SCCKN does contain several compute nodes with GPUs for use in simulations. Use the queue "gpu" to run these jobs. You need to specify the resource "gpu" with the number of needed GPUs: "#$ -l gpu=X" (X=1..8). You can also use a parallel environments to additionally reserve multiple CPU cores for the job. The IDs of the reserved GPUs are set as SGE_GPU. Since CUDA_VISIBLE_DEVICES is also set, applications counting GPUs itself (like mumax3) need to use GPU 0.
There are also workstations with consumer GPUs that can be used (see below). On all nodes and workstations the OpenCL toolkit of Intel is installed to use the CPU from OpenCL applications. On systems with NVIDIA cards the specified CUDA toolkit is installed.
Node/Workstation | OS | Driver | Graphics Card | CUDA Capability | CUDA Version | OpenCL | OpenACC | Memory | Speed | Number of Cores | Notes |
---|---|---|---|---|---|---|---|---|---|---|---|
scc192 | openSUSE 15.5 | 550.54.14 | 4xL40 | 8.9 | 12.4 | 4x48GB | 2490 MHz | 4x18176 | OK | ||
spiderman | openSUSE 15.5 | 550.54.14 | 2xA100 | 8.0 | 12.4 | 1.2 | Yes tesla:cc80 | 2x40GB | 1410 MHz | 2x6912 | OK |
hulk | openSUSE 15.5 | 550.54.14 | 3xRTX 5000 | 7.5 | 12.4 | 1.2 | Yes tesla:cc75 | 3x16GB | 1815 MHz | 3x3072 | OK |
scc195-scc199 | openSUSE 15.5 | 550.54.14 | 8xRTX 2080 Ti | 7.5 | 12.4 | 1.2 | Yes tesla:cc75 | 8x11 GB | 1545 MHz | 8x4352 | OK |
scc146 | openSUSE 15.3 | 470.57 | 4xTesla V100 | 7.0 | TODO | 1.2 | Yes tesla:cc70 | 2x32GB 2x16GB | 1455 MHz | 4x5120 | TODO |
scc116, scc117 | openSUSE 15.5 | 470.239.06 | Tesla K80 | 3.7 | 11.4 | 1.2 | Yes tesla:cc35 | 2x12 GB | 824 MHz | 2x2496 | OK |
scc066 | openSUSE 15.5 | 470.239.06 | Tesla K20Xm | 3.5 | 11.4 | 1.2 | Yes tesla:cc35 | 6 GB | 732 MHz | 2688 | OK |
fano | openSUSE 15.3 | 14.12(14.50.2), 3.0.130.135 | Radeon HD 7970 | --- | - | 1.2 | --- | 2 GB | 925 MHz | 2048 | TODO |