Can the NVIDIA RTX 3090 run gpt-oss 20B?
- FP16
- Q8_0
- Q6_K
- Q5_K_M
- Q4_K_M
- Q3_K_M
- Q2_K
Yes. The NVIDIA RTX 3090 can run gpt-oss 20B fully in VRAM at Q8_0 (near-lossless), using about 23.2 GB of the 24 GB available, at roughly 147 tokens/second.
Run it with Ollama
With Ollama installed, one command downloads and starts gpt-oss 20B:
ollama run gpt-oss:20bOllama's default download is Q4_K_M (about 13.7 GB here). You have room for Q8_0, which is near-lossless: pick that GGUF in LM Studio or from Hugging Face.
New to this? Step-by-step setup for the RTX 3090
Run it in the cloud
No NVIDIA RTX 3090? Run gpt-oss 20B in the cloud.
Rent a cloud GPU by the hour. No upfront cost, stop any time.
Memory needed by quantization
At 8,192 tokens of context. The black line marks the 24 GB the NVIDIA RTX 3090 can use.
| Quant | Quality | Memory | Fits? | Speed | |
|---|---|---|---|---|---|
| FP16 | Full quality | 42.8 GB | Too big | — | |
| Q8_0 | Near-lossless | 23.2 GB | Fits | ~147 tok/s | |
| Q6_K | Excellent | 18.1 GB | Fits | ~190 tok/s | |
| Q5_K_M | Very good | 15.9 GB | Fits | ~219 tok/s | |
| Q4_K_M | Good, most popular | 13.7 GB | Fits | ~257 tok/s | |
| Q3_K_M | Noticeably degraded | 11.2 GB | Fits | ~319 tok/s | |
| Q2_K | Heavily degraded | 8.8 GB | Fits | ~416 tok/s |
GPUs that run gpt-oss 20B comfortably
Frequently asked questions
Can NVIDIA RTX 3090 run gpt-oss 20B?
Yes. The NVIDIA RTX 3090 can run gpt-oss 20B fully in VRAM at Q8_0 (near-lossless), using about 23.2 GB of the 24 GB available, at roughly 147 tokens/second.
How much VRAM does gpt-oss 20B need?
About 13.7 GB at Q4_K_M with 8,192 tokens of context, and about 42.8 GB at full FP16 precision.
How fast is gpt-oss 20B on the NVIDIA RTX 3090?
Roughly 147 tokens per second at Q8_0, estimated from the NVIDIA RTX 3090's 936 GB/s memory bandwidth.