The cluster is made up of 1 head node and 10 compute nodes. Of the compute nodes, 8 are AMD based systems while the other 2 are Intel based systems. Here is the following hardware stack for all of the nodes:
Repairs The DGX1 node is currently unavailable due to faulty GPU board. There is no ETA on when this will be repaired.
|Node Type||Node Name||Processor||RAM||HDD (
||AMD Opteron 6320 (8 cores)||64 GB||916 GB||—|
|AMD Compute Node||
||4x AMD Opteron 6376 (64 cores)||512 GB||40 GB||NVIDIA K20 GPU, (2x in gpu06)|
|AMD Compute Node (Large RAM)||
||4x AMD Opteron 6376 (64 cores)||1024 GB||40 GB||NVIDIA K20 GPU (only in
|Intel Compute Node||
||2x Intel Xeon E5-2650v2 (16 cores)||128 GB||40 GB||—|
|NVIDIA DGX1 Node||
||2x Intel Xeon E5-2698v4 (40 cores)||512 GB||440 GB (7 TB at
||8x NVIDIA Tesla P100|
In addition to the above listed nodes, there are also some special-purpose nodes — specifically, the cluster has eight Intel MIC nodes:
|Node Type||Node Name||Processor||Onboard Memory||Extra Components|
|Intel MIC Compute Units||
||Intel Xeon Phi 5120D||8 GB||—|
All of the nodes are connected by both a 1Gbps network and an InfiniBand 4x 10Gbps network. We support the full OFED stack.
The nodes are made available through the SLURM resource management system — more information on queues can be found here.
Details about the OS and available software packages can be found on the software information page.