Hardware / NVIDIA /B200 (SXM 192GB)

NVIDIA · GPU

B200 (SXM 192GB)

Runs 157 models fully in memory, 15 more with CPU offload. 2 of those sit close enough to 192 GB that our size estimate is what decides them.

Memory

192 GB

Bandwidth

8000 GB/s

Runs

157

01

Specifications

Memory

192 GB

Bandwidth

8000 GB/s

FP16 compute

2250 TFLOPS

TDP

1000 W

MSRP

—

Street price

—

Released

Nov 18, 2024

Architecture

Blackwell

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Nex-N2-Mini35.1BQ5_K_M262K218 tok/sestFits in memory Granite 4.1 8B8.8BQ4_K_M131K1200 tok/sestFits in memory Qwen3.6 35B A3B36BQ4_K_M262K2991 tok/sestFits in memory Gemma 4 26B A4B26.5BQ5_K_M262K1912 tok/sestFits in memory Gemma 4 31B32.7BQ8_0131K184 tok/sestFits in memory Qwen3.5-9B9.7BQ8_0262K621 tok/sestFits in memory Qwen3.5-27B27.8BQ4_K_M262K380 tok/sestFits in memory Qwen3.5-122B-A10B125BQ5_K_M262K765 tok/sestFits in memory GLM 4.7 Flash31.2BQ4_K_M131K288 tok/sestFits in memory Nemotron 3 Nano 30B A3B31.6BQ5_K_M262K2550 tok/sestFits in memory Ministral 3 14B 251213.9BQ8_0262K433 tok/sestFits in memory Ministral 3 8B 25128.9BQ5_K_M262K1011 tok/sestFits in memory Ministral 3 3B 25123.8BQ4_K_M131K2778 tok/sestFits in memory Olmo 3 32B Think32.2BQ4_K_M66K328 tok/sestFits in memory gpt-oss-safeguard-20b21.5BQ5_K_M131K356 tok/sestFits in memory Qwen3 VL 32B Instruct33.4BQ4_K_M131K316 tok/sestFits in memory Granite 4.0 Micro3.2BQ4_K_M66K2804 tok/sestFits in memory Qwen3 VL 8B Thinking8.8BQ8_0131K685 tok/sestFits in memory Qwen3 VL 30B A3B Thinking31.1BQ4_K_M262K2991 tok/sestFits in memory Cydonia 24B V4.123.6BQ5_K_M131K381 tok/sestFits in memory Qwen3 VL 235B A22B Thinking236BQ4_K_M131K408 tok/sestFits in memory Qwen3 Next 80B A3B Thinking81.3BQ4_K_M262K2991 tok/sestFits in memory Hermes 4 70B70.6BQ4_K_M131K150 tok/sestFits in memory UI-TARS 7B8.3BQ8_066K726 tok/sestFits in memory Uncensored24BQ4_K_M66K440 tok/sestFits in memory Hunyuan A13B Instruct80.4BQ5_K_M131K588 tok/sestFits in memory Mistral Small 3.2 24B24BQ4_K_M131K440 tok/sestFits in memory Llama Guard 4 12B12BQ5_K_M262K750 tok/sestFits in memory Qwen3 14B14.8BQ8_0131K407 tok/sestFits in memory Gemma 3 12B12.2BQ5_K_M131K738 tok/sestFits in memory Gemma 3 27B27.4BQ8_0262K220 tok/sestFits in memory Qwen2.5 VL 72B Instruct73.4BQ5_K_M66K123 tok/sestFits in memory Llama 3.3 Euryale 70B70.6BQ4_K_M131K150 tok/sestFits in memory Rocinante 12B12.2BQ8_066K494 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ4_K_M131K3299 tok/sestFits in memory Llama 3.1 Euryale 70B v2.270.6BQ5_K_M131K128 tok/sestFits in memory Mistral Nemo12.2BQ4_K_M131K865 tok/sestFits in memory Mixtral 8x22B Instruct141BQ8_066K131 tok/sestFits in memory Laguna XS 2.133.4BQ8_0262K153 tok/sestFits in memory Qwen3.6 27B27.8BQ4_K_M262K380 tok/sestFits in memory Qwen3.5-35B-A3B36BQ5_K_M262K2550 tok/sestFits in memory Qwen3 VL 8B Instruct8.8BQ8_0262K685 tok/sestFits in memory Qwen3 VL 30B A3B Instruct31.1BQ5_K_M262K2550 tok/sestFits in memory Qwen3 VL 235B A22B Instruct236BQ4_K_M131K408 tok/sestFits in memory Qwen3 Next 80B A3B Instruct81.3BQ5_K_M262K2550 tok/sestFits in memory Qwen3 30B A3B Thinking 250730.5BQ5_K_M66K2550 tok/sestFits in memory Qwen3 Coder 30B A3B Instruct30.5BQ4_K_M262K2991 tok/sestFits in memory Qwen3 30B A3B Instruct 250730.5BQ8_0262K1707 tok/sestFits in memory Qwen3 235B A22B Thinking 2507235BQ4_K_M131K408 tok/sestFits in memory Qwen3 235B A22B Instruct 2507235BQ4_K_M131K408 tok/sestFits in memory Qwen3 8B8.2BQ8_0131K735 tok/sestFits in memory Qwen3 32B32.8BQ8_0131K184 tok/sestFits in memory Mistral Small 3.1 24B24BQ5_K_M66K375 tok/sestFits in memory Gemma 3 4B4.3BQ8_0131K1401 tok/sestFits in memory Reka Flash 320.9BQ8_066K288 tok/sestFits in memory Hermes 3 70B Instruct70.6BQ5_K_M131K128 tok/sestFits in memory Llama 3.1 70B Instruct70.6BQ5_K_M131K128 tok/sestFits in memory Llama 3.1 8B Instruct8BQ5_K_M131K1125 tok/sestFits in memory Llama 3.3 70B Instruct70.6BQ5_K_M131K128 tok/sestFits in memory Niagara 19m Batch.en20MQ4_K_M~262K527835 tok/sestFits in memory

A speed marked est is calculated from this card's memory bandwidth and the size of the weights it has to read per word. It is not a measurement.

Something wrong on this page? Tell us