Choose memory first
Model size, batch size, context length, precision, and framework determine the minimum VRAM boundary.
Loading AuraOne ComputePreparing GPU search and provider inventory.
Compute / GPU catalog
Use the reference catalog to narrow the hardware class, then search the live API for a date, quantity, workload, region, and hourly budget.
The hardware catalog does not claim provider inventory, regional availability, launch time, or price. Those values appear only in current API-backed offers.
| Accelerator | Architecture | VRAM | Reference workloads | Inspect |
|---|---|---|---|---|
| NVIDIA H100 80GB | Hopper | 80 GB | training, fine-tuning, inference | GPU details |
| NVIDIA A100 80GB | Ampere | 80 GB | training, fine-tuning, inference | GPU details |
| NVIDIA L40S 48GB | Ada Lovelace | 48 GB | fine-tuning, inference, rendering | GPU details |
| NVIDIA RTX A6000 48GB | Ampere | 48 GB | fine-tuning, inference, rendering | GPU details |
| NVIDIA RTX 4090 24GB | Ada Lovelace | 24 GB | fine-tuning, inference, rendering | GPU details |