The Fermilab Wilson cluster Institutional Cluster (WC-IC) is an HPC resource available for use by all of Fermilab. The cluster worker nodes are connected by a low latency high speed Infiniband network fabric. For maximum flexibility, the code uses the Open MPI package for controlling parallel calculations that can make use of any parallel network hardware.
The Wilson cluster is a development cluster and not designed or intended for long term use for production tasks.
A summary of the current configuration of the WC-IC is as follows:
- Conventional Worker Nodes (x86 with NO GPUs):
- 100 worker nodes with dual-socket eight-core Intel E5-2650v2 “Ivy Bridge” (2.6 GHz) processors, (total 16 cores/node), 8GB/core memory, ~280GB local scratch per worker node, Inter-node QDR (40Gbps) Infiniband connectivity for low latency data transfers.
- Accelerated Worker Nodes (hosts with GPUs, QDR Infiniband inter-node interconnect):
- 27 worker nodes with similar specifications as above but with 4 NVIDIA K40 GPUs. (Total 108 GPUs)
- 1 worker node with 2 NVIDIA P100 w/NVLINK GPUs. (wcgpu01)
- 1 worker node with 8 NVIDIA P100 GPUs. (wcgpu02)
- 4 worker nodes with 2 NVIDIA V100 GPU/host. (wcgpu03-06)
- 1 IBM Power9 with 4 NVIDIA V100 GPUs w/NVLINK. This worker node is similar to a SUMMIT worker node. (wcibmpower01)
- 1 Intel Knights Landing, 512 cores total. (wcknl01)
- Filesystems available on WC-IC:
- 334TB Lustre file-system accessible across the entire cluster & connected to the QDR Infiniband. [NOT BACKED UP]
- NAS attached work directory area with a modest quota per user accessible from the entire cluster. [NOT BACKED UP]
- NAS attached home area with a modest quota per user which requires a valid Kerberos ticket to access. [BACKED UP]
- For additional information about the filesystems please click here.