These are the available nodes in the MCC cluster (updated 25/10/2023).
To obtain this information, run sinfo --long --Node --noheader | awk '!seen[$1]++' on the cluster for the partition dimensions. For CPU and GPU look at /proc/cpuinfo or similar tools.
Indices of partition indicate number of available nodes within a partition, e.g. rome partition has seven available nodes.
| Node | Cores | Threads:Cores:Sockets | CPU | GPU | Memory GB | OS | Python |
|---|---|---|---|---|---|---|---|
| dhabi[1-9] | 64 | 8:8:1 | AMD Opteron 6376 | N/A | 1000 | Ubuntu 18.04.5 bionic | 2.7.18, 3.8.6, 3.9.2 |
| naples[1-9] | 64 | 8:8:1 | AMD EPYC 7551 | N/A | 500 | Debian oldstable-updates sid | 2.7.18, 3.8.6, 3.9.2 |
| rome[1-7] | 96 | 2:48:1 | AMD EPYC 7642 | N/A | 1000 | Ubuntu 20.04 focal | 2.7.18, 3.8.10 |
| turing[1-2] | 64 | 2:16:2 | AMD EPYC 7302 | 6x Tesla T4 (TU104GL) | 500 | Ubuntu 20.04 focal | 2.7.18, 3.8.10 |
| vmware[1-4] | 1 | 1:1:1 | Intel Xeon E5-2680 v4 | N/A | 1 | Ubuntu 18.04.5 bionic | N/A |
The two nodes of the turing partition has different driver- and software-stacks as depicted below, this is to accomodate projects which require newer features.
Note that turing02 might be unstable due to this newer stack, report any issues and requests to falke.
| Node Name | CUDA Driver | CUDA Toolkit |
|---|---|---|
turing01 |
cuda-driver-450 | cuda-toolkit-11-0 |
turing02 |
cuda-driver-535 | cuda-toolkit-12.2 |