← AI Hashrate

B200 192GB SXM

NVIDIA · 192 GB HBM3e · 8000 GB/s · 1000 W · MSRP $40,000

The B200 192GB SXM is a NVIDIA accelerator with 192 GB HBM3e and about 8000 GB/s peak memory bandwidth. AI Hashrate estimates local LLM decode (batch 1) at 8K context: fit counts weights + KV cache + 1 GB runtime overhead — not weights alone. At Q4, about 50 curated models fit fully on this card; at FP16, about 44 fit. Top Q4 speeds: gemma-3-1b-it ≈ 5000 tok/s (estimated, fits); Qwen3-0.6B ≈ 5000 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 4242.4 tok/s (estimated, fits); Qwen3-1.7B ≈ 2545.5 tok/s (estimated, fits); Qwen3-30B-A3B ≈ 1697.0 tok/s (estimated, fits). Relative ranking is more reliable than absolute tok/s. See methodology for the bandwidth formula and measured-anchor policy. Methodology.

Table: decode speed estimates (Q4 / FP16). Measured rows override estimates. MSRP is list price, not live retail. Context default 8K.

ModelParams (B)Quanttok/sFits?
gemma-3-1b-it1.0Q45000 est.Yes
Qwen3-0.6B0.8Q45000 est.Yes
Llama-3.2-1B-Instruct1.2Q44242.4 est.Yes
Qwen3-1.7B2.0Q42545.5 est.Yes
Qwen3-0.6B0.8FP161750.0 est.Yes
Qwen3-30B-A3B30.5Q41697.0 est.Yes
Qwen3-Coder-30B-A3B-Instruct30.5Q41697.0 est.Yes
Qwen3-30B-A3B-Instruct-250730.5Q41697.0 est.Yes
Qwen3-Next-80B-A3B-Instruct81.3Q41697.0 est.Yes
Qwen3-Coder-Next79.7Q41697.0 est.Yes
Llama-3.2-3B-Instruct3.2Q41590.9 est.Yes
gpt-oss-20b21.5Q41414.1 est.Yes
gemma-3-1b-it1.0FP161400.0 est.Yes
Phi-4-mini-instruct3.8Q41339.7 est.Yes
Phi-4-mini-reasoning3.8Q41339.7 est.Yes
gemma-3-4b-it4.3Q41305.4 est.Yes
Qwen3-4B-Instruct-25074.0Q41272.7 est.Yes
Qwen3-4B-Thinking-25074.0Q41272.7 est.Yes
Llama-3.2-1B-Instruct1.2FP161166.7 est.Yes
Mistral-7B-Instruct-v0.37.2Q4707.1 est.Yes
Qwen3-1.7B2.0FP16700.0 est.Yes
Qwen2.5-7B-Instruct7.6Q4669.9 est.Yes
Qwen2.5-Coder-7B-Instruct7.6Q4669.9 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6Q4669.9 est.Yes
Llama-3.1-8B-Instruct8.0Q4636.4 est.Yes
gpt-oss-120b120.4Q4636.4 est.Yes
Qwen3-8B8.2Q4620.8 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2Q4620.8 est.Yes
Llama-3.1-70B-Instruct70.6Q4520.0 measured (nvidia-gtc)Yes
MiniMax-M2.7228.7Q4509.1 est.Yes
Qwen3-30B-A3B30.5FP16466.7 est.Yes
Qwen3-Coder-30B-A3B-Instruct30.5FP16466.7 est.Yes
Qwen3-30B-A3B-Instruct-250730.5FP16466.7 est.Yes
Qwen3-Next-80B-A3B-Instruct81.3FP16466.7 est.Yes
Qwen3-Coder-Next79.7FP16466.7 est.Yes
Llama-3.2-3B-Instruct3.2FP16437.5 est.Yes
gemma-3-12b-it12.2Q4424.2 est.Yes
GLM-4.7-Flash31.2Q4424.2 est.Yes
GLM-4.5-Air110.5Q4424.2 est.Yes
GLM-4.5358.3Q4400.8 est.No
Mixtral-8x7B-Instruct-v0.146.7Q4394.6 est.Yes
DeepSeek-V4-Flash158.1Q4391.6 est.Yes
gpt-oss-20b21.5FP16388.9 est.Yes
Phi-4-mini-instruct3.8FP16368.4 est.Yes
