Computational Resources

CPU Resources

Each of the 6 EPYC Genoa CPU nodes has 384 logical cores, 1.5TB of memory, and 50Gb networking. This yields approximately 17 TFLOPS FP64 performance per node. 

Each of the 24 EPYC Rome CPU nodes has 256 logical cores, 1TB of memory, and 50Gb networking. This yields approximately 8.4 TFLOPS FP64 performance per node. 

Each of the 77 legacy Xeon CPU nodes has 32-40 CPUs, 64-256 (1 with 1536) GB of memory, and 10Gb networking. This yields approximately 0.8 TFLOPS FP64 performance per node.

GPU Resources

Each of the 5 quad-GPU nodes has 4 A100 or H100 GPUs with NVLink interconnect, 96 or 128 logical cores, 1024GB of ECC memory, and 100Gb networking.

Each of the 16 dual-GPU nodes has 2 A40 or A100 GPUs, 32 or 64 logical cores, 250GB or 500GB of ECC memory, and 50Gb networking.

Each H100 GPU has 18,432 CUDA cores, 456 fourth-generation Tensor cores, and 80GB HBM3 ECC memory. This results in 67 TFLOPS FP32, 34 TFLOPS FP64, 67 TFLOPS FP64 Tensor, 986 TFLOPS FP32 Tensor, 1979 TFLOPS FP 16 Tensor per GPU.

Each A100 GPU has 10,752 CUDA cores, 432 third-generation Tensor cores, and 40GB HBM2E ECC memory. This results in 78 TFLOPS FP16, 19.5 TFLOPS FP32, 9.7 TFLOPS FP64, 19.5 TFLOPS FP64 Tensor, 156 TFLOPS FP32 Tensor, 312 TFLOPS FP 16 Tensor per GPU.

Each A40 GPU has 10,752 CUDA cores, 84 second-generation RT cores, 336 third-generation Tensor cores, and 48GB of GDDR6 ECC memory. This results in 37.4 TFOPS FP32, 149.7 TFLOPS Tensor FP16, 74.8 TFLOPS Tensor FP32, and 73.1 TFLOPS RT per GPU.