On-demand GPUs, ready in seconds
Train, fine-tune and serve models on 1x to 8x NVIDIA GPU instances. Billed by the minute, no commitments, 10 TB of free egress a month, and a software stack that works the moment you log in.
From $0.49 per GPU-hour. B200 available in every region.
Everything you need, nothing you have to set up
Launch in under a minute
Instances boot GreenAI Stack with drivers, CUDA, PyTorch, JAX and Hugging Face libraries pre-installed and tested together. SSH in and start training.
1x, 2x, 4x and 8x configurations
SXM instances are a full NVLink domain on an NVIDIA HGX baseboard, and PCIe instances such as L40S and RTX 6000 Ada are clearly marked. No shared GPUs, no surprises.
Dashboard, CLI and API
Launch, scale and terminate from the web console, our CLI, the REST API or the Terraform provider. Automate everything.
Per-minute pricing, no commitments
Prices per GPU per hour, billed per minute. Persistent storage and egress priced separately.
| GPU | GPUs | VRAM / GPU | vCPUs | RAM | Storage | Price / GPU / hr | Action |
|---|---|---|---|---|---|---|---|
| NVIDIA B200 SXM6 New | 8x | 180 GB | 208 | 2,900 GiB | 22 TiB NVMe | $6.42 | Launch |
| NVIDIA H200 SXM Popular | 8x | 141 GB | 208 | 2,000 GiB | 22 TiB NVMe | $4.55 | Launch |
| NVIDIA H100 SXM | 8x | 80 GB | 208 | 1,800 GiB | 22 TiB NVMe | $3.06 | Launch |
| NVIDIA A100 SXM | 8x | 80 GB | 240 | 1,800 GiB | 19 TiB NVMe | $1.09 | Launch |
| NVIDIA A100 SXM | 8x | 40 GB | 124 | 1,800 GiB | 5.8 TiB NVMe | $0.85 | Launch |
| GPU | GPUs | VRAM / GPU | vCPUs | RAM | Storage | Price / GPU / hr | Action |
|---|---|---|---|---|---|---|---|
| NVIDIA B200 SXM6 | 4x | 180 GB | 104 | 1,440 GiB | 11 TiB NVMe | $6.52 | Launch |
| NVIDIA H100 SXM | 4x | 80 GB | 104 | 900 GiB | 11 TiB NVMe | $3.16 | Launch |
| NVIDIA A100 PCIe | 4x | 40 GB | 120 | 900 GiB | 1 TiB NVMe | $0.85 | Launch |
| NVIDIA L40S | 4x | 48 GB | 96 | 768 GiB | 4 TiB NVMe | $0.99 | Launch |
| NVIDIA RTX 6000 Ada | 4x | 48 GB | 56 | 400 GiB | 1 TiB NVMe | $0.89 | Launch |
| GPU | GPUs | VRAM / GPU | vCPUs | RAM | Storage | Price / GPU / hr | Action |
|---|---|---|---|---|---|---|---|
| NVIDIA B200 SXM6 | 2x | 180 GB | 52 | 720 GiB | 5.5 TiB NVMe | $6.62 | Launch |
| NVIDIA H100 SXM | 2x | 80 GB | 52 | 450 GiB | 5.5 TiB NVMe | $3.26 | Launch |
| NVIDIA A100 PCIe | 2x | 40 GB | 60 | 450 GiB | 1 TiB NVMe | $0.85 | Launch |
| NVIDIA L40S | 2x | 48 GB | 48 | 384 GiB | 2 TiB NVMe | $0.99 | Launch |
| NVIDIA RTX 6000 Ada | 2x | 48 GB | 28 | 200 GiB | 1 TiB NVMe | $0.89 | Launch |
| GPU | GPUs | VRAM / GPU | vCPUs | RAM | Storage | Price / GPU / hr | Action |
|---|---|---|---|---|---|---|---|
| NVIDIA B200 SXM6 | 1x | 180 GB | 26 | 360 GiB | 2.75 TiB NVMe | $6.72 | Launch |
| NVIDIA GH200 | 1x | 96 GB | 64 | 432 GiB | 4 TiB NVMe | $2.89 | Launch |
| NVIDIA H100 SXM | 1x | 80 GB | 26 | 225 GiB | 2.75 TiB NVMe | $3.36 | Launch |
| NVIDIA H100 PCIe | 1x | 80 GB | 26 | 225 GiB | 1 TiB NVMe | $2.49 | Launch |
| NVIDIA A100 PCIe | 1x | 40 GB | 30 | 225 GiB | 512 GiB NVMe | $0.85 | Launch |
