NVIDIA · datacenter

B200

B200 has 192 GB of VRAM at 8000 GB/s — about 178.56 GiB usable after driver and compositor overhead. 2108 of 2118 indexed models fit at 32K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
192 GB
HBM3e
Bandwidth
8000 GB/s
8192-bit bus
Tensor FP16
2250 TF
dense
TDP
1000 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1812vision language 192image 2audio tts 21audio asr 39video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 2108 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiMo-V2.5MoEKV unresolvedQ4_K_L311B176.24 GiB1.05 GiB178.34 GiB0.22 GiB143±37%
MiniMax-M2.7MoEUD-Q6_K229B175.15 GiB2.18 GiB178.31 GiB0.25 GiB136±37%
DeepSeek-R1-0528MoEIQ2_S685B176.61 GiB0.60 GiB178.29 GiB0.27 GiB147±37%
DeepSeek-V3.1-TerminusMoEIQ2_S685B176.61 GiB0.60 GiB178.29 GiB0.27 GiB147±37%
DeepSeek-V3.1MoEIQ2_S685B176.61 GiB0.60 GiB178.29 GiB0.27 GiB147±37%
Nex-N2-ProMoEIQ3_M397B176.93 GiB0.26 GiB178.25 GiB0.31 GiB168±37%
Trinity-Large-PreviewMoEQ3_K_M399B176.43 GiB0.75 GiB178.21 GiB0.35 GiB170±37%
command-a-plus-05-2026-bf16MoEQ6_K219B176.79 GiB0.40 GiB178.20 GiB0.36 GiB115±37%
Trinity-Large-TrueBaseMoEI1-Q3_K_M399B176.30 GiB0.75 GiB178.08 GiB0.48 GiB170±37%
MiniMax-M2.1MoEQ6_K229B174.91 GiB2.18 GiB178.08 GiB0.48 GiB136±37%
MiniMax-M2MoEQ6_K229B174.91 GiB2.18 GiB178.08 GiB0.48 GiB136±37%
MiniMax-M2.5MoEQ6_K229B174.87 GiB2.18 GiB178.03 GiB0.53 GiB136±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ6_K229B174.87 GiB2.18 GiB178.03 GiB0.53 GiB136±37%
Trinity-Large-ThinkingMoEIQ3_M399B176.24 GiB0.75 GiB178.02 GiB0.54 GiB170±37%
Ornith-1.0-397BMoEIQ3_M397B176.54 GiB0.26 GiB177.86 GiB0.70 GiB168±37%
ERNIE-4.5-300B-A47B-PTQ4_1300B174.57 GiB1.90 GiB177.59 GiB0.97 GiB26±22%
GLM-4.6-REAP-268B-A32BMoEQ5_K_S269B172.64 GiB3.23 GiB176.92 GiB1.64 GiB101±37%
MiMo-V2-FlashMoEKV unresolvedQ4_K_L310B174.72 GiB1.05 GiB176.82 GiB1.74 GiB144±37%
Qwen3.5-397B-A17BMoEQ3_K_M403B175.29 GiB0.26 GiB176.60 GiB1.96 GiB169±37%
grok-2MoEQ5_K_S270B173.10 GiB2.25 GiB176.49 GiB2.07 GiB48±37%
MiniMax-M3MoEQ3_K_S427B174.39 GiB1.05 GiB176.46 GiB2.10 GiB145±37%
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16MoEUD-IQ1_M561B175.12 GiB0.00 GiB176.15 GiB2.41 GiB137±37%
DeepSeek-V3-0324MoEUD-IQ1_S685B173.45 GiB0.60 GiB175.13 GiB3.43 GiB149±37%
Hermes-3-Llama-3.1-405BIQ3_M406B169.26 GiB4.43 GiB174.97 GiB3.59 GiB26±22%
cogito-v2-preview-deepseek-671B-MoEMoEUD-IQ1_S671B173.12 GiB0.60 GiB174.80 GiB3.76 GiB149±37%
cogito-671b-v2.1MoEUD-IQ1_S671B173.11 GiB0.60 GiB174.79 GiB3.77 GiB149±37%
Qwen3.5-REAP-262B-A17BMoEQ5_K_M262B172.96 GiB0.26 GiB174.27 GiB4.29 GiB154±37%
DeepSeek-TNG-R1T2-ChimeraMoEUD-IQ1_S685B172.39 GiB0.60 GiB174.07 GiB4.49 GiB150±37%
Hy3MoEQ4_K_L299B169.99 GiB2.81 GiB173.84 GiB4.72 GiB124±37%
DeepSeek-Prover-V2-671BMoEUD-IQ1_S685B172.10 GiB0.60 GiB173.78 GiB4.78 GiB150±37%
DeepSeek-V3.2MoEUD-IQ1_S685B171.48 GiB0.60 GiB173.16 GiB5.40 GiB150±37%
Minimax-M3-abliterated-cleanMoEQ3_K_S427B171.06 GiB1.05 GiB173.13 GiB5.43 GiB147±37%
GLM-4.7-REAP-218B-A32BMoEQ6_K218B167.57 GiB3.23 GiB171.84 GiB6.72 GiB94±37%
GLM-5.2MoEQ3_K_M753B169.33 GiB0.77 GiB171.15 GiB7.41 GiB151±37%
Solar-Open2-250BMoEQ5_K_M250B165.65 GiB1.69 GiB168.36 GiB10.20 GiB154±37%
Qwen3-Coder-480B-A35B-InstructMoEQ2_K_L480B162.87 GiB2.18 GiB166.08 GiB12.48 GiB124±37%
GLM-5MoEUD-TQ1_0754B164.05 GiB0.77 GiB165.87 GiB12.69 GiB155±37%
GLM-5.1MoEIQ1_M754B163.93 GiB0.77 GiB165.75 GiB12.81 GiB155±37%
Step-3.7-FlashQ6_K_L201B160.20 GiB3.67 GiB164.89 GiB13.67 GiB28±22%
Qwen3-Coder-REAP-363B-A35BMoEQ3_K_M363B161.57 GiB2.18 GiB164.79 GiB13.77 GiB112±37%
r1-1776MoEIQ2_XXS671B162.45 GiB0.60 GiB164.13 GiB14.43 GiB158±37%
DeepSeek-R1MoEIQ2_XXS685B162.45 GiB0.60 GiB164.13 GiB14.43 GiB158±37%
GLM-4.5MoEQ3_K_M358B159.51 GiB3.23 GiB163.78 GiB14.78 GiB120±37%
GLM-4.7MoEQ3_K_M358B159.51 GiB3.23 GiB163.78 GiB14.78 GiB120±37%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedQ3_K_S402B160.80 GiB1.69 GiB163.51 GiB15.05 GiB187±37%
GLM-4.6MoEQ3_K_M357B158.89 GiB3.23 GiB163.16 GiB15.40 GiB121±37%
GLM-4.6-Derestricted-v3MoEQ3_K_L357B158.25 GiB3.23 GiB162.52 GiB16.04 GiB121±37%
Qwen3-235B-A22B-Thinking-2507MoEQ5_K_M235B155.43 GiB1.65 GiB158.11 GiB20.45 GiB121±37%
Qwen3-235B-A22B-Instruct-2507MoEQ5_K_M235B155.43 GiB1.65 GiB158.11 GiB20.45 GiB121±37%
Qwen3-VL-235B-A22B-ThinkingMoEQ5_K_M236B155.36 GiB1.65 GiB158.04 GiB20.52 GiB121±37%
Qwen3-VL-235B-A22B-InstructMoEQ5_K_M236B155.36 GiB1.65 GiB158.04 GiB20.52 GiB121±37%
Qwen3-235B-A22BMoEQ5_K_M235B155.36 GiB1.65 GiB158.04 GiB20.52 GiB121±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q5_K_M235B155.36 GiB1.65 GiB158.04 GiB20.52 GiB121±37%
DeepSeek-Coder-V2-Instruct-0724MoEQ5_K236B155.74 GiB0.59 GiB157.37 GiB21.19 GiB155±37%
DeepSeek-V2.5MoEQ5_K236B155.74 GiB0.59 GiB157.37 GiB21.19 GiB155±37%
DeepSeek-Coder-V2-InstructMoEQ5_K_M236B155.74 GiB0.59 GiB157.37 GiB21.19 GiB155±37%
step-3.5-flashQ6_K199B150.81 GiB3.67 GiB155.51 GiB23.05 GiB29±22%
DeepSeek-V4-FlashMoEQ4_K291B153.33 GiB0.02 GiB154.39 GiB24.17 GiB186±37%
Hunyuan-A13B-InstructMoEBF1680.4B149.76 GiB1.13 GiB151.89 GiB26.67 GiB30±22%
dots.llm1.instMoEQ8_0143B141.37 GiB8.72 GiB151.12 GiB27.44 GiB105±37%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a B200 run?
2108 of 2118 indexed open-weight models fit a B200 at 32,768 context with q4_0 KV cache, the largest being MiMo-V2.5 at Q4_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a B200 actually have?
Its nameplate is 192 GB, but about 178.56 GiB is available to a model once driver and compositor overhead is accounted for.
Is a B200 fast for local AI?
Its memory bandwidth is 8000 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.