## page was renamed from GR2DOC/Tools/Grid/GPU/GPUsOnGrid
## page was renamed from Documentation/Tools/Grid/GPU/GPUsOnGrid
## page was renamed from Documentation/Tools/Grid/GPUsOnGrid
## page was renamed from Cluster/Operational/Tools/Grid/GPUsOnGrid
## page was renamed from Cluster/Operational/Tools/Grid/GPUOnGrid
#acl All:read

<<TableOfContents>>

= GPU on GRID =

{i} ~+See the required [[Documentation/Tools/Grid/GPU|underlying steps]] to enable GPU job submission.+~

== Use Case 1: Direct job submission to a CREAM service ==
 * Straightforward means of GPU computing on Grid.
 * Acomplished through `gpu` resource under JDL's `CERequirements`.
 * There is no GPU matchmaking: the grid user needs to know the endpoint to submit to.


 * A basic JDL example could be:
    {{{#!highlight bash numbers=disable
    # cat test_cream.jdl
    [ 
        JobType = "Normal"; 
        Executable = "foo.sh"; 
        StdOutput="out.out"; 
        StdError="err.err"; 
        InputSandbox={"foo.sh"}; 
        OutputSandbox={"out.out", "err.err" }; 
        OutputSandboxBaseDestUri="gsiftp://localhost";

        CERequirements="gpu==2";
    ]
    }}}

 * Submit the job directly to the CREAM service with `glite-job-ce-submit`.
 * Target CE: amargus01.ifca.es

== Use Case 2: Job submission through WMS ==