SBG Nodes¶
We have 2 Lenovo ThinkSystem SR670 nodes containing Nvidia Tesla V100 cards for GPU jobs.
| SBG | Lenovo ThinkSystem SR670 |
|---|---|
| Count | 2 |
| Processor | 2 x 16 Core Intel Xeon Gold 6142 (Skylake) |
| Cores/Node | 32 |
| RAM | 384GB |
| TMP Size | 1.2TB |
| Interconnect | 25Gb Ethernet |
| GPU | 2 x NVIDIA Tesla V100 (up to maximum of 4) |
| GPU architecture | Volta |
| Tensor Cores | 640 |
| CUDA Cores | 5,120 |
| GPU Memory | 16GiB per GPU |
| CUDA Compute | 7.0 (CUDA version 9 or greater required) |
Accessing the GPU nodes¶
Access is permitted to QMUL researchers upon request. Note that access to GPU nodes is not permitted for Undergraduate and MSc students. Please raise a support ticket by emailing its-research-support@qmul.ac.uk with a brief overview of intended use, so that we can verify that the jobs will use the GPUs correctly. More detail is on the using GPUs section.
Half-node jobs¶
Please ensure that you specify system (as opposed to GPU) RAM carefully when
requesting a "half-node job" (i.e. to make use of a single GPU).
Remember that CPU memory is per core, so if you are using 1 GPU and 16 CPU
cores, setting -l h_vmem=11.5G, will request 184GB of system RAM.