Rent NVIDIA H200 NVL GPU by the hour
The NVIDIA H200 NVL is a Hopper-architecture datacenter GPU built for memory-intensive workloads, offering 140 GB of VRAM and a memory bandwidth of about 3884 GB/s. It operates at a 600 W power envelope, supports compute capability 9.0, and delivers about 48.3 TFLOPS of standard floating-point compute. You can deploy this processor on ONIE with per-hour billing, direct SSH access, and ready templates.
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 H200 NVL is good for
Large Language Model Inference
Run large models in memory without sharding, taking advantage of the 140 GB VRAM capacity and about 3884 GB/s of memory bandwidth.
Model Fine-Tuning and Training
Train and adapt deep learning models using Hopper compute capability 9.0 and spacious GPU memory for large batch sizes.
High-Throughput Batch Processing
Process massive datasets and long context windows efficiently using pre-configured PyTorch/CUDA, vLLM, or Ollama templates.
Scientific Computing and Simulation
Execute complex datacenter simulations that require about 48.3 TFLOPS of standard floating-point performance and substantial onboard memory.
Is it the right GPU for you?
Choose the H200 NVL when your primary bottleneck is VRAM capacity and memory bandwidth, particularly for hosting large models on a single card. If you are running lighter inference, standard image generation, or development tasks that do not require 140 GB of memory, a smaller workstation or consumer GPU is a more practical choice.
Compare with similar GPUs
Frequently asked questions
Why choose the H200 NVL over a standard Hopper GPU?
The H200 NVL provides 140 GB of VRAM and about 3884 GB/s of memory bandwidth, which is higher than the memory capacity and bandwidth of earlier Hopper-generation cards. This allows larger models and context windows to fit on a single GPU.
Is the H200 NVL suitable for LLM serving with vLLM or Ollama?
Yes, the 140 GB memory pool and high memory bandwidth make it well-suited for serving large language models. ONIE provides ready templates for both vLLM and Ollama to help you start quickly.
What software environments and access methods are available?
Every instance includes full SSH access for custom configuration. You can also launch instances with pre-installed templates including PyTorch/CUDA, vLLM, Ollama, and ComfyUI.
How does billing work for the H200 NVL on ONIE?
Instances are billed on a per-hour basis. You only pay for the time your H200 NVL instance is active.
Rent the H200 NVL by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent H200 NVL