← AI Hashrate

A100 80GB SXM4

NVIDIA · 80 GB HBM2e · 2039 GB/s · 400 W · MSRP $15,000

The A100 80GB SXM4 is a NVIDIA accelerator with 80 GB HBM2e and about 2039 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 46 curated models fit fully on this card; at FP16, about 37 fit. Top Q4 speeds: Qwen3-0.6B ≈ 1621.9 tok/s (estimated, fits); gemma-3-1b-it ≈ 1297.5 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 1081.3 tok/s (estimated, fits); Qwen3-1.7B ≈ 648.8 tok/s (estimated, fits); Qwen3-30B-A3B ≈ 432.5 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?
Qwen3-0.6B0.8Q41621.9 est.Yes
gemma-3-1b-it1.0Q41297.5 est.Yes
Llama-3.2-1B-Instruct1.2Q41081.3 est.Yes
Qwen3-1.7B2.0Q4648.8 est.Yes
Qwen3-0.6B0.8FP16446.0 est.Yes
Qwen3-30B-A3B30.5Q4432.5 est.Yes
Qwen3-Coder-30B-A3B-Instruct30.5Q4432.5 est.Yes
Qwen3-30B-A3B-Instruct-250730.5Q4432.5 est.Yes
Qwen3-Next-80B-A3B-Instruct81.3Q4432.5 est.Yes
Qwen3-Coder-Next79.7Q4432.5 est.Yes
Llama-3.2-3B-Instruct3.2Q4405.5 est.Yes
gpt-oss-20b21.5Q4360.4 est.Yes
gemma-3-1b-it1.0FP16356.8 est.Yes
Phi-4-mini-instruct3.8Q4341.5 est.Yes
Phi-4-mini-reasoning3.8Q4341.5 est.Yes
gemma-3-4b-it4.3Q4332.7 est.Yes
Qwen3-4B-Instruct-25074.0Q4324.4 est.Yes
Qwen3-4B-Thinking-25074.0Q4324.4 est.Yes
Llama-3.2-1B-Instruct1.2FP16297.4 est.Yes
DeepSeek-V4-Flash158.1Q4222.6 est.No
Qwen2.5-7B-Instruct7.6FP16195.0 measured (vllm-bench)Yes
Mistral-7B-Instruct-v0.37.2Q4180.2 est.Yes
Llama-3.1-8B-Instruct8.0FP16180.0 measured (vllm-bench)Yes
Qwen3-1.7B2.0FP16178.4 est.Yes
Qwen2.5-Coder-7B-Instruct7.6Q4170.7 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6Q4170.7 est.Yes
Llama-3.1-8B-Instruct8.0Q4162.2 est.Yes
gpt-oss-120b120.4Q4162.2 est.Yes
Qwen3-8B8.2Q4158.2 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2Q4158.2 est.Yes
Qwen2.5-7B-Instruct7.6Q4154.05 measured (qwen-docs-speed-benchmark)Yes
MiniMax-M2.7228.7Q4143.0 est.No
Qwen3-30B-A3B30.5FP16118.9 est.Yes
Qwen3-Coder-30B-A3B-Instruct30.5FP16118.9 est.Yes
Qwen3-30B-A3B-Instruct-250730.5FP16118.9 est.Yes
Llama-3.2-3B-Instruct3.2FP16111.5 est.Yes
gemma-3-12b-it12.2Q4108.1 est.Yes
GLM-4.7-Flash31.2Q4108.1 est.Yes
GLM-4.5-Air110.5Q4108.1 est.Yes
Mixtral-8x7B-Instruct-v0.146.7Q4100.6 est.Yes
gpt-oss-20b21.5FP1699.1 est.Yes
Qwen2.5-14B-Instruct14.8Q498.02 measured (qwen-docs-speed-benchmark)Yes
Mistral-7B-Instruct-v0.37.2FP1696.8 measured (rbgo/llm-inference-benchmark)Yes
