GPU Queue

With the addition of a few GPU nodes to Fusion you can now take advantage of the processing power of these graphical processors to further increase the amount of work you can get done on Fusion with codes that can take full advantage of GPU's as a computing resource. These nodes are a bit beefier than normal fusion nodes not only in terms of having two GPU's at their disposal but also having 16 CPU cores and 64GB of memory. The GPU's are Nvidia's Tesla M2090 graphics cards which having 6GB's of on card memory and 512 cuda cores to give an estimated double precision floating point performance of 665 Gigaflops.

We have made a separate queue in which the GPU's can be used called 'GPU'. To submit to this queue you can simply add #PBS -q GPU to your batch script. * Note that there are only 2 nodes with 2 GPU's per node.

The GPU nodes are charged differently than normal nodes. For each node you consume you will be charged for at two times the rate or in the case of the GPU nodes 32 core hours.

Also make sure to add the following key in order to have the cuda toolkit at your disposal: +cudatoolkit

Everything else about these nodes are the same on any other Fusion node including having access to both the GPFS home filesystem and PVFS scratch space.

If you have any further questions about these nodes or using them please let us know by sending an email to support@lcrc.anl.gov.