The Cluster
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 (/tmp) |
Extra Components |
|---|---|---|---|---|---|
| Head Node | robotarium |
AMD Opteron 6320 (8 cores) | 64 GB | 916 GB | — |
| AMD Compute Node | gpu01, gpu02, gpu03, gpu04, gpu05, gpu06 |
4x AMD Opteron 6376 (64 cores) | 512 GB | 40 GB | NVIDIA K20 GPU, (2x in gpu06) |
| AMD Compute Node (Large RAM) | gpu07, gpu08 |
4x AMD Opteron 6376 (64 cores) | 1024 GB | 40 GB | NVIDIA K20 GPU (only in gpu07), 2x NVIDIA Titan Xp GPUs (only in gpu08) |
| Intel Compute Node | mic01, mic02 |
2x Intel Xeon E5-2650v2 (16 cores) | 128 GB | 40 GB | — |
| NVIDIA DGX1 Node | dgx01 |
2x Intel Xeon E5-2698v4 (40 cores) | 512 GB | 440 GB (7 TB at /raid/scratch) |
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 | mic01-0, mic01-1, mic01-2, mic01-3, mic02-0, mic02-1, mic02-2, mic02-3 |
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.