GPU quota by request
Dedicated RTX Pro Blackwell cards, billed by the hour.
Three cards today: RTX 4500 Pro (32 GiB), RTX 5000 Pro (48 GiB), and RTX 6000 Pro (96 GiB). All dedicated, all backed by NVMe. GPU access is quota-gated — write to us at support@excloud.dev with the workload context and we'll bump you up.
$ exc compute create --name train-1 --instance_type nv2a.xlarge --wait
The catalogue
Three cards, three price points.
All rates are on-demand, metered hourly, in INR. Pick the card by VRAM budget — the bigger the model, the bigger the card.
| Instance | Card | vCPU | Memory | VRAM | Rate |
|---|---|---|---|---|---|
| nv2a.xlarge | RTX 4500 Pro Blackwell | 4 | 16 GiB | 32 GiB | ₹44.554 /hr |
| nv3a.2xlarge | RTX 5000 Pro Blackwell | 8 | 32 GiB | 48 GiB | ₹63.849 /hr |
| nv1a.4xlarge | RTX 6000 Pro Blackwell | 16 | 64 GiB | 96 GiB | ₹126.784 /hr |
How access works
Request a quota, get a yes or a no.
GPUs are quota-gated because the supply is finite. Email support@excloud.dev with the workload, the model size, and the duration you need. We reply quickly, with a real decision.
- Quota is per-org. Request it once, get it once.
- Same region. GPU instances run in mum-1a, like every other Excloud resource.
- LLM inference already exists. If you just want to run Qwen3.6-27B without a dedicated GPU, see LLM Inference.
$ exc compute instancetype list NAME vCPU VRAM RATE nv2a.xlarge 4 32 GiB ₹44.554/hr nv3a.2xlarge 8 48 GiB ₹63.849/hr nv1a.4xlarge 16 96 GiB ₹126.784/hr $ exc compute create \ --name train-1 \ --instance_type nv2a.xlarge \ --image_id 4 --wait ✓ instance train-1 running train-1 · 4 vCPU · 16 GiB · RTX 4500 Pro
The honest part
A note on what GPU pages do not claim.
The 50 Gbps networking and AMD EPYC numbers on the Compute page are documented for t1a and m1a only. We don't quote them for GPU instances — the GPU spec sheets specify what they specify, and the rest you measure on your workload.