Phi-4-mini-reasoning3.8FP16368.4 est.Yes
gemma-3-4b-it4.3FP16359.0 est.Yes
Qwen3-4B-Instruct-25074.0FP16350.0 est.Yes
Qwen3-4B-Thinking-25074.0FP16350.0 est.Yes
Phi-414.7Q4346.3 est.Yes
Phi-4-reasoning14.7Q4346.3 est.Yes
Qwen3-14B14.8Q4344.0 est.Yes
Qwen2.5-14B-Instruct14.8Q4344.0 est.Yes
Qwen2.5-Coder-14B-Instruct14.8Q4344.0 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8Q4344.0 est.Yes
gpt-oss-120b120.4FP16311.1 est.No
GLM-4.5-Air110.5FP16244.0 est.No
Qwen3-235B-A22B235.1Q4231.4 est.Yes
Qwen3-235B-A22B-Instruct-2507235.1Q4231.4 est.Yes
Devstral-Small-250723.6Q4215.7 est.Yes
Mistral-Small-3.2-24B-Instruct-250624.0Q4212.1 est.Yes
Magistral-Small-250924.0Q4212.1 est.Yes
Qwen3-Coder-480B-A35B-Instruct480.2Q4206.9 est.No
Mistral-7B-Instruct-v0.37.2FP16194.4 est.Yes
gemma-3-27b-it27.4Q4187.2 est.Yes
Qwen2.5-7B-Instruct7.6FP16184.2 est.Yes
Qwen2.5-Coder-7B-Instruct7.6FP16184.2 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6FP16184.2 est.Yes
Llama-3.1-8B-Instruct8.0FP16175.0 est.Yes
MiniMax-M1456.1Q4171.4 est.No
Qwen3-8B8.2FP16170.7 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2FP16170.7 est.Yes
Qwen3-32B32.8Q4155.2 est.Yes
Qwen2.5-32B-Instruct32.8Q4155.2 est.Yes
Qwen2.5-Coder-32B-Instruct32.8Q4155.2 est.Yes
DeepSeek-R1-Distill-Qwen-32B32.8Q4155.2 est.Yes
gemma-3-12b-it12.2FP16116.7 est.Yes
GLM-4.7-Flash31.2FP16116.7 est.Yes
DeepSeek-V4-Flash158.1FP16110.9 est.No
Mixtral-8x7B-Instruct-v0.146.7FP16108.5 est.Yes
DeepSeek-R1684.5Q497.8 est.No
DeepSeek-V3.2685.4Q497.6 est.No
Phi-414.7FP1695.2 est.Yes
Phi-4-reasoning14.7FP1695.2 est.Yes
Qwen3-14B14.8FP1694.6 est.Yes
Qwen2.5-14B-Instruct14.8FP1694.6 est.Yes
Qwen2.5-Coder-14B-Instruct14.8FP1694.6 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8FP1694.6 est.Yes
GLM-5.2753.3Q474.8 est.No
GLM-5.1753.9Q474.7 est.No
Llama-3.3-70B-Instruct70.6Q472.7 est.Yes
DeepSeek-R1-Distill-Llama-70B70.6Q472.1 est.Yes
Qwen2.5-72B-Instruct72.7Q470.0 est.Yes
MiniMax-M2.7228.7FP1669.6 est.No
Devstral-Small-250723.6FP1659.3 est.Yes
Mistral-Small-3.2-24B-Instruct-250624.0FP1658.3 est.Yes
Magistral-Small-250924.0FP1658.3 est.Yes
gemma-3-27b-it27.4FP1651.5 est.Yes
Kimi-K2-Instruct-09051026.5Q451.2 est.No
Kimi-K2-Thinking1058.1Q448.2 est.No
DeepSeek-V4-Pro861.6Q446.6 est.No
Qwen3-32B32.8FP1642.7 est.Yes
Qwen2.5-32B-Instruct32.8FP1642.7 est.Yes
Qwen2.5-Coder-32B-Instruct32.8FP1642.7 est.Yes
DeepSeek-R1-Distill-Qwen-32B32.8FP1642.7 est.Yes
Qwen3-235B-A22B235.1FP1629.6 est.No
Qwen3-235B-A22B-Instruct-2507235.1FP1629.6 est.No
Llama-3.1-70B-Instruct70.6FP1620.0 est.Yes
Llama-3.3-70B-Instruct70.6FP1620.0 est.Yes
DeepSeek-R1-Distill-Llama-70B70.6FP1619.8 est.Yes
Qwen2.5-72B-Instruct72.7FP1619.3 est.Yes
Llama-3.1-405B-Instruct405.9Q414.2 est.No
GLM-4.5358.3FP168.8 est.No