Hardware

NVIDIA B200

NVIDIA B200 has 192 GB HBM3e and 8000 GB/s. See which open models fit, at which quant, and roughly how fast. Figures below use the default memory (192 GB), 32 GB system RAM, and DDR5-5600 for anything that spills. Updated 2026-10-17.

Memory192 GB HBM3e

Bandwidth8000 GB/s

Kindgpu

Year2024

192GB HBM3e, 8 TB/s class bandwidth. Source.

RSS for new models on this device

Models that fit

Model	Verdict	Quant	Context	tok/s
MiMo V2 Flash 309B	Tight fit	UD-Q4_K_XL	8K	132–180
DeepSeek V4 Flash 0731 284B	Runs	UD-Q8_K_XL	8K	153–208
DeepSeek V4 Flash Vision Exp 284B	Runs	UD-Q8_K_XL	8K	153–208
GLM 5.3 Flash 321B	Runs	UD-IQ4_XS	8K	157–215
Llama 4 Maverick 17B 128E Instruct 401B	Tight fit	UD-Q3_K_XL	8K	147–200
MiniMax M2 229B	Tight fit	Q6_K	4K	125–171
Qwen3 235B A22B Thinking 2507 235B	Tight fit	UD-Q5_K_XL	8K	256–349
Qwen3 235B A22B Instruct 2507 235B	Tight fit	UD-Q5_K_XL	8K	253–345
Qwen3.5 397B A17B 396B	Tight fit	UD-Q3_K_XL	8K	556–758
GLM 4.6 357B	Runs	UD-Q3_K_XL	8K	127–173
DeepSeek R1 671B	Tight fit	UD-IQ1_M	8K	141–192
DeepSeek V3.2 671B	Tight fit	UD-IQ1_S	8K	130–177
Step 3.7 Flash 197B	Tight fit	UD-Q6_K_XL	4K	116–159
Qwen3.8 Flash Next 177B	Tight fit	Q8_0	8K	137–187
Qwen3 Coder 480B A35B Instruct 480B	Tight fit	UD-Q2_K_XL	8K	289–394
Mistral Medium 3.5 128B 125B	Runs	UD-Q8_K_XL	8K	30–41
Devstral 2 123B Instruct 2512 125B	Runs	UD-Q8_K_XL	8K	30–40
Qwen3.5 122B A10B MTP 125B	Runs	UD-Q8_K_XL	8K	399–544
Mistral Small 4 119B 2603 119B	Runs	UD-Q8_K_XL	8K	182–248
gpt oss 120b 117B	Runs great	UD-Q8_K_XL	8K	358–488
NVIDIA Nemotron 3 Super 120B A12B 121B	Runs	UD-Q8_K_XL	8K	316–431
GLM 4.5 Air 110B	Runs	UD-Q8_K_XL	8K	153–209
Llama 4 Scout 17B 16E Instruct 108B	Runs	UD-Q8_K_XL	8K	153–208
GLM 4.6V 107B	Runs	UD-Q8_K_XL	8K	158–215
Qwen2.5 72B Instruct 72.7B	Runs great	Q8_0	8K	55–75
Llama 3.3 70B Instruct 70.6B	Runs	BF16	8K	31–42
DeepSeek R1 Distill Llama 70B 70.6B	Runs	BF16	8K	31–42
Hermes 4 70B 70.6B	Runs	BF16	8K	31–42
Qwen3 Coder Next 79.7B	Runs	BF16	8K	158–216
Qwen3 Next 80B A3B Instruct 79.7B	Runs	BF16	8K	646–881
Qwen3 Next 80B A3B Thinking 79.7B	Runs	BF16	8K	646–881
Qwen2.5 Coder 32B Instruct 32.8B	Runs great	Q8_0	8K	119–162
Qwen2.5 32B Instruct 32.8B	Runs great	F16	8K	65–89
Qwen3 32B 32.8B	Runs great	BF16	8K	65–89
DeepSeek R1 Distill Qwen 32B 32.8B	Runs great	F16	8K	65–89
Qwen3 VL 32B Instruct 32.8B	Runs great	BF16	8K	65–89
GLM 4 32B 0414 32.6B	Runs great	BF16	8K	67–91
gemma 4 31B it 30.7B	Runs great	BF16	8K	63–86
Qwen3.8 27B 27.3B	Runs great	BF16	8K	80–109
gemma 3 27b it 27.0B	Runs great	BF16	8K	76–103

90 more models in the catalog do not fit this default profile, or ranked below the first 40 that do.

Example: Tight fit fully on your NVIDIA B200 at UD-Q4_K_XL, 8K context, about 132–180 tok/s. Weights 165.5 GB plus 1.13 GB of KV cache — 168.5 GB of 182.4 GB usable. Up to 64K context still fits. Weights shown in the verdict use GiB (the same 1024³ counting as GPU memory).