gpt-oss-20b: how much VRAM?
In Q4_K_M with 8k tokens of context, gpt-oss-20b needs about 14 GB of memory. The cheapest option: RTX 4060 Ti 16 GB (€1,350 for the complete machine).
Model sheet
ollama run gpt-oss:20bWeights on Hugging Face: openai/gpt-oss-20b
VRAM by precision and context
| Precision | 4k | 8k | 32k | 128k |
|---|---|---|---|---|
| FP16 | 45 GB | 45 GB | 46 GB | 51 GB |
| FP8 | 23 GB | 23 GB | 24 GB | 29 GB |
| Q8_0 | 24 GB | 24 GB | 25 GB | 30 GB |
| Q6_K | 19 GB | 19 GB | 20 GB | 25 GB |
| Q4_K_M | 14 GB | 14 GB | 15 GB | 20 GB |
| Q3_K_M | 11 GB | 12 GB | 13 GB | 18 GB |
One user. Each simultaneous user adds their own context cache.
Frequently asked questions
How much VRAM does gpt-oss-20b need?
With an 8k-token context and one user, gpt-oss-20b needs about 14 GB in Q4, 24 GB in Q8 and 45 GB in FP16.
What is the cheapest hardware to run gpt-oss-20b?
In Q4, the cheapest option in our catalogue to run gpt-oss-20b is RTX 4060 Ti 16 GB, about €1,350 for the complete machine.
How fast does gpt-oss-20b run?
Expect around 69 tokens per second for one user on RTX 4060 Ti 16 GB in Q4 (estimate based on memory bandwidth).
On which hardware?
| Hardware | Cards | Speed | Machine |
|---|---|---|---|
| RTX 4060 Ti 16 GB | 1 is enough | ~69 tok/s | €1,350 |
| Radeon RX 7800 XT 16 GB | 1 is enough | ~149 tok/s | €1,400 |
| RTX 3060 12 GB | 2 cards | ~129 tok/s | €1,460 |
| RTX 3090 24 GB (used) | 1 is enough | ~223 tok/s | €1,650 |
| RTX 4070 Ti SUPER 16 GB | 1 is enough | ~160 tok/s | €1,750 |
| Radeon RX 7900 XTX 24 GB | 1 is enough | ~229 tok/s | €1,850 |
| RTX 4080 SUPER 16 GB | 1 is enough | ~175 tok/s | €1,950 |
| RTX 4070 12 GB | 2 cards | ~180 tok/s | €2,000 |
| RTX 5080 16 GB | 1 is enough | ~229 tok/s | €2,100 |
| Mini-PC Ryzen AI Max+ 395 128 GB | 1 is enough | ~61 tok/s | €2,300 |
| RTX 4090 24 GB | 1 is enough | ~240 tok/s | €2,900 |
| RTX 5090 32 GB | 1 is enough | ~427 tok/s | €3,500 |
| NVIDIA DGX Spark 128 GB | 1 is enough | ~65 tok/s | €4,000 |
| Mac Studio M4 Max 128 GB | 1 is enough | ~130 tok/s | €4,400 |
| RTX 6000 Ada 48 GB | 1 is enough | ~229 tok/s | €7,900 |
| RTX PRO 6000 Blackwell 96 GB | 1 is enough | ~427 tok/s | €9,900 |
| Mac Studio M3 Ultra 512 GB | 1 is enough | ~195 tok/s | €11,500 |
| L40S 48 GB | 1 is enough | ~206 tok/s | €13,500 |
| A100 80 GB | 1 is enough | ~485 tok/s | €17,000 |
| H100 80 GB | 1 is enough | ~798 tok/s | €34,000 |
| H200 141 GB | 1 is enough | ~1,143 tok/s | €39,000 |
| B200 192 GB | 1 is enough | ~1,905 tok/s | €48,000 |
Estimated complete machine: cards + host PC (€900 for 1–2 consumer cards, €2,800 beyond, a €6,000 server for datacenter cards). Macs and mini-PCs are sold complete.
Similar models
From: RTX 3060 12 GB · €1,180
From: RTX 3060 12 GB · €1,180
From: 2 × RTX 3060 12 GB · €1,460
From: 2 × RTX 3060 12 GB · €1,460