What GPU do you need to run SmallThinker 3B?

3.09B params 32K max context Reasoning

The smallest GPU memory that runs SmallThinker 3B well is 4 GB (Q8_0, e.g. T1000). 131 of 132 GPUs run it well and 1 more are a tight fit.

MMLU 65

VRAM needed per quantization

Quantization	Weights	16K context	Full 32K
Q4_K_M	1.9 GB	2.5 GB	3.0 GB
Q8_0	3.3 GB	3.8 GB	4.4 GB

Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.

Cheapest GPUs that can run SmallThinker 3B

Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.

Arc A580 8 GB · ~175 tok/s at Q4_K_M · Launch MSRP $179
Check price on Amazon
Arc B570 10 GB · ~130 tok/s at Q4_K_M · Launch MSRP $219
Check price on Amazon
Arc B580 12 GB · ~156 tok/s at Q4_K_M · Launch MSRP $249
Check price on Amazon

As an Amazon Associate, I earn from qualifying purchases.

GPUs that can run SmallThinker 3B

Best quantization that fits on each GPU, with the context and speed it gets. For chips with several memory sizes, the smallest that works is shown.

NVIDIA

T400 2 GB Q4_K_M 2K context · ~27 tok/s tight fit
T1000 4 GB Q8_0 20K context · ~37 tok/s
T400 4GB 4 GB Q8_0 20K context · ~19 tok/s
T600 4 GB Q8_0 20K context · ~37 tok/s
GTX 1060 6GB 6 GB Q8_0 32K context · ~45 tok/s
GTX 1660 6 GB Q8_0 32K context · ~45 tok/s
GTX 1660 SUPER 6 GB Q8_0 32K context · ~78 tok/s
GTX 1660 Ti 6 GB Q8_0 32K context · ~67 tok/s
RTX 2060 6 GB Q8_0 32K context · ~78 tok/s
RTX 3050 6GB 6 GB Q8_0 32K context · ~39 tok/s
RTX A2000 6 GB Q8_0 32K context · ~67 tok/s
GTX 1070 8 GB Q8_0 32K context · ~60 tok/s
GTX 1070 Ti 8 GB Q8_0 32K context · ~60 tok/s
GTX 1080 8 GB Q8_0 32K context · ~75 tok/s
RTX 2060 SUPER 8 GB Q8_0 32K context · ~104 tok/s
RTX 2070 8 GB Q8_0 32K context · ~104 tok/s
RTX 2070 SUPER 8 GB Q8_0 32K context · ~104 tok/s
RTX 2080 8 GB Q8_0 32K context · ~104 tok/s
RTX 2080 SUPER 8 GB Q8_0 32K context · ~116 tok/s
RTX 3050 8 GB Q8_0 32K context · ~52 tok/s
RTX 3060 Ti 8 GB Q8_0 32K context · ~104 tok/s
RTX 3060 Ti GDDR6X 8 GB Q8_0 32K context · ~142 tok/s
RTX 3070 8 GB Q8_0 32K context · ~104 tok/s
RTX 3070 Ti 8 GB Q8_0 32K context · ~142 tok/s
RTX 4060 8 GB Q8_0 32K context · ~63 tok/s
RTX 4060 Ti 8 GB Q8_0 32K context · ~67 tok/s
RTX 5050 8 GB Q8_0 32K context · ~75 tok/s
RTX 5060 8 GB Q8_0 32K context · ~104 tok/s
RTX 5060 Ti 8 GB Q8_0 32K context · ~104 tok/s
