OpenAI on NVIDIA Ampere
Can I run gpt-oss 20B-A3.6B on an RTX 3080 10GB?
Not at Q4_K_M — it needs 13.1 GB against 8.6 GB available. You would need 2 of these cards.
Won't fit
13.1 GB of 8.6 GB · 152%014 GB
Weights
11.9 GB
KV cache
0.4 GB
Runtime overhead
0.8 GB
Over the limit
4.5 GB
Short by 4.5 GB. You can run it with 14 of 24 layers on the RTX 3080 10GB and the rest in system RAM, at roughly 36.3 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.
Generation36.3tok/s
Prompt processing3500tok/s
Max context0tokens
KV per 1K tokens0GB
Every quantisation of gpt-oss 20B-A3.6B on a RTX 3080 10GB
Highlighted row is the highest quality that still fits at 8K context.
| Quantisation | bpw | Weights | Total | Verdict | Max ctx | tok/s |
|---|---|---|---|---|---|---|
| FP16 / BF16 | 16.00 | 38.9 GB | 40.1 GB | Won't fit | — | 6.78 |
| INT8 / W8A8 | 8.50 | 20.4 GB | 21.6 GB | Won't fit | — | 15.2 |
| FP8 (E4M3) | 8.00 | 19.5 GB | 20.6 GB | Won't fit | — | 16.8 |
| Q6_K | 6.56 | 16.0 GB | 17.1 GB | Won't fit | — | 22.7 |
| Q5_K_M | 5.67 | 13.8 GB | 15.0 GB | Won't fit | — | 27.6 |
| Q5_K_S | 5.52 | 13.4 GB | 14.6 GB | Won't fit | — | 30.3 |
| AWQ 4-bit | 4.25 | 11.9 GB | 13.1 GB | Won't fit | — | 36.2 |
| GPTQ 4-bit | 4.25 | 11.9 GB | 13.1 GB | Won't fit | — | 36.2 |
| MXFP4 | 4.25 | 11.9 GB | 13.1 GB | Won't fit | — | 36.2 |
| Q4_K_M | 4.85 | 11.9 GB | 13.1 GB | Won't fit | — | 36.3 |
| Q4_K_S | 4.58 | 11.3 GB | 12.4 GB | Won't fit | — | 41.2 |
| Q4_0 | 4.55 | 11.2 GB | 12.4 GB | Won't fit | — | 41.5 |
| IQ4_XS | 4.25 | 10.5 GB | 11.7 GB | Won't fit | — | 47.8 |
| Q3_K_M | 3.91 | 9.7 GB | 10.9 GB | Won't fit | — | 62.1 |
| IQ3_M | 3.70 | 9.2 GB | 10.4 GB | Won't fit | — | 72.8 |
| IQ3_XXS | 3.06 | 7.8 GB | 8.9 GB | Won't fit | 512 | 131 |
| Q2_K | 2.63 | 6.8 GB | 8.0 GB | Fits, but tight | 22K | 231 |
| IQ2_XXS | 2.06 | 5.5 GB | 6.7 GB | Runs comfortably | 50K | 264 |
| IQ1_M | 1.75 | 4.8 GB | 5.9 GB | Runs comfortably | 65K | 287 |