RunLocalAI
nvidia
GPU
192GB VRAM
workstation

NVIDIA B200

Datacenter Blackwell. 192GB HBM3e per chip, ~8 TB/s bandwidth. Cloud-tier — you rent these by the hour.

Released 2024

Overview

Datacenter Blackwell. 192GB HBM3e per chip, ~8 TB/s bandwidth. Cloud-tier — you rent these by the hour.

Specs

VRAM192 GB
Power draw1000 W
Released2024
MSRP$40000
Backends
CUDA

Models that fit

Open-weight models small enough to run on NVIDIA B200 with usable context.

Llama 3.1 8B Instruct
8B · llama
Llama 4 Scout
109B · llama
Qwen 3 30B-A3B
30B · qwen
Qwen 2.5 Coder 32B Instruct
32B · qwen
Llama 3.3 70B Instruct
70B · llama
Qwen 3 32B
32B · qwen
Gemma 4 31B Dense
31B · gemma
Qwen 3 8B
8B · qwen

Frequently asked

What models can NVIDIA B200 run?

With 192GB VRAM, the NVIDIA B200 runs 70B models in 4-bit quantization, plus everything smaller. See the model list below for tested combinations.

Does NVIDIA B200 support CUDA?

Yes — NVIDIA B200 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.