Written by Jakub Rusinowski · Last updated August 15, 2026
Yes — comfortably
Yes, comfortably — GPT-OSS 20B at Q8_0 needs about 22.5 GB of the 32 GB usable on RTX 5090 Desktop (32 GB VRAM, 64 GB RAM), leaving ~9.5 GB spare and running at ~54.9 tok/s (estimated), with room for about 131,072 tokens of context.
Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~54.9 tok/s
| Usable memory for models | 32 GB |
| Memory bandwidth | 1792 GB/s |
| Form factor | Desktop |
| Operating system | Windows or Linux |
| Memory upgradeable | Yes |
| Quant | Memory needed | Fits 32 GB? | Max context | Est. speed | Download |
|---|---|---|---|---|---|
| F16 | 41.2 GB | ✗ No | — | — | 40 GB |
| Q8_0 | 22.5 GB | ✓ Yes | 128K | ~54.9 tok/s | 21.3 GB |
| Q6_K | 17.6 GB | ✓ Yes | 128K | ~68.5 tok/s | 16.4 GB |
| Q5_K_M | 15.4 GB | ✓ Yes | 128K | ~77.3 tok/s | 14.2 GB |
| Q4_K_M | 13.3 GB | ✓ Yes | 128K | ~87.9 tok/s | 12.1 GB |
| Q3_K_M | 9.7 GB | ✓ Yes | 128K | ~114.4 tok/s | 8.5 GB |
| Q2_K | 7.8 GB | ✓ Yes | 128K | ~137.2 tok/s | 6.6 GB |
| Variant | Needs at Q4_K_M | Fits? | Est. speed |
|---|---|---|---|
| GPT-oss 120B | 73.9 GB | ✗ Too large | — |
| GPT-OSS 20B | 13.3 GB | ✓ Fits | ~87.9 tok/s |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes, comfortably — GPT-OSS 20B at Q8_0 needs about 22.5 GB of the 32 GB usable on RTX 5090 Desktop (32 GB VRAM, 64 GB RAM), leaving ~9.5 GB spare and running at ~54.9 tok/s (estimated), with room for about 131,072 tokens of context.
Q8_0 — it needs about 22.5 GB of the 32 GB available, downloads as roughly 21.3 GB, and runs at an estimated 54.9 tokens/sec with up to 128K of context.
Nothing binding — the model fits with headroom and generates at a usable speed on this hardware.
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving