Comparing Tahoma and Cascade
Cascade |
Tahoma |
|
|---|---|---|
Overall Size |
960 Intel Sandy-Bridge nodes, clock speed 2.6GHz Theoretical peak performance: 465 TFlops |
184 Intel Cascade Lake(tm) nodes, clock speed 3.1 GHz Theoretical peak performance: 1015 TFlops |
CPU/cores per node |
2 eight-core sockets = 16 Intel (c) Xeon(tm) E5-2670 cores |
2 x 18-core sockets = 36 Intel(c) Xeon(tm) Gold 6254 cores |
Memory per node |
128GB total = 8 GB/core |
384 GB total = 10 2/3 GB/core on t[1-160] 1536 GB total = 42 2/3 GB/core on ta[1-24] |
Disk |
2.8 PB global storage in Lustre file system Aggregate of 702 TB of local disk |
10 PB global storage in BeeGFS file system Aggregate of 536 TB local disk |
Accelerators |
56 nodes have 2X NVIDIA Tesla V100 16GB GPUs attached by PCI-express (PCIe x16) |
24 nodes (ta[1-24]) have 2X NVIDIA Tesla V100 32GB GPUs attached by PCI-express (PCIe x16) |
Network |
FDR Infiniband fabric (ConnectX-3) connected through two 648-port XS6536 Mellanox switches |
HDR-100 Infiniband fabric (ConnectX-6) connected through seven 40-port Mellanox MQM8790 switches. |
Current status * |
Note
The website showing the current status is only available for users onsite or within PNNL’s firewall.