Model training
Spin up single GPUs or multi-GPU instances for training runs, then release them the moment the job finishes — you only pay while the work runs.
GPU RENTALS
Accelerated computing, on demand.
Reserve NVIDIA H100, A100, and L40S accelerators by the hour or month for training, fine-tuning, and inference. Provisioned on our global network with NVMe scratch storage, fast networking, and full root access — no long-term commitment.
NVIDIA acceleratorsNVMe scratch10Gbps networkingHourly billing
- From
- $0.99/hr
- Network
- Global
- GPUs
- H100 · A100
- Support
- 24/7
USE CASES
Built for the AI lifecycle.
From the first experiment to production inference, rent exactly the acceleration each stage needs and release it when you're done.
Fine-tuning
Adapt foundation models to your data on dedicated accelerators with fast NVMe scratch space and high-bandwidth interconnect between GPUs.
Low-latency inference
Serve models close to your users with predictable performance, autoscaling-ready instances, and per-second-grade billing for bursty traffic.
Research & experimentation
Give teams self-service access to top-tier GPUs without capital expense or procurement delays. Reserve what you need, when you need it.
GPU LINEUP
Pick your accelerator.
Single and multi-GPU configurations, billed per GPU-hour. Monthly reservations available.
Flagship training
NVIDIA H100 80GB
Large-model training and high-throughput fine-tuning.
$2.49/GPU-hr
- GPU memory
- 80GB HBM3
- vCPU
- 26 vCPU
- System RAM
- 200GB RAM
- Local storage
- 1TB NVMe
Proven workhorse
NVIDIA A100 80GB
Distributed training, fine-tuning, and batch inference.
$1.49/GPU-hr
- GPU memory
- 80GB HBM2e
- vCPU
- 22 vCPU
- System RAM
- 160GB RAM
- Local storage
- 1TB NVMe
Inference & graphics
NVIDIA L40S 48GB
Real-time inference, rendering, and visual workloads.
$0.99/GPU-hr
- GPU memory
- 48GB GDDR6
- vCPU
- 16 vCPU
- System RAM
- 96GB RAM
- Local storage
- 512GB NVMe
Need a reserved multi-node cluster or a GPU not listed? Contact sales.
WHAT'S INCLUDED
Everything around the GPU.
Accelerators are only useful when the rest of the machine keeps up. Every GPU instance ships with the storage, networking, and access your workloads need.
Hourly or monthly
Pay by the hour for bursts or commit monthly for sustained workloads at a lower effective rate. No long-term contracts.
NVMe scratch storage
Local NVMe keeps datasets and checkpoints close to the GPU so data loading never starves your accelerators.
High-throughput networking
Up to 10Gbps uplinks and private networking between instances for distributed training and fast dataset transfer.
Full root access
Bring your own CUDA, drivers, and frameworks. Install whatever your stack needs with complete control of the machine.
Snapshots & images
Capture a configured environment as an image and relaunch it in seconds, so every run starts from a known-good state.
24/7 expert support
Our team is on call around the clock to size an instance, plan a cluster, or work through an issue with you.
FAQ
GPU rentals, answered.
Still deciding on a configuration? Our team can size an instance or plan a cluster with you.
Contact salesGPUs are billed per GPU-hour while an instance is running, with monthly reservations available at a lower effective rate for sustained workloads. There are no long-term contracts — release an instance and billing stops.
We offer NVIDIA H100, A100, and L40S accelerators in single and multi-GPU configurations. Talk to our team if you need a specific topology or a reserved multi-node cluster.
Yes. Choose multi-GPU instances for single-node training, or connect instances over private high-throughput networking for distributed, multi-node runs.
Every GPU instance ships with full root access. Bring your own CUDA, drivers, and frameworks, or start from one of our prebuilt images and customize from there.
GPU instances run in ezghcloud regions with NVMe scratch storage and private networking available between instances, tuned for low latency to your users.
Get started with GPU rentals
Put a GPU to work in minutes.
Launch an instance with hourly billing, or talk to us about a monthly reservation for sustained workloads.