Current Cluster Comparisons
| Attribute | Terra | ViDal | Grace | FASTER | Lonestar6 * | ACES | Launch |
|---|---|---|---|---|---|---|---|
| Total SUs per Fiscal Year | 84,376,320 | 9,811,200 | 391,186,560 | 100,915,200 | 12,800,000 | 104,138,880 | ? |
| Peak DP FLOPS | 517 TF | 67 TF | 6.2 PF | 1.2 PF | 2.9 PF | 1.6 PF ** | 436 TFLOPS |
| Login Resources | |||||||
| # Nodes | 3 | N/A | 5 | 4 | 3 | 2 | 2 |
| # Cores | 84 | 240 | 256 | 384 | 192 | 8640 | |
| # Cores/Node | 28 | 48 | 64 | 128 | 96 | 32 | |
| Node Configurations | 2x 128GB 1x 128GB with 2 GPU | 2x 384GB 1x 384GB with 1 A100 GPU 1x 384GB with 2 T4 GPUs 1x 384GB with 2 RTX 6000 GPUs | 4x 256GB | 3x 256GB | 2x 512GB | 1x 384 GB 1x 384 GB with 1 A30 GPU | |
| Compute Resources | |||||||
| # Nodes | 320 | 24 | 925 | 180 | 560 | 130 | 45 |
| # Cores | 9,632 | 1,120 | 44,656 | 11,520 | 73,984 | 11,888 | 8,640 |
| # Cores/Node | 304x 28/node 8x 68/node 8x 72/node | 20x 40/node 4x 80/node | 917x 48/node 8x 80/node | 180x 64/node | 578x 128/node | 110x 96/node 1x 48/node 1x 128/node 18x 64/node | 45x 192/node |
| Node Configurations | 256x 64GB 48x 128GB and 2 GPU 4x 192GB and 2 32GB V100 16x 96GB and 16GB MCDRAM (KNL) | 16x 192GB 4x 192GB and 2 16GB V100 4x 1.5TB | 800x 384GB 100x 384GB and 2 A100 9x 384GB and 2x RTX 6000 8x 384GB and 4x T4 15x 384GB and 2x A40 8x 3TB | 180x 256GB See below for Liqid fabric devices. *** | 560x 256GB 18x 256GB and 2 A100 | 110x 512GB 17x 512GB 1x 768GB and 8 VE 1x 768GB with 16 classic IPUs 1x 512GB with 16 Bow2000 IPUs See below for Liqid fabric devices. **** | 35x 384 GB 10x 768 GB and 2 A30 |
| Local Disk/Node | 1x 1TB disk | 1.6TB or 3.2TB NVMe disks | 1x 1.6TB NVMe disk | 1x 3.84TB NVMe disk | 1x 144GB SDD disk | 1x 1.6TB NVMe disk | 1x 200GB disk |
| Other | |||||||
| Interconnect | Omni-Path Architecture | 40Gb Ethernet | HDR100 InfiniBand | HDR100 InfiniBand | HDR100 InfiniBand | NDR200 InfiniBand | HDR100 InfiniBand |
| Global Storage | 7.4 PB | 2 PB | 5.0 PB (Shared between Grace and FASTER) | 8 PB | 2.6 PB | 2.2 PB | |
* Lonestar6 is hosted at TACC
** Will be updated when remaining hardware has been released to market.
*** FASTER cluster accelerators on Liqid fabrics: 200x T4, 40x A100, 8x A10, 4x A30, 8x A40 GPUs
**** ACES cluster accelerators on Liqid fabrics: 30x H100 GPUs, 120x PVC GPUs, 4x A30 GPUs, 2 D5005 FPGAs, 2 IA-840F FPGAs, and 2 Nextsilicon accelerators