NVIDIA capacity, billed by the hour or reserved by the month
Dedicated KVM instances with root access, persistent volumes and prebuilt CUDA images. Boot in minutes; cancel an hourly instance any time.
| Class | GPU | VRAM | vCPU / RAM | Storage | Network | Hourly | Monthly |
|---|---|---|---|---|---|---|---|
| RTX 4090 Best value | NVIDIA RTX 4090 | 24 GB | 16 / 64 GB | 500 GB NVMe | 2 Gbps | $0.55 | $349.00 |
| RTX A6000 | NVIDIA RTX A6000 | 48 GB | 24 / 96 GB | 1000 GB NVMe | 4 Gbps | $0.89 | $569.00 |
| L40S Inference | NVIDIA L40S | 48 GB | 32 / 128 GB | 1000 GB NVMe | 4 Gbps | $1.29 | $799.00 |
| A100 80GB | NVIDIA A100 SXM | 80 GB | 48 / 256 GB | 2000 GB NVMe | 8 Gbps | $1.79 | $1,149.00 |
| H100 80GB Fastest | NVIDIA H100 SXM | 80 GB | 64 / 320 GB | 2000 GB NVMe | 10 Gbps | $2.99 | $1,899.00 |
| H200 141GB Frontier | NVIDIA H200 SXM | 141 GB | 96 / 512 GB | 4000 GB NVMe | 10 Gbps | $3.99 | $2,499.00 |
Monthly reservations assume 720 hours (30 days) — reserve and the effective rate drops to the monthly column.
Best price/performance for 7B–14B inference and fine-tuning.
- 1× RTX 4090 24GB GDDR6X
- 16 vCPU · 64GB RAM · 500GB NVMe
- 2 Gbps network · static IP
48GB cards for larger context inference and multi-model serving.
- 1× RTX A6000 48GB GDDR6
- 24 vCPU · 96GB RAM · 1TB NVMe
- 4 Gbps network · static IP
Purpose-built for generative AI inference at scale.
- 1× L40S 48GB GDDR6 ECC
- 32 vCPU · 128GB RAM · 1TB NVMe
- 4 Gbps network · static IP
The workhorse for training and large-model serving.
- 1× A100 80GB HBM2e
- 48 vCPU · 256GB RAM · 2TB NVMe
- 8 Gbps network · NVLink ready
Fastest transformer engine for frontier training runs.
- 1× H100 80GB HBM3
- 64 vCPU · 320GB RAM · 2TB NVMe
- 10 Gbps network · NVLink ready
Frontier capacity for the largest models and context windows.
- 1× H200 141GB HBM3e
- 96 vCPU · 512GB RAM · 4TB NVMe
- 10 Gbps network · NVLink ready
Bring your own stack
Root SSH, Docker, CUDA 12.4 and PyTorch 2.4 images, or upload your own. Jupyter available on every instance.
Clusters on request
Multi-node NVLink and InfiniBand fabrics for training runs — talk to sales for sizing and reserved pricing.
Pay how you already pay
Same invoice flow as tokens: wire or USDT, USD-denominated, activated by a billing specialist.