The AMD-Xilinx Heterogeneous Compute Cluster is hosted within the Coordinated Science Lab at UIUC. The cluster was designed to support high-performance computing (HPC), machine learning (ML), and genomics applications, and for investigating programming models, compilers, schedulers, systems and architectures. It scalable and expandable, and encompasses modern FPGAs, multi-GPU nodes, near-memory acceleration devices, and emerging compute devices.
The cluster is comprised of 6 compute nodes + 4 Support nodes.
- 1 x Head node – Many core, large memory development node.
- 2 x Compute nodes – Accelerated compute node
- 2 x Big compute nodes – Accelerated compute nodes with support for up to eight accelerators per node
- 1 x Big Data node – FPGA + high end GPU node for big data analytics research
- 4 x Login nodes
The nodes are interconnected on a 100 Gb/s ethernet rack switch with a 1 Gb/s uplink to communicate outside of the cluster. For storage, there is a shared 24TB home filesystem. Each compute node also includes high speed scratch space.
Overall compute infrastructure details:
Node Open to External Use CPU Cores/Threads Memory Accelerator Resources
Head - 0 Yes 2 x Xeon Gold 6230 40/80 384 GB N/A
Compute - 0 Yes 2 x Xeon Gold 5217 16/32 192 GB 3 x Alveo U250
Compute - 1 Yes 2 x Xeon Gold 5217 16/32 192 GB 2 x Alveo U280+ 1 VCK 5000
Compute - 2 No 2 x Xeon Gold 5218 32/64 192 GB 1x Alveo U280 + 2 x Alveo U250
Compute - 3 No 2 x Xeon Gold 5218 32/64 192 GB 1 x Alveo U280 + Development boards
Compute - 4 2 x Xeon Gold 6230 40/80 128 GB 1 Alveo U250 + 2x Tesla V100
Login - 0 Yes Login - 1 Yes Login - 2 Yes XACC Web Login - 0 Yes