H100 SXM
In stock$2.49/hr
The workhorse for multi-node LLM training. NVLink across all 8 GPUs.
- VRAM80 GB
- FP161,979 TFLOPS
- vCPU / RAM26 / 200GB
Rent NVIDIA H100, H200, A100 and RTX-class accelerators by the hour. No quotas, no sales calls, no minimum term — spin a box up, run your job, tear it down.
The workhorse for multi-node LLM training. NVLink across all 8 GPUs.
Proven, widely supported, and the best price-per-VRAM on the fleet.
Ada-generation inference and rendering with hardware video encode.
Best raw throughput per dollar on the fleet for single-GPU jobs.
Top up from $25. Credit never expires and covers every GPU class.
Pick a GPU, region and image. Pre-baked CUDA and PyTorch stacks included.
Live provisioning logs stream into the console while the node boots.
Billing stops the second you destroy the instance. Nothing rounds up.
Multi-node H100 clusters with InfiniBand for pre-training, or a single A100 for LoRA runs.
L40S and H100 PCIe nodes tuned for throughput, with autoscaling recipes in the docs.
RTX 4090 boxes give the best images-per-dollar on the fleet for SD and Flux workloads.
Hardware encode on Ada cards for Blender, Octane, Unreal and Omniverse pipelines.
Short bursts without a procurement cycle. Educational discounts on request.
FP64-capable A100 and H100 nodes for simulation, genomics and computational chemistry.