T1000 8GB 8 GB Q8_0 32K context · ~37 tok/s
RTX 3080 10 GB Q8_0 32K context · ~177 tok/s
GTX 1080 Ti 11 GB Q8_0 32K context · ~113 tok/s
RTX 2080 Ti 11 GB Q8_0 32K context · ~144 tok/s
RTX 2060 12GB 12 GB Q8_0 32K context · ~78 tok/s
RTX 3060 12 GB Q8_0 32K context · ~84 tok/s
RTX 3080 12GB 12 GB Q8_0 32K context · ~213 tok/s
RTX 3080 Ti 12 GB Q8_0 32K context · ~213 tok/s
RTX 4070 12 GB Q8_0 32K context · ~117 tok/s
RTX 4070 SUPER 12 GB Q8_0 32K context · ~117 tok/s
RTX 4070 Ti 12 GB Q8_0 32K context · ~117 tok/s
RTX 5070 12 GB Q8_0 32K context · ~157 tok/s
RTX A2000 12GB 12 GB Q8_0 32K context · ~67 tok/s
RTX 4060 Ti 16GB 16 GB Q8_0 32K context · ~67 tok/s
RTX 4070 Ti SUPER 16 GB Q8_0 32K context · ~157 tok/s
RTX 4080 16 GB Q8_0 32K context · ~167 tok/s
RTX 4080 SUPER 16 GB Q8_0 32K context · ~172 tok/s
RTX 5060 Ti 16GB 16 GB Q8_0 32K context · ~104 tok/s
RTX 5070 Ti 16 GB Q8_0 32K context · ~209 tok/s
RTX 5080 16 GB Q8_0 32K context · ~224 tok/s
RTX 2000 Ada 16 GB Q8_0 32K context · ~52 tok/s
RTX A4000 16 GB Q8_0 32K context · ~104 tok/s
RTX PRO 2000 Blackwell 16 GB Q8_0 32K context · ~67 tok/s
Tesla T4 16 GB Q8_0 32K context · ~75 tok/s
Tesla V100 16GB 16 GB Q8_0 32K context · ~210 tok/s
RTX 4000 Ada 20 GB Q8_0 32K context · ~84 tok/s
RTX 3090 24 GB Q8_0 32K context · ~218 tok/s
RTX 3090 Ti 24 GB Q8_0 32K context · ~235 tok/s
RTX 4090 24 GB Q8_0 32K context · ~235 tok/s
RTX 4500 Ada 24 GB Q8_0 32K context · ~101 tok/s
RTX A5000 24 GB Q8_0 32K context · ~179 tok/s
RTX PRO 4000 Blackwell 24 GB Q8_0 32K context · ~157 tok/s
RTX Spark 24 GB Q8_0 32K context · ~70 tok/s
Tesla P40 24 GB Q8_0 32K context · ~81 tok/s
RTX 5090 32 GB Q8_0 32K context · ~418 tok/s
RTX 5000 Ada 32 GB Q8_0 32K context · ~134 tok/s
RTX PRO 4500 Blackwell 32 GB Q8_0 32K context · ~209 tok/s
Tesla V100 32GB 32 GB Q8_0 32K context · ~210 tok/s
L40S 48 GB Q8_0 32K context · ~201 tok/s
RTX 6000 Ada 48 GB Q8_0 32K context · ~224 tok/s
RTX A6000 48 GB Q8_0 32K context · ~179 tok/s
RTX PRO 5000 Blackwell 48 GB Q8_0 32K context · ~313 tok/s
RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 32K context · ~313 tok/s
RTX PRO 5500 Blackwell 84 GB Q8_0 32K context · ~326 tok/s
RTX PRO 6000 Blackwell 96 GB Q8_0 32K context · ~418 tok/s
DGX Spark (GB 10) 128 GB Q8_0 32K context · ~64 tok/s