| NVIDIA L40S | 1x | 48 GB | 24 | 192 GiB | 1 TiB NVMe | $0.99 | Launch |
| NVIDIA RTX 6000 Ada | 1x | 48 GB | 14 | 100 GiB | 512 GiB NVMe | $0.89 | Launch |
| NVIDIA A10 | 1x | 24 GB | 30 | 226 GiB | 1.3 TiB NVMe | $0.49 | Launch |
| NVIDIA RTX A6000 | 1x | 48 GB | 14 | 100 GiB | 512 GiB NVMe | $0.59 | Launch |
Need reserved capacity or volume discounts? Talk to our team. Startups and researchers may qualify for credits.
Dedicated hardware, live telemetry and storage that grows with you
- Bare-metal performance. Dedicated GPUs with full NVLink bandwidth, local NVMe at 7 GB/s per drive and 200 Gb/s of network per node.
- Live telemetry. GPU utilization, memory, temperature, power draw and NVLink throughput per instance, streamed to the dashboard and exportable to Prometheus.
- Persistent filesystems. NVMe-backed shared storage that mounts on any instance in the region. Scale from 100 GB to 100 TB without downtime.
- S3-compatible object storage. Keep datasets and checkpoints next to your GPUs with zero transfer fees.
- Private networking. Every account gets an isolated VPC with optional peering to your cloud or on-prem environment.
NVIDIA HGX B200, now on demand
Blackwell brings 180 GB of HBM3e per GPU, fifth-generation NVLink at 1.8 TB/s and FP4 Transformer Engine support. In our tests a 70B model trains about 2.6x faster per GPU than on H100.
- GPU memory180 GB HBM3e per GPU
- Memory bandwidth8 TB/s per GPU
- NVLink1.8 TB/s per GPU
- FP4 (dense)Up to 72 PFLOPS per node
- Configurations1x, 2x, 4x, 8x
- On-demand from$6.42 / GPU / hr
Scale to a cluster without changing a line of code
GPU Clusters give you 16 to 1,024+ GPUs on NVIDIA Quantum-2 InfiniBand with managed Slurm or Kubernetes. Same images, same storage, same dashboard.
How is billing calculated?
Instances are billed per minute from the moment they are ready until you terminate them. There is no minimum charge and no charge while an instance is being provisioned. Persistent storage is billed per GB-month.
Can I stop an instance and keep my data?
Yes. Attach a persistent filesystem to your instance and your data survives termination. Instance-local NVMe is ephemeral and wiped when the instance is terminated.
Do you charge for egress?
The first 10 TB per month of internet egress is free. Traffic between GREENAIBOX regions and to our object storage is always free.
What comes pre-installed?
Every instance boots GreenAI Stack: Ubuntu 24.04 LTS, NVIDIA drivers, CUDA, cuDNN, NCCL, PyTorch, JAX, TensorFlow, Hugging Face libraries, vLLM and Docker with the NVIDIA container toolkit. You can also bring your own image.
Are the GPUs shared?
No. Every instance is bare metal or a dedicated VM with GPU passthrough. You get the whole GPU, the whole NVLink domain and predictable performance.
What if I need more than 8 GPUs?
Move to GPU Clusters for 16 to 1,024+ GPUs on InfiniBand, or ask us to build you a cluster. The software stack is identical, so your code moves without changes.
Build on GREENAIBOX
Launch an instance in minutes, reserve a cluster in days, or talk to us about building your own.