Question 1

What AI models can run on H100 PCIe?

Accepted Answer

The H100 PCIe (80 GB VRAM) can run models including Llama-2-7b-chat, Llama-3.2-1B-Instruct-QLORA_INT4_EO8, MediPhi-Instruct, Llama-3.2-3B-Instruct-SpinQuant_INT4_EO8, Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8 and more — 10 compatible models indexed so far. Compatible models are determined by whether the model's memory footprint fits within the GPU's 80 GB VRAM, potentially using quantization (Q8, Q6, or Q4) to reduce memory usage.

Question 2

What is the cheapest cloud option for H100 PCIe?

Accepted Answer

The cheapest on-demand option for the H100 PCIe is Vast.ai at $1.80/hr (1x-H100_SXM-32-65536). Spot/preemptible pricing is available at $0.00/hr. We track pricing across 76 providers to help you find the best deal.

Question 3

Is H100 PCIe good for fine-tuning?

Accepted Answer

Yes, the H100 PCIe is excellent for fine-tuning. With 80 GB of VRAM, it can handle full fine-tuning of models up to ~30B parameters and LoRA/QLoRA fine-tuning of much larger models. It's a top-tier choice for training workloads.

Question 4

How much VRAM does H100 PCIe have?

Accepted Answer

The H100 PCIe has 80 GB of VRAM (HBM2e). It also offers 2000 GB/s memory bandwidth, 756.5 TFLOPS of FP16 compute, FP8 support for efficient inference.

Question 5

What quantization formats work on H100 PCIe?

Accepted Answer

The H100 PCIe supports the following quantization formats: FP16 (full precision), FP8, Q8 (8-bit), Q6 (6-bit), Q4 (4-bit). FP16 works for models up to ~40B parameters. FP8 halves memory vs FP16 with minimal quality loss. Q4 quantization lets models up to ~272B parameters (by VRAM estimate) fit on the H100 PCIe.

Provider	On-demand	Spot	Actions
Vast.ai	$1.800/hr	$0.000/hr	Rent
RunPod	$2.890/hr	$2.390/hr	Rent Get $5–$500 free
Nebius	$2.950/hr	$1.250/hr	—
Lambda	$3.290/hr	—	—
Google Cloud	$4.201/hr	$0.980/hr	—
Amazon Web Services	$6.880/hr	$2.806/hr	—
Microsoft Azure	$6.980/hr	$6.980/hr	—

H100 PCIe

Specifications

Cloud pricing

Compatible models

Compare this GPU

Frequently asked questions

VRAM	80 GB
Memory type	HBM2e
Bandwidth	2000 GB/s
FP16 TFLOPS	756.5
FP8 support	Yes