Vulcan cluster¶
Overview¶
The NEC Cluster (Vulcan) is a heterogeneous university cluster with AMD and NVIDIA GPUs, available to the University of Stuttgart for research, visualization, AI, and big data.
See page NEC Cluster Hardware and Architecture for more information. See the HLRS status page for maintenance information.
Entgeltordnung |
Intended use |
Nodes |
Cores |
RAM |
GPU |
Node list |
|
|---|---|---|---|---|---|---|---|
smp Knoten |
(Pre- / Post-) processing |
1 |
2x20 |
1536 GB |
1x K6000 |
|
|
Cascadelake 384 GB |
Data processing |
96 |
2x20 |
384 GB |
|
||
CS500-Spark |
Data processing & storage |
8 |
2x20 |
384 GB |
|
||
CS Storm |
AI and big data |
8 |
2x18 |
768 GB |
8x V100 |
|
|
Haswell 128 GB |
Data processing |
76 |
2x10 |
128 GB |
|
||
Haswell 256 GB |
Data processing |
4 |
2x10 |
256 GB |
|
||
Haswell 128 GB |
Data processing |
76 |
2x12 |
128 GB |
|
||
Haswell 256 GB |
Data processing |
16 |
2x12 |
256 GB |
|
||
SKL-192GB |
Data processing |
88 |
2x20 |
192 GB |
|
||
Graphikserver AMD |
Visualization |
6 |
2x4 |
96 GB |
1x WX 8200 |
|
|
Graphikserver NVIDIA |
Visualization |
1 |
2x4 |
96 GB |
1x RTX 4000 |
|
|
VISP 100 |
Data processing & Visualization |
10 |
2x8 |
256 GB |
1x P100 |
|
|
Partitions and nodes¶
Vulcan uses the PBS scheduling system!
smp Knoten¶
There is 1 node equipped with 2 Intel Xeon Gold 6148 (20 cores, 40 threads, 2.40 GHz, 150 W), 1536 GB of RAM, 15 TB SSD. Each node has 1 NVIDIA Quadro K6000 (288 Gbps, 225 W, 12 GB GDDR5).
Cascadelake 384 GB¶
There are 96 nodes equipped with 2 Intel Xeon Gold 6248 (20 cores, 40 threads, 2.50 GHz, 150 W), 384 GB of RAM. There are no GPUs.
CS500-Spark¶
There are 8 nodes equipped with 2 Intel Xeon Gold 6230 (20 cores, 40 threads, 2.10 GHz, 125 W), 384 GB of RAM, 1.8 TB NVMe SSD. There are no GPUs.
CS Storm¶
There are 8 nodes equipped with 2 Intel Xeon Gold 6240 (18 cores, 36 threads, 2.60 GHz, 150 W), 768 GB of RAM, 7.3 TB NVMe SSD, 220 GB SSD. Each node has 8 NVIDIA Tesla V100 SXM2 (300 Gbps, 300 W, 32 GB HBM2).
Haswell 128 GB 20 cores¶
There are 76 nodes equipped with 2 Intel Xeon E5-2660v3 (10 cores, 20 threads, 2.60 GHz, 105 W), 128 GB of RAM. There are no GPUs.
Haswell 256 GB 20 cores¶
There are 4 nodes equipped with 2 Intel Xeon E5-2660v3 (10 cores, 20 threads, 2.60 GHz, 105 W), 256 GB of RAM. There are no GPUs.
Haswell 128 GB 24 cores¶
There are 152 nodes equipped with 2 Intel Xeon E5-2668v3 (12 cores, 24 threads, 2.50 GHz, 120 W), 128 GB of RAM. There are no GPUs.
Haswell 256 GB 24 cores¶
There are 16 nodes equipped with 2 Intel Xeon E5-2668v3 (12 cores, 24 threads, 2.50 GHz, 120 W), 128 GB of RAM. There are no GPUs.
SKL-192GB¶
There are 16 nodes equipped with 2 Intel Xeon Gold 6138 (20 cores, 40 threads, 2.00 GHz, 125 W), 192 GB of RAM. There are no GPUs.
Graphikserver AMD¶
There are 6 nodes equipped with 2 Intel Xeon Silver 4112 (4 cores, 8 threads, 2.60 GHz, 80 W), 96 GB of RAM. Each node has 1 AMD Radeon Pro WX 8200 (512 Gbps, 230 W, 8 GB HBM2).
Graphikserver NVIDIA¶
There is 1 node equipped with 2 Intel Xeon Silver 4112 (4 cores, 8 threads, 2.60 GHz, 80 W), 96 GB of RAM. Each node has 1 NVIDIA Quadro RTX 4000 (400 Gbps, 160 W, 8 GB GDDR6).
VISP 100¶
There are 16 nodes equipped with 2 Intel Xeon E5-2667v4 (8 cores, 16 threads, 3.20 GHz, 135 W), 256 GB of RAM, 4.1 TB SSD. Each node has 1 NVIDIA Tesla P100 (549 Gbps, 250 W, 12 GB CoWoS HBM2).
Access¶
Please refer to the HLRS access policy.
Once access is granted, refer to the Vulcan user documentation.
Publications¶
[Zhong et al., 2023]: benchmarking of Tensorflow DNN on CS-Storm coupled to FEM on Hawk
[Zhou et al., 2019]: benchmarking of MPI communication strategies on Vulcan and Hazel Hen
[Bispo et al., 2021]: guidelines for NVIDIA accelerators on various hardware, including Vulcan
[HPCG, 2020]: Vulcan ranked in 123rd place of the conjugate gradient benchmark