Compute Infrastructure
The lab runs a fleet of on-prem GPU workstations for day-to-day development, and bursts to the cloud for anything that needs a bigger accelerator.
On-Prem GPU Fleet
29 GPUs available on-prem, as of 2026-08-03.
| GPU Model | Count |
|---|---|
| NVIDIA RTX A6000 | 12 |
| NVIDIA RTX A5000 | 4 |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | 3 |
| NVIDIA RTX PRO 5000 Blackwell | 3 |
| NVIDIA RTX 5000 Ada Generation | 3 |
| NVIDIA RTX 4000 Ada Generation | 2 |
| Quadro RTX 8000 | 2 |
| Total | 29 |
Cloud Compute
For large-scale training runs that outgrow the on-prem workstations — multi-GPU foundation model training, large batch experiments — the lab uses cloud GPU services for on-demand access to NVIDIA A100, H100, and H200 accelerators.