Technical Specifications

Compute Nodes

Compute nodes rely on remote file systems (like the parallel file system BeeGFS), as they do not have local storage capabilities. The local file system resides in memory (RAM).

Regular

571 compute nodes each equipped with 256 GB RAM and 2 Intel Icelake sockets; each socket features a Intel(R) Xeon (R) Platinum 8360Y processor with (up to) 36 cores, yielding a total of 72 cores per node

Fat Memory

5 compute nodes each equipped with 1 TB RAM and 2 Intel Icelake sockets; each socket features a Intel(R) Xeon (R) Platinum 8360Y processor with (up to) 36 cores, yielding a total of 72 cores per node

GPU

GPU nodes have local storage and provide access to local scratch storage.

gpu01-05: A100-based Nodes

5 compute nodes each equipped with 256 GB RAM, 2 Intel Icelake sockets, 2 NVIDIA A100 (40GB) GPUs and 894 GB local scratch storage; each socket features an Intel(R) Xeon (R) Platinum 8360Y processor with (up to) 36 cores, yielding a total of 72 cores per node

• Recommended for: Double-precision computational tasks.
• Supported CUDA versions: 11.x or later. (cuda_arch=80)
• See the A100 datasheet and product brief for more information.

gpu06-09: L40S-based Nodes

4 compute nodes each equipped with 1 TB RAM, 2 Intel Icelake sockets, 8 NVIDIA L40S GPUs (with 48 GB memory) and 3.4 TB local scratch storage; each socket features an Intel(R) Xeon (R) Platinum 8360Y processor with (up to) 36 cores, yielding a total of 72 cores per node

• Recommended for: Machine learning or computations that primarily use half precision.
• Supported CUDA versions: 12.2 or later. (cuda_arch=89)
• See the L40S datasheet or product brief for more information.

File Systems

Besides, HSUper provides access to a 1PB BeeGFS file system, as well as a 1 PB Ceph-based file system.

Network

All nodes are connected by a non-blocking NVIDIA InfiniBand HDR100 fabric.