H100 GPU Rental
Rent H100 GPUs on Lium from $2.00 per GPU-hour, built for large-scale AI training, inference, and HPC. Billed per second with no minimum, deployed in seconds from the browser or the CLI.
Price per GPU-hour
$2.00 – $3.75
Live asks across all available pods, billed per second
Available now
4 pods
32 GPUs in total
Reference price
$1.30
Marketplace baseline per GPU-hour
About the H100
The H100 is NVIDIA’s Hopper-generation data center GPU and still the default choice for serious training and high-throughput inference. Its transformer engine adds native FP8, which roughly doubles arithmetic throughput over FP16 for models that tolerate the lower precision, and 80 GB of high-bandwidth memory keeps weights, activations and KV cache on the card instead of spilling to host RAM. Teams reach for it to fine-tune language models, to serve production inference where latency is part of the product, and to run diffusion, speech and scientific workloads that are limited by memory bandwidth rather than by raw arithmetic.
What people run on it
- Fine-tuning 7B–70B language models
- Production LLM inference at low latency
- Diffusion, speech and vision training
- Bandwidth-bound HPC and simulation
H100 variants compared
| H100 80GB HBM3 | H100 PCIe | |
|---|---|---|
| VRAM | 80 GB HBM3 | 80 GB HBM2e |
| Memory bandwidth | 3,350 GB/s | 2,000 GB/s |
| FP16 tensor | 989 TFLOPS | 756 TFLOPS |
| FP8 tensor | 1,979 TFLOPS | 1,513 TFLOPS |
| Interconnect | NVLink 900 GB/s | NVLink bridge 600 GB/s |
| Power | 700 W | 350 W |
| Architecture | Hopper | Hopper |
| Released | 2022 | 2022 |
| Listed on the marketplace as | NVIDIA H100 80GB HBM3 | NVIDIA H100 PCIe |
| Price per GPU-hour | $2.00 – $3.75 | reference $1.30 |
| Available now | 4 pods, 32 GPUs | None right now |
| Rent it | Browse H100 80GB HBM3 pods | Request this GPU |
Tensor figures are NVIDIA’s published dense numbers for the card, without sparsity; sparsity doubles them. What a pod actually delivers depends on the model, the kernels and the host it runs on.
Both carry 80 GB, but they are not the same part. The SXM card uses HBM3 at 3.35 TB/s and sits on an NVLink mesh at 900 GB/s, so tensors move between GPUs in one node without crossing PCIe. The PCIe card uses HBM2e at 2 TB/s, draws 350 W instead of 700 W, and links cards in pairs over a 600 GB/s bridge. For single-GPU inference the gap is modest and the PCIe card is usually the cheaper of the two per GPU-hour. For multi-GPU training, tensor parallelism, or anything bandwidth-bound, take SXM.
What runs on one H100
With 80 GB of memory, one card holds roughly the following open models for inference, counting the weights plus about 20% for activations and cache. Rent several GPUs in one pod to go bigger.
| Precision | Largest model that fits |
|---|---|
| FP16 | 13B parameters |
| 8-bit | 34B parameters |
| 4-bit | 70B parameters |
Running one H100 around the clock for a month (730 hours) costs about $1,460 at the lowest live ask; you pay only for the seconds a pod runs.
Frequently asked questions
How much does it cost to rent an H100 on Lium?
H100 pods are listed at $2.00 to $3.75 per GPU-hour right now, across 4 pods with 32 GPUs. Each provider sets its own price and the range moves as pods are added and removed.
Where is the cheapest place to rent an H100 by the hour?
On Lium, H100 pods start at $2.00 per GPU-hour right now, the lowest live ask across 4 pods, and billing is per second with no minimum, so a 20-minute job costs a third of that. Providers set their own prices, so the lowest ask moves as pods are listed and rented.
Is there a minimum rental period?
No. Lium bills per second with no minimum and no reservation: start the pod when you need it, delete it when the job is done, and pay for the seconds it ran.
How do I rent an H100 pod?
Top up a balance by card or crypto, pick an H100 pod on the marketplace, choose a Docker template, add your SSH key and deploy. From a script, run lium up with the node id after exporting LIUM_API_KEY.
Where do the machines run, and what happens to my data?
Providers run their own hardware in their own facilities, and every pod on the marketplace names its city and country, so you can pick a region before you start. Your data lives on the machine you rented and is released when you delete the pod; attach a volume if you need storage that outlives a single pod.
Compare with
| GPU | VRAM | Price per GPU-hour |
|---|---|---|
| H100 NVL | 94 GB | reference $1.11/hr |
| H800 80GB HBM3 | 80 GB | reference $0.88/hr |
For AI agents — CLI install, agent skill, headless signup: llms-full.txt