it should use the graphical card CPU 
> instead of normal CPU? How to identify those resources?

Hi Jacek

   The way we have done things like this is to define 1 GPU per GPU node 
(that is all the GPU cores are considered part of a single processing 
element on the node).  Each node that has a GPU gets 1 GPU resource. 
When you submit a job for the GPU, it will run, in part, on the host 
processor, as well as on the GPU itself.  You cannot run a job on one 
core of the GPU, these are dedicated specialized processors.

   To submit the job requesting a GPU (assuming you named the consumable 
resource to be "GPU") you would

	qsub -l GPU=1 ...

where ... are the rest of your job arguments and script.  To define the 
consumable, you can do this through the gui.  Create a complex, make it 
requestable, consumable, default of 1, and an integer type.

Then add this complex to any compute nodes with the GPU.  Once this is 
done, you can submit jobs as above.

I think Chris D has some pages on the wiki about this.

Joe


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
        http://jackrabbit.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 866 888 3112
cell : +1 734 612 4615

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list