Phi-4-mini-instruct3.8FP1693.9 est.Yes
Phi-4-mini-reasoning3.8FP1693.9 est.Yes
gemma-3-4b-it4.3FP1691.5 est.Yes
Qwen3-4B-Instruct-25074.0FP1689.2 est.Yes
Qwen3-4B-Thinking-25074.0FP1689.2 est.Yes
Phi-414.7Q488.3 est.Yes
Phi-4-reasoning14.7Q488.3 est.Yes
Qwen3-14B14.8Q487.7 est.Yes
Qwen2.5-Coder-14B-Instruct14.8Q487.7 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8Q487.7 est.Yes
Qwen3-Coder-Next79.7FP1683.9 est.No
Qwen3-Next-80B-A3B-Instruct81.3FP1680.7 est.No
Qwen3-235B-A22B235.1Q459.4 est.No
Qwen3-235B-A22B-Instruct-2507235.1Q459.4 est.No
Qwen2.5-32B-Instruct32.8Q455.83 measured (qwen-docs-speed-benchmark)Yes
Devstral-Small-250723.6Q455.0 est.Yes
Mistral-Small-3.2-24B-Instruct-250624.0Q454.1 est.Yes
Magistral-Small-250924.0Q454.1 est.Yes
Mixtral-8x7B-Instruct-v0.146.7FP1653.8 est.No
Qwen2.5-72B-Instruct72.7Q448.0 measured (vllm-bench)Yes
gemma-3-27b-it27.4Q447.7 est.Yes
Qwen2.5-Coder-7B-Instruct7.6FP1647.0 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6FP1647.0 est.Yes
Qwen3-8B8.2FP1643.5 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2FP1643.5 est.Yes
Qwen3-32B32.8Q439.6 est.Yes
Qwen2.5-Coder-32B-Instruct32.8Q439.6 est.Yes
DeepSeek-R1-Distill-Qwen-32B32.8Q439.6 est.Yes
gemma-3-12b-it12.2FP1629.7 est.Yes
GLM-4.7-Flash31.2FP1629.7 est.Yes
Phi-414.7FP1624.3 est.Yes
Phi-4-reasoning14.7FP1624.3 est.Yes
Qwen3-14B14.8FP1624.1 est.Yes
Qwen2.5-14B-Instruct14.8FP1624.1 est.Yes
Qwen2.5-Coder-14B-Instruct14.8FP1624.1 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8FP1624.1 est.Yes
Llama-3.1-70B-Instruct70.6Q418.5 est.Yes
Llama-3.3-70B-Instruct70.6Q418.5 est.Yes
DeepSeek-R1-Distill-Llama-70B70.6Q418.4 est.Yes
GLM-4.5358.3Q417.7 est.No
Devstral-Small-250723.6FP1615.1 est.Yes
Mistral-Small-3.2-24B-Instruct-250624.0FP1614.9 est.Yes
Magistral-Small-250924.0FP1614.9 est.Yes
gpt-oss-120b120.4FP1613.8 est.No
gemma-3-27b-it27.4FP1613.1 est.Yes
Qwen3-32B32.8FP1610.9 est.Yes
Qwen2.5-32B-Instruct32.8FP1610.9 est.Yes
Qwen2.5-Coder-32B-Instruct32.8FP1610.9 est.Yes
DeepSeek-R1-Distill-Qwen-32B32.8FP1610.9 est.Yes
GLM-4.5-Air110.5FP1610.8 est.No
Qwen3-Coder-480B-A35B-Instruct480.2Q49.2 est.No
MiniMax-M1456.1Q47.6 est.No
DeepSeek-V4-Flash158.1FP164.9 est.No
Llama-3.1-70B-Instruct70.6FP163.8 est.No
Llama-3.3-70B-Instruct70.6FP163.8 est.No
DeepSeek-R1-Distill-Llama-70B70.6FP163.8 est.No
Qwen2.5-72B-Instruct72.7FP163.5 est.No
Llama-3.1-405B-Instruct405.9Q40.6 est.No