AMD

Radeon RX 5600 XT 6 GB Q8_0 32K context · ~67 tok/s
Radeon RX 5700 8 GB Q8_0 32K context · ~104 tok/s
Radeon RX 5700 XT 8 GB Q8_0 32K context · ~104 tok/s
Radeon RX 6600 8 GB Q8_0 32K context · ~52 tok/s
Radeon RX 6600 XT 8 GB Q8_0 32K context · ~60 tok/s
Radeon RX 6650 XT 8 GB Q8_0 32K context · ~65 tok/s
Radeon RX 7600 8 GB Q8_0 32K context · ~67 tok/s
Radeon RX 9060 8 GB Q8_0 32K context · ~67 tok/s
Radeon RX 9060 XT 8 GB Q8_0 32K context · ~75 tok/s
Radeon RX 6700 XT 12 GB Q8_0 32K context · ~90 tok/s
Radeon RX 6750 XT 12 GB Q8_0 32K context · ~101 tok/s
Radeon RX 7700 XT 12 GB Q8_0 32K context · ~101 tok/s
Radeon RX 9070 GRE 12 GB Q8_0 32K context · ~101 tok/s
AMD BC-250 16 GB Q8_0 32K context · ~104 tok/s
Radeon RX 6800 16 GB Q8_0 32K context · ~119 tok/s
Radeon RX 6800 XT 16 GB Q8_0 32K context · ~119 tok/s
Radeon RX 6900 XT 16 GB Q8_0 32K context · ~119 tok/s
Radeon RX 6950 XT 16 GB Q8_0 32K context · ~134 tok/s
Radeon RX 7600 XT 16 GB Q8_0 32K context · ~67 tok/s
Radeon RX 7800 XT 16 GB Q8_0 32K context · ~145 tok/s
Radeon RX 7900 GRE 16 GB Q8_0 32K context · ~134 tok/s
Radeon RX 9060 XT 16GB 16 GB Q8_0 32K context · ~75 tok/s
Radeon RX 9070 16 GB Q8_0 32K context · ~149 tok/s
Radeon RX 9070 XT 16 GB Q8_0 32K context · ~149 tok/s
Radeon RX 7900 XT 20 GB Q8_0 32K context · ~186 tok/s
Radeon RX 7900 XTX 24 GB Q8_0 32K context · ~224 tok/s
Radeon AI PRO R9700 32 GB Q8_0 32K context · ~149 tok/s
Ryzen AI Max (Strix Halo) 32 GB Q8_0 32K context · ~60 tok/s
Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 32K context · ~64 tok/s

Intel

Arc A580 8 GB Q8_0 32K context · ~119 tok/s
Arc A750 8 GB Q8_0 32K context · ~119 tok/s
Arc A770 8GB 8 GB Q8_0 32K context · ~119 tok/s
Arc B570 10 GB Q8_0 32K context · ~89 tok/s
Arc B580 12 GB Q8_0 32K context · ~106 tok/s
Arc A770 16GB 16 GB Q8_0 32K context · ~131 tok/s
Arc Pro B65 32 GB Q8_0 32K context · ~142 tok/s
Arc Pro B70 32 GB Q8_0 32K context · ~142 tok/s

Apple

M1 8 GB Q8_0 32K context · ~16 tok/s
M2 8 GB Q8_0 32K context · ~23 tok/s
M3 8 GB Q8_0 32K context · ~23 tok/s
M1 Pro 16 GB Q8_0 32K context · ~47 tok/s
M2 Pro 16 GB Q8_0 32K context · ~47 tok/s
M4 16 GB Q8_0 32K context · ~28 tok/s
M5 16 GB Q8_0 32K context · ~36 tok/s
M6 16 GB Q8_0 32K context · ~36 tok/s
M3 Pro 18 GB Q8_0 32K context · ~35 tok/s
M4 Pro 24 GB Q8_0 32K context · ~64 tok/s
M5 Pro 24 GB Q8_0 32K context · ~72 tok/s
M1 Max 32 GB Q8_0 32K context · ~93 tok/s
M2 Max 32 GB Q8_0 32K context · ~93 tok/s
M3 Max 36 GB Q8_0 32K context · ~70 tok/s
M4 Max 36 GB Q8_0 32K context · ~96 tok/s
M5 Max 36 GB Q8_0 32K context · ~107 tok/s
M1 Ultra 64 GB Q8_0 32K context · ~186 tok/s
M2 Ultra 64 GB Q8_0 32K context · ~186 tok/s
M3 Ultra 96 GB Q8_0 32K context · ~191 tok/s
M5 Ultra 96 GB Q8_0 32K context · ~280 tok/s

Check your exact setup in the full tool →