Casper

4/7/2021: Casper nodes are now accessible through PBS Pro rather than Slurm. See Starting Casper jobs with PBS and this related documentation for more information about the transition to PBS Pro.

The Casper cluster is a heterogeneous system of specialized data analysis and visualization resources; large-memory, multi-GPU nodes; and high-throughput computing nodes.

Three photos of the Casper cluster of nodes.

Casper is composed of 100 nodes featuring Intel Skylake or Cascade Lake processors.

  • 22 Supermicro SuperWorkstation nodes are used for data analysis and visualization jobs. Each node has 36 cores and up to 384 GB memory.
  • 10 nodes feature large-memory, dense GPU configurations to support explorations in machine learning (ML) and deep learning (DL) and general-purpose GPU (GPGPU) computing in atmospheric and related sciences.
  • 64 high-throughput computing (HTC) nodes for small computing tasks using 1 or 2 CPUs.
    • 62 HTC nodes have 384 GB of available memory
    • 2 HTC nodes have 1.5 TB of available memory
  • 4 nodes are reserved for Research Data Archive workflows.

See the hardware summary table below for detailed specifications.

Operating system: CentOS 7.8


Hardware

Data Analysis
& Visualization nodes

22 Supermicro 7049GP-TRT SuperWorkstation nodes
Up to 384 GB DDR4-2666 memory per node
2 18-core 2.3-GHz Intel Xeon Gold 6140 (Skylake) processors per node
2 TB local NVMe Solid State Disk
1 Mellanox ConnectX-4 100Gb Ethernet connection (GLADE, Campaign Storage, external connectivity)
1 Mellanox ConnectX-6 HDR100 InfiniBand link
1 NVIDIA Quadro GP100 GPU 16GB PCIe on each of 9 nodes

Machine Learning/Deep Learning 
& General Purpose GPU (GPGPU) nodes

4 Supermicro SuperServer nodes with 4 V100 GPUs
768 GB DDR4-2666 memory per node
2 18-core 2.3-GHz Intel Xeon Gold 6140 (Skylake) processors per node
2 18-core 2.6-GHz Intel Xeon Gold 6240 (Cascade Lake) processors per node
2 TB local NVMe Solid State Disk
1 Mellanox ConnectX-4 100Gb Ethernet connection (GLADE, Campaign Storage, external connectivity)
2 Mellanox ConnectX-6 HDR200 InfiniBand adapters. HDR100 link on each CPU socket
4 NVIDIA Tesla V100 32GB SXM2 GPUs with NVLink

6 Supermicro SuperServer nodes with 8 V100 GPUs
1152 GB DDR4-2666 memory per node
2 18-core 2.3-GHz Intel Xeon Gold 6140 (Skylake) processors per node
2 TB local NVMe Solid State Disk
1 Mellanox ConnectX-4 100Gb Ethernet connection (GLADE, Campaign Storage, external connectivity)
2 Mellanox ConnectX-6 HDR200 InfiniBand adapters, HDR100 link on each CPU socket
8 NVIDIA Tesla V100 32GB SXM2 GPUs with NVLink

High-Throughput Computing nodes

62 small-memory workstation nodes
394 GB DDR4-2666 memory per node 
2 18-core 2.6-GHz Intel Xeon Gold 6240 (Cascade Lake) processors per node
1.6 TB local NVMe Solid State Disk
1 Mellanox ConnectX-5 100Gb Ethernet VPI adapter (GLADE, Campaign Storage, external connectivity)
1 Mellanox ConnectX-6 HDR200 InfiniBand VPI adapter. HDR100 link on each CPU socket

2 large-memory workstation nodes
1.5 TB DDR4-2666 memory per node 
2 18-core 2.3-GHz Intel Xeon Gold 6240 (Cascade Lake) processors per node
1.6 TB local NVMe Solid State Disk
1 Mellanox ConnectX-5 100Gb Ethernet VPI adapter (GLADE, Campaign Storage, external connectivity)
1 Mellanox ConnectX-6 HDR200 InfiniBand VPI adapter, HDR100 link on each CPU socket

Research Data Archive
nodes (reserved for
RDA use)

4 Supermicro Workstation nodes
94 GB DDR4-2666 memory per node
2 16-core 2.3-GHz Intel Xeon Gold 5218 (Cascade Lake) processors per node
1.92 TB local Solid State Disk
1 Mellanox ConnectX-6 VPI 100Gb Ethernet connection (GLADE, Campaign Storage, internal connectivity)