GPU Partitions

Nodes in these partitons provide GPUs for parallelizing calculations. See GPU Usage for more details on how to use GPU partitions, particularly those where GPUs are split into MiG slices.

Partitions

The partitions are listed in the table below without hardware details.

Cluster	Partition	OS	Shared	Max. walltime	Max. nodes per job	Core-hours per node
NHR	grete	Rocky 8		48 hr	16	600
	grete:shared	Rocky 8	yes	48 hr	1	150 per GPU
	grete:interactive	Rocky 8	yes	48 hr	1	150/47 per GPU/slice
	grete:preemptible	Rocky 8	yes	48 hr	1	150/47 per GPU/slice
	grete-h100	Rocky 8		48 hr	16	1050
	grete-h100:shared	Rocky 8	yes	48 hr	16	262.5
	kisski	Rocky 8		48 hr	16	600
	kisski-h100	Rocky 8		48 hr	16	1050
	react	Rocky 8		48 hr	16	600
	scc-a100	Rocky 8		48 hr	16	600
	jupyter:gpu (jupyter)	Rocky 8	yes	24 hr	1	150/47 per GPU/slice
SCC	gpu	Rocky 8	yes	48 hr	max	24 per GPU
	gpu-int (jupyter)	Rocky 8	yes	48 hr	max	5 per GPU
	vis	Rocky 8	yes	48 hr	max	5 per GPU

Warning

The partitions you are allowed to use may be restricted by your kind of account. For example, only KISSKI users can use the kisski and kisski-h100 partitions.

Info

JupyterHub sessions run on the partitions marked with jupyter in the table above.

Info

Three GPU nodes featuring 4x A100 80GB cards are not directly accessible from SCC for technical reasons, but are instead connected in the NHR cluster temporarily (scc-a100 partition). SCC users wanting to use them should request a project in our HPC Project Portal to access these nodes by emailing hpc-support@gwdg.de. The nodes can be accessed via the NHR login nodes and they use the NHR software stacks as well as our new VAST filesystem.

The NHR nodes are grouped into subclusters of different hardware. The partitions for each NHR subcluster are listed in the table below where the parts of partition names in square brackets are optional (e.g. grete-h100[:*] includes grete-h100 and grete-h100:shared).

NHR Subcluster	GPUs	Partitions
Grete Phase 1	Nvidia V100	`jupyter:gpu`
Grete Phase 2	Nvidia A100	`grete[:*]`, `kisski`, `react`, `scc-a100`
Grete Phase 3	Nvidia H100	`grete-h100[:*]`, `kisski-h100`

The hardware for the different nodes in each partition are listed in the table below. Note that some partitions are heterogeneous, having nodes with different hardware. Additionally, many nodes are in more than one partition.

