Rent NVIDIA H100 PCIE GPU by the hour
The NVIDIA H100 PCIE is a Hopper-architecture datacenter GPU built for demanding deep learning and scientific computing workloads. It provides 80 GB of VRAM, about 1760 GB/s of memory bandwidth, and about 41.0 TFLOPS of standard floating-point compute within a 350 W power envelope. On ONIE, you can deploy this compute capability 9.0 accelerator with per-hour billing and full SSH access.
Specifications
VRAM per GPU
Architecture
Compute capability
FP32 compute
Memory bandwidth
Power limit
Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.
Available configurations
Live stock and prices, updated continuously. The final price is confirmed before you order.
Loading live prices…
What the H100 PCIE is good for
Large Language Model Inference
Serve large language models in memory using the 80 GB of VRAM and high memory bandwidth alongside pre-configured vLLM or Ollama templates.
Model Fine-Tuning and Training
Train or fine-tune deep neural networks using the Hopper architecture and ready PyTorch/CUDA environments.
Batch Image Generation
Run high-throughput diffusion pipelines and complex ComfyUI workflows that benefit from substantial memory capacity.
Data Science and Simulation
Execute compute-intensive simulations and data processing tasks with about 41.0 TFLOPS of standard floating-point performance and about 1760 GB/s of memory throughput.
Is it the right GPU for you?
Choose the H100 PCIE when your workloads require 80 GB of VRAM, datacenter-grade memory bandwidth, and Hopper compute capability 9.0 features. For lighter inference or smaller models, a workstation or consumer GPU with less VRAM is more economical, while extreme throughput requirements may warrant an SXM-based datacenter variant.
Compare with similar GPUs
Frequently asked questions
Is the H100 PCIE suitable for LLM fine-tuning and inference?
Yes, its 80 GB of VRAM and about 1760 GB/s of memory bandwidth make it well-suited for hosting and training large models. You can start immediately using pre-configured templates for PyTorch/CUDA, vLLM, or Ollama.
How does the H100 PCIE compare to SXM or smaller workstation GPUs?
The H100 PCIE operates at a 350 W TDP, offering lower power consumption and standard PCIe compatibility compared to higher-wattage SXM datacenter variants. Compared to workstation or consumer cards, it provides significantly more VRAM at 80 GB and higher memory bandwidth.
What software support and compute capability does this GPU have?
Being based on the NVIDIA Hopper architecture, the H100 PCIE supports CUDA compute capability 9.0. You can launch instances with ready templates including PyTorch/CUDA and ComfyUI, or configure your own stack via SSH access.
How does billing work for H100 PCIE instances on ONIE?
Instances on ONIE use straightforward per-hour billing, so you only pay for the time your GPU is active. Full SSH access is included with every deployment.
Rent the H100 PCIE by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent H100 PCIE