What GPU do you need to run GPT-OSS 20B?
21B params 128K max context ReasoningTool use
The smallest GPU memory that runs GPT-OSS 20B well is 16 GB (Q8_0, e.g. RTX PRO 2000 Blackwell). 72 of 132 GPUs run it well and 14 more are a tight fit.
MMLU 75 SWE-bench Verified 60.7
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 128K |
|---|---|---|---|---|
| Q4_K_M | 11.7 GB | 12.0 GB | 12.4 GB | 14.6 GB |
| Q8_0 | 12.1 GB | 12.5 GB | 12.8 GB | 15.0 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run GPT-OSS 20B
Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.
As an Amazon Associate, I earn from qualifying purchases.
GPUs that can run GPT-OSS 20B
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
- RTX 2060 12GB 12 GB Q4_K_M 14K context · ~27 tok/s tight fit
- RTX 3060 12 GB Q4_K_M 14K context · ~28 tok/s tight fit
- RTX 3080 12GB 12 GB Q4_K_M 14K context · ~72 tok/s tight fit
- RTX 3080 Ti 12 GB Q4_K_M 14K context · ~72 tok/s tight fit
- RTX 4070 12 GB Q4_K_M 14K context · ~40 tok/s tight fit
- RTX 4070 SUPER 12 GB Q4_K_M 14K context · ~40 tok/s tight fit
- RTX 4070 Ti 12 GB Q4_K_M 14K context · ~40 tok/s tight fit
- RTX 5070 12 GB Q4_K_M 14K context · ~53 tok/s tight fit
- RTX A2000 12GB 12 GB Q4_K_M 14K context · ~23 tok/s tight fit
- RTX 4060 Ti 16GB 16 GB Q8_0 128K context · ~22 tok/s
- RTX 4070 Ti SUPER 16 GB Q8_0 128K context · ~51 tok/s
- RTX 4080 16 GB Q8_0 128K context · ~55 tok/s
- RTX 4080 SUPER 16 GB Q8_0 128K context · ~56 tok/s
- RTX 5060 Ti 16GB 16 GB Q8_0 128K context · ~34 tok/s
- RTX 5070 Ti 16 GB Q8_0 128K context · ~68 tok/s
- RTX 5080 16 GB Q8_0 128K context · ~73 tok/s
- RTX 2000 Ada 16 GB Q8_0 128K context · ~17 tok/s
- RTX A4000 16 GB Q8_0 128K context · ~34 tok/s
- RTX PRO 2000 Blackwell 16 GB Q8_0 128K context · ~22 tok/s
- Tesla T4 16 GB Q8_0 128K context · ~24 tok/s
- Tesla V100 16GB 16 GB Q8_0 128K context · ~69 tok/s
- RTX 4000 Ada 20 GB Q8_0 128K context · ~27 tok/s
- RTX 3090 24 GB Q8_0 128K context · ~71 tok/s
- RTX 3090 Ti 24 GB Q8_0 128K context · ~77 tok/s
- RTX 4090 24 GB Q8_0 128K context · ~77 tok/s
- RTX 4500 Ada 24 GB Q8_0 128K context · ~33 tok/s
- RTX A5000 24 GB Q8_0 128K context · ~59 tok/s
- RTX PRO 4000 Blackwell 24 GB Q8_0 128K context · ~51 tok/s
- RTX Spark 24 GB Q8_0 128K context · ~23 tok/s
- Tesla P40 24 GB Q8_0 128K context · ~26 tok/s
- RTX 5090 32 GB Q8_0 128K context · ~137 tok/s
- RTX 5000 Ada 32 GB Q8_0 128K context · ~44 tok/s
- RTX PRO 4500 Blackwell 32 GB Q8_0 128K context · ~68 tok/s
- Tesla V100 32GB 32 GB Q8_0 128K context · ~69 tok/s
- L40S 48 GB Q8_0 128K context · ~66 tok/s
- RTX 6000 Ada 48 GB Q8_0 128K context · ~73 tok/s
