GeForce RTX 5090 · gpt-oss-120b · MXFP4
No. gpt-oss-120b at MXFP4 does not fit on GeForce RTX 5090 at 8,192 tokens, even with 32 GB of system RAM.
| Piece | Value | Note |
|---|---|---|
| Weights | 59.0 GiB | gpt-oss-120b-MXFP4.gguf |
| KV cache | 0.29 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 2.99 GiB | On the sum above |
| Total | 62.8 GiB | No |
| Fast memory | 32.0 GiB | 32 GB VRAM |
| Speed | — | Active-expert bandwidth range |
ollama run gpt-oss:120b
GeForce RTX 5090 · gpt-oss-120b ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 59.0 GiB, KV cache 0.29 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 62.8 GiB. Fast memory available is 32.0 GiB.
What if I change the quant or the context?
This page is MXFP4 at 8,192 tokens. A smaller quant or a shorter context lowers the total. The homepage calculator changes both without a new page.
Can I trust the tokens per second?
No speed is shown because the model does not load.
What do I type?
ollama run gpt-oss:120b
Is GeForce RTX 5090 assumed to have extra RAM?
Yes. Discrete GPU pages assume 32 GB of system RAM. Offload uses whatever remains after 8 GB for Windows, and it is limited by the slower of DDR5-5600 and the PCIe link.