RunLocalAI
nvidia
GPU
80GB VRAM
workstation

NVIDIA H100 PCIe

PCIe Hopper. Lower power, lower bandwidth than SXM. Server-tier.

Released 2022

Overview

PCIe Hopper. Lower power, lower bandwidth than SXM. Server-tier.

Specs

VRAM80 GB
Power draw350 W
Released2022
MSRP$25000
Backends
CUDA

Models that fit

Open-weight models small enough to run on NVIDIA H100 PCIe with usable context.

Llama 3.1 8B Instruct
8B · llama
Qwen 3 30B-A3B
30B · qwen
Qwen 2.5 Coder 32B Instruct
32B · qwen
Qwen 3 32B
32B · qwen
Gemma 4 31B Dense
31B · gemma
Qwen 3 8B
8B · qwen
DeepSeek R1 Distill Qwen 32B
32B · deepseek
Gemma 4 26B MoE
26B · gemma

Frequently asked

What models can NVIDIA H100 PCIe run?

With 80GB VRAM, the NVIDIA H100 PCIe runs 70B models in 4-bit quantization, plus everything smaller. See the model list below for tested combinations.

Does NVIDIA H100 PCIe support CUDA?

Yes — NVIDIA H100 PCIe is an NVIDIA card with full CUDA support, the most mature local-AI backend. llama.cpp, Ollama, vLLM, and ExLlamaV2 all run natively.

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.