Partition	Nodes	GPU + slices	VRAM each	CPU	RAM per node	Cores
grete	27	4 × Nvidia A100	40 GB	2 × Zen3 EPYC 7513	497 810 MB	64
	25	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	497 810 MB	64
grete:shared	33	4 × Nvidia A100	40 GB	2 × Zen3 EPYC 7513	497 810 MB	64
	22	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	497 810 MB	64
	5	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	1 013 000 MB	64
	2	8 × Nvidia A100	80 GB	2 × Zen2 EPYC 7662	1 013 620 MB	128
grete:interactive	3	4 × Nvidia A100 (2g.10gb and 3g.20gb)	10/20 GB	2 × Zen3 EPYC 7513	497 810 MB	64
grete:preemptible	3	4 × Nvidia A100 (2g.10gb and 3g.20gb)	10/20 GB	2 × Zen3 EPYC 7513	497 810 MB	64
grete-h100	5	4 × Nvidia H100	94 GB	2 × Xeon Platinum 8468	1 029 200 MB	96
grete-h100:shared	5	4 × Nvidia H100	94 GB	2 × Xeon Platinum 8468	1 029 200 MB	96
kisski	59	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	497 810 MB	64
	4	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	1 013 000 MB	64
kisski-h100	11	4 × Nvidia H100	94 GB	2 × Xeon Platinum 8468	1 029 200 MB	96
react	22	4 x Nvidia A100	80 GB	2 × Zen3 EPYC 7513	497 810 MB	64
scc-a100	2	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	1 013 000 MB	64
	1	4 × Nvidia A100	80 GB	2 × Zen3 EPYC 7513	497 810 MB	64
jupyter:gpu	3	4 × Nvidia V100	32 GB	2 × Skylake 6148	746 000 MB	40
gpu	2	8 × Nvidia V100	32 GB	2 × Cascadelake 6252	362 000 MB	48
	13	4 × Nvidia RTX5000	16 GB	2 × Cascadelake 6242	172 000 MB	32
	3	4 × Nvidia GTX1080	8 GB	2 × Broadwell E5-2650v4	128 000 MB	24
gpu-int	2	4 × Nvidia GTX980	4 GB	2 × Broadwell E5-2650v4	128 000 MB	24
vis	3	4 × Nvidia GTX980	4 GB	2 × Broadwell E5-2650v4	128 000 MB	24

The CPUs and GPUs

For partitions that have heterogeneous hardware, you can give Slurm options to request the particular hardware you want. For CPUs, you can specify the kind of CPU you want by passing a -C/--constraint option to slurm to get the CPUs you want. For GPUs, you can specify the name of the GPU when you pass the -G/--gpus option (or --gpus-per-task) and larger VRAM using a -C/--constraint option. See Slurm and GPU Usage for more information.

The GPUs, the options to request them, and some of their properties are given in the table below.

GPU	VRAM	FP32 cores	Tensor cores	`-G` option	`-C` option	Compute Cap.
Nvidia A100	40 GB	6912	432	`A100`		80
	80 GB	6912	432	`A100`	`80gb`	80
Nvidia H100	94 GB	8448	528	`H100`	`96gb`	90
2g.10gb slice of Nvidia A100	10 GB	1728	108	`2g.10gb`		80
3g.20gb slice of Nvidia A100	20 GB	2592	162	`3g.20gb`		80
Nvidia V100	32 GB	5120	640	`V100`		70
Nvidia Quadro RTX 5000	16 GB	3072	384	`RTX5000`		75
Nvidia GeForce GTX 1080	8 GB	2560		`GTX1080`		61
Nvidia GeForce GTX 980	4 GB	2048		`GTX980`		52

The CPUs, the options to request them, and some of their properties are give in the table below.

CPU	Cores	`-C` option	Architecture
AMD Zen3 EPYC 7513	32	`zen3` or `milan`	`zen3`
AMD Zen2 EPYC 7662	64	`zen2` or `rome`	`zen2`
Intel Sapphire Rapids Xeon Platinum 8468	48	`sapphirerapids`	`sapphirerapids`
Intel Cascadelake Xeon Gold 6252	24	`cascadelake`	`cascadelake`
Intel Cascadelake Xeon Gold 6242	16	`cascadelake`	`cascadelake`
Intel Skylake Xeon Gold 6148	20	`skylake`	`skylake_avx512`
Intel Broadwell Xeon E5-2650 V4	12	`broadwell`	`broadwell`

Hardware Totals

The total nodes, cores, GPUs, RAM, and VRAM for each cluster and sub-cluster are given in the table below.

Cluster	Sub-cluster	Nodes	GPUs	VRAM (TiB)	Cores	RAM (TiB)
NHR	Grete Phase 1	3	12	0.375	120	2.1
	Grete Phase 2	103	420	27.1	6,720	47.6
	Grete Phase 3	16	64	6.0	1,536	15.7
	TOTAL	122	496	33.5	8,376	65.4
SCC	TOTAL	21	92	1.4	656	3.5