← AI Hashrate

T4 16GB

NVIDIA · 16 GB GDDR6 · 320 GB/s · 70 W · MSRP $2,000 · Search Amazon (affiliate)

The T4 16GB is a NVIDIA accelerator with 16 GB GDDR6 and about 320 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 25 curated models fit fully on this card; at FP16, about 10 fit. Top Q4 speeds: Qwen3-0.6B ≈ 254.5 tok/s (estimated, fits); gemma-3-1b-it ≈ 203.6 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 169.7 tok/s (estimated, fits); Qwen3-1.7B ≈ 101.8 tok/s (estimated, fits); Llama-3.2-3B-Instruct ≈ 63.6 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?
Qwen3-0.6B0.8Q4254.5 est.Yes
gemma-3-1b-it1.0Q4203.6 est.Yes
Llama-3.2-1B-Instruct1.2Q4169.7 est.Yes
Qwen3-30B-A3B30.5Q4147.0 est.No
Qwen3-Coder-30B-A3B-Instruct30.5Q4147.0 est.No
Qwen3-30B-A3B-Instruct-250730.5Q4147.0 est.No
Qwen3-1.7B2.0Q4101.8 est.Yes
Qwen3-0.6B0.8FP1670.0 est.Yes
Llama-3.2-3B-Instruct3.2Q463.6 est.Yes
gpt-oss-20b21.5Q456.6 est.Yes
gemma-3-1b-it1.0FP1656.0 est.Yes
Phi-4-mini-instruct3.8Q453.6 est.Yes
Phi-4-mini-reasoning3.8Q453.6 est.Yes
gemma-3-4b-it4.3Q452.2 est.Yes
Qwen3-4B-Instruct-25074.0Q450.9 est.Yes
Qwen3-4B-Thinking-25074.0Q450.9 est.Yes
Llama-3.2-1B-Instruct1.2FP1646.7 est.Yes
GLM-4.7-Flash31.2Q429.4 est.No
Mistral-7B-Instruct-v0.37.2Q428.3 est.Yes
Qwen3-1.7B2.0FP1628.0 est.Yes
Qwen2.5-7B-Instruct7.6Q426.8 est.Yes
Qwen2.5-Coder-7B-Instruct7.6Q426.8 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6Q426.8 est.Yes
Llama-3.1-8B-Instruct8.0Q425.5 est.Yes
Qwen3-8B8.2Q424.8 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2Q424.8 est.Yes
Qwen3-Coder-Next79.7Q424.0 est.No
Qwen3-Next-80B-A3B-Instruct81.3Q423.1 est.No
Mistral-7B-Instruct-v0.37.2FP1620.1 est.No
Devstral-Small-250723.6Q418.0 est.No
Llama-3.2-3B-Instruct3.2FP1617.5 est.Yes
Qwen2.5-7B-Instruct7.6FP1617.2 est.No
Qwen2.5-Coder-7B-Instruct7.6FP1617.2 est.No
DeepSeek-R1-Distill-Qwen-7B7.6FP1617.2 est.No
Mistral-Small-3.2-24B-Instruct-250624.0Q417.2 est.No
Magistral-Small-250924.0Q417.2 est.No
gemma-3-12b-it12.2Q417.0 est.Yes
Llama-3.1-8B-Instruct8.0FP1614.8 est.No
Phi-4-mini-instruct3.8FP1614.7 est.Yes
Phi-4-mini-reasoning3.8FP1614.7 est.Yes
gemma-3-4b-it4.3FP1614.4 est.Yes
Qwen3-4B-Instruct-25074.0FP1614.0 est.Yes
Qwen3-4B-Thinking-25074.0FP1614.0 est.Yes
Phi-414.7Q413.9 est.Yes
Phi-4-reasoning14.7Q413.9 est.Yes
Qwen3-8B8.2FP1613.8 est.No
Qwen3-14B14.8Q413.8 est.Yes
Qwen2.5-14B-Instruct14.8Q413.8 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2FP1613.8 est.No
Qwen2.5-Coder-14B-Instruct14.8Q413.8 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8Q413.8 est.Yes
Mixtral-8x7B-Instruct-v0.146.7Q413.5 est.No
gemma-3-27b-it27.4Q411.8 est.No
Qwen3-32B32.8Q46.9 est.No
Qwen2.5-32B-Instruct32.8Q46.9 est.No
Qwen2.5-Coder-32B-Instruct32.8Q46.9 est.No
DeepSeek-R1-Distill-Qwen-32B32.8Q46.9 est.No
gpt-oss-20b21.5FP165.7 est.No
gemma-3-12b-it12.2FP164.4 est.No
Qwen3-30B-A3B30.5FP163.5 est.No
Qwen3-Coder-30B-A3B-Instruct30.5FP163.5 est.No
Qwen3-30B-A3B-Instruct-250730.5FP163.5 est.No
Qwen3-14B14.8FP162.5 est.No
Qwen2.5-14B-Instruct14.8FP162.5 est.No
Qwen2.5-Coder-14B-Instruct14.8FP162.5 est.No
DeepSeek-R1-Distill-Qwen-14B14.8FP162.5 est.No
Phi-414.7FP162.5 est.No
Phi-4-reasoning14.7FP162.5 est.No
Qwen2.5-72B-Instruct72.7Q40.7 est.No
Llama-3.1-70B-Instruct70.6Q40.7 est.No
Llama-3.3-70B-Instruct70.6Q40.7 est.No
DeepSeek-R1-Distill-Llama-70B70.6Q40.7 est.No
Mistral-Small-3.2-24B-Instruct-250624.0FP160.6 est.No
Magistral-Small-250924.0FP160.6 est.No
Devstral-Small-250723.6FP160.6 est.No
gemma-3-27b-it27.4FP160.4 est.No