- RTX A6000 48 GB Q8_0 128K context · ~59 tok/s
- RTX PRO 5000 Blackwell 48 GB Q8_0 128K context · ~103 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 128K context · ~103 tok/s
- RTX PRO 5500 Blackwell 84 GB Q8_0 128K context · ~107 tok/s
- RTX PRO 6000 Blackwell 96 GB Q8_0 128K context · ~137 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 128K context · ~21 tok/s
AMD
- Radeon RX 6700 XT 12 GB Q4_K_M 14K context · ~30 tok/s tight fit
- Radeon RX 6750 XT 12 GB Q4_K_M 14K context · ~34 tok/s tight fit
- Radeon RX 7700 XT 12 GB Q4_K_M 14K context · ~34 tok/s tight fit
- Radeon RX 9070 GRE 12 GB Q4_K_M 14K context · ~34 tok/s tight fit
- AMD BC-250 16 GB Q8_0 128K context · ~34 tok/s
- Radeon RX 6800 16 GB Q8_0 128K context · ~39 tok/s
- Radeon RX 6800 XT 16 GB Q8_0 128K context · ~39 tok/s
- Radeon RX 6900 XT 16 GB Q8_0 128K context · ~39 tok/s
- Radeon RX 6950 XT 16 GB Q8_0 128K context · ~44 tok/s
- Radeon RX 7600 XT 16 GB Q8_0 128K context · ~22 tok/s
- Radeon RX 7800 XT 16 GB Q8_0 128K context · ~48 tok/s
- Radeon RX 7900 GRE 16 GB Q8_0 128K context · ~44 tok/s
- Radeon RX 9060 XT 16GB 16 GB Q8_0 128K context · ~24 tok/s
- Radeon RX 9070 16 GB Q8_0 128K context · ~49 tok/s
- Radeon RX 9070 XT 16 GB Q8_0 128K context · ~49 tok/s
- Radeon RX 7900 XT 20 GB Q8_0 128K context · ~61 tok/s
- Radeon RX 7900 XTX 24 GB Q8_0 128K context · ~73 tok/s
- Radeon AI PRO R9700 32 GB Q8_0 128K context · ~49 tok/s
- Ryzen AI Max (Strix Halo) 32 GB Q8_0 128K context · ~20 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 128K context · ~21 tok/s
Intel
- Arc B580 12 GB Q4_K_M 14K context · ~36 tok/s tight fit
- Arc A770 16GB 16 GB Q8_0 128K context · ~43 tok/s
- Arc Pro B65 32 GB Q8_0 128K context · ~46 tok/s
- Arc Pro B70 32 GB Q8_0 128K context · ~46 tok/s
Apple
- M1 16 GB Q8_0 128K context · ~5 tok/s
- M1 Pro 16 GB Q8_0 128K context · ~15 tok/s
- M2 16 GB Q8_0 128K context · ~8 tok/s
- M2 Pro 16 GB Q8_0 128K context · ~15 tok/s
- M3 16 GB Q8_0 128K context · ~8 tok/s
- M4 16 GB Q8_0 128K context · ~9 tok/s
- M5 16 GB Q8_0 128K context · ~12 tok/s
- M6 16 GB Q8_0 128K context · ~12 tok/s
- M3 Pro 18 GB Q8_0 128K context · ~11 tok/s
- M4 Pro 24 GB Q8_0 128K context · ~21 tok/s
- M5 Pro 24 GB Q8_0 128K context · ~23 tok/s
- M1 Max 32 GB Q8_0 128K context · ~31 tok/s
- M2 Max 32 GB Q8_0 128K context · ~31 tok/s
- M3 Max 36 GB Q8_0 128K context · ~23 tok/s
- M4 Max 36 GB Q8_0 128K context · ~31 tok/s
- M5 Max 36 GB Q8_0 128K context · ~35 tok/s
- M1 Ultra 64 GB Q8_0 128K context · ~61 tok/s
- M2 Ultra 64 GB Q8_0 128K context · ~61 tok/s
- M3 Ultra 96 GB Q8_0 128K context · ~63 tok/s
- M5 Ultra 96 GB Q8_0 128K context · ~92 tok/s