Rent NVIDIA H100 SXM GPU by the hour
The NVIDIA H100 SXM is a Hopper-architecture datacenter GPU equipped with 80 GB of VRAM and a 700 W power envelope. It delivers about 53.5 TFLOPS of standard floating-point compute and a memory bandwidth of about 2869 GB/s, supporting CUDA compute capability 9.0. Rent it on ONIE with per-hour billing, full SSH access, and ready templates for PyTorch/CUDA, vLLM, Ollama, and ComfyUI.
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 SXM is good for
Large Language Model Inference
Deploy demanding language models into production using the 80 GB VRAM capacity and about 2869 GB/s of memory bandwidth via vLLM or Ollama templates.
Deep Learning Fine-Tuning
Train and fine-tune neural networks using the Hopper architecture and CUDA compute capability 9.0 with pre-configured PyTorch/CUDA environments.
High-Throughput Batch Processing
Run intensive data processing and simulation workloads that benefit from about 53.5 TFLOPS of compute performance in a 700 W datacenter form factor.
Heavy Generative Media Workflows
Execute complex ComfyUI pipelines and diffusion models with large batch sizes that fit comfortably within the 80 GB memory pool.
Is it the right GPU for you?
Choose the H100 SXM when your training or inference workloads require 80 GB of high-bandwidth datacenter memory and Hopper architecture features. A smaller workstation or consumer GPU is a more practical choice for lightweight development or small models, while a newer or higher-capacity datacenter GPU is preferable if a single instance cannot hold your entire model in 80 GB of VRAM.
Compare with similar GPUs
Frequently asked questions
Is the H100 SXM suitable for LLM fine-tuning and inference?
Yes, its 80 GB of VRAM and memory bandwidth of about 2869 GB/s make it well-suited for both serving and fine-tuning large language models. You can start immediately on ONIE using ready templates for vLLM, Ollama, or PyTorch/CUDA.
How does the SXM variant compare to lower-power datacenter or workstation GPUs?
The H100 SXM operates at a 700 W thermal design power, allowing it to sustain higher compute performance and memory bandwidth than PCIe datacenter variants or workstation cards. If your workload does not saturate about 53.5 TFLOPS of compute or 80 GB of memory, a smaller GPU may be sufficient.
What software compatibility does CUDA compute capability 9.0 provide?
Compute capability 9.0 indicates hardware support for NVIDIA Hopper architecture instructions and features. Our pre-configured PyTorch/CUDA templates come ready to utilize this hardware without manual driver setup.
How do billing and instance access work on ONIE?
Instances are billed on a per-hour basis for the duration of your rental. Once deployed, you can connect directly to your environment via SSH access or interact with pre-installed tools like ComfyUI.
Rent the H100 SXM by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent H100 SXM