Raj System Overview

Compute Nodes

General Compute Nodes

38 nodes consisting of two AMD Rome 64 core 2 GHz processors, 512 GB RAM and 1.92 TB SSD scratch space.

Large Memory Compute Nodes

11 nodes consisting of two AMD Rome 64 core 2 GHz processors, 1 TB RAM and 1.92 TB SSD scratch space

Massive Memory Compute Nodes

Four nodes consisting of two AMD Rome 64 core 2 GHz processors, 2 TB RAM and 1.92 TB SSD scratch space.

GPU Compute Nodes

12 nodes consisting of two AMD Rome 64 core 2 GHz processors, 2 NVIDIA Tesla V100 GPU accelerators, 512 GB RAM and 1.92 TB SSD scratch space.

Artificial Intelligence/Machine Learning Nodes

Three nodes consisting of two Intel Cascade Lake 18 core 2.6 GHz processors, eight NVIDIA Tesla V100 GPU accelerators, NVLink high speed GPU interconnect, 768 GB RAM and 7 TB SSD NVMe scratch space.

Storage

Raj has a two-tiered storage system utilizing IBM's Spectrum Scale GPFS to present both tiers as a single namespace, optimizing performance and capacity while maintaining simplicity for the end user.

Tier 0 Storage

123 TB NVMe SSD storage utilizing Excelero NVMesh software.

Tier 1 Storage

1.2 PB HDD storage.

Network

Compute nodes and storage are connected via a 100 GB/s Infiniband network.