Hawk supercomputer¶
Overview¶
Hawk is a Tier-0, 26 PFLOPS peak performance, Hewlett Packard Enterprise Apollo supercomputer at the HLRS, available to the University of Stuttgart for research.
See the PDF slides linked on the HPE Hawk Hardware and Architecture page for more information. See the HLRS status page for maintenance information.
Type |
Nodes |
Cores |
RAM |
GPU |
Partition |
|---|---|---|---|---|---|
CPU |
5632 |
2x64 |
256 GB |
|
|
GPU |
24 |
? |
? |
8x A100 |
|
Partitions and nodes¶
Hawk uses the PBS scheduling system!
CPU nodes¶
There are 5632 nodes equipped with 2 AMD EPYC 7742 (64 cores, 128 threads, 2.25 GHz, 200 W), 256 GB of RAM (380 Gbps), Dual Rail InfiniBand HDR interconnect (200 Gbps). There are no GPUs.
GPU nodes¶
There are 24 nodes equipped with 2 CPUs, Dual Rail InfiniBand HDR interconnect (200 Gbps). Each node has 8 NVIDIA Tesla A100.
Access¶
Please refer to the HLRS access policy.
Access can also be gained through calls from the Gauss Centre for Supercomputing (instructions).
Once access is granted, refer to the Hawk user documentation.
Obligations¶
Usage of Hawk comes with reporting obligations and acknowledgment requirements. The specific details vary depending on whether Hawk resources were acquired through EuroHPC (instructions), the HLRS (instructions), or the Gauss Centre for Supercomputing (instructions).
Publications¶
[Zhong et al., 2023]: benchmarking of Tensorflow DNN on CS-Storm coupled to FEM on Hawk
[HPCG, 2020]: Hawk ranked in 18th place of the conjugate gradient benchmark