Skip to content
Run LLM Locally

gpt-oss-20b on the RTX 5090

With 32 GB of VRAM, the RTX 5090 runs gpt-oss-20b at MXFP4 entirely on the GPU with an 8K context (about 13 GB needed).

ShareFacebookXWhatsAppLinkedInEmail

gpt-oss-20b

MXFP4 · 8K context

Excellent

The model needs about 13 GB and fits in the 31 GB available, with 18.1 GB to spare.

Your hardware
RTX 5090 32 GB · 64 GB RAM
Estimated memory
~13 GB estimate
VRAM (usable memory)
31 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
264–291 tok/s
Largest context in fast memory
128K tokens
Confidence: medium

Based on: same hardware · same model · same quantization · 3 benchmark(s)

Memory breakdown

~13 GB (12.4 GB–13.6 GB) needed · estimate

Fits entirely in VRAM · 18.1 GB left

  • Model weights11.3 GB
  • KV cache0.2 GB
  • Runtime overhead0.9 GB
  • Safety margin0.6 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size · sliding-window layers keep 128 tokens

EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • mixture-of-experts: speed depends on expert routing

Other quantizations on the RTX 5090

gpt-oss-20b on RTX 5090 + 64 GB RAM · 8K context
gpt-oss-20b MXFP48K~13 GB · 264–291 tok/s measuredExcellent

Alternatives

GPUs that run gpt-oss-20b MXFP4 entirely in VRAM

Other models that fit the RTX 5090

Where to buy a RTX 5090

Prices and links follow the market you choose.

GeForce RTX 5090

32 GB GDDR7 · 1,792 GB/s · 575 W

ExcellentRuns gpt-oss-20b MXFP4 entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the RTX 5090 run gpt-oss-20b?
The model needs about 13 GB and fits in the 31 GB available, with 18.1 GB to spare.
Which quantization should I use on the RTX 5090?
MXFP4 is the highest-precision quantization that fits entirely in VRAM at 8K context. ~13 GB / 31 GB.
What is the largest context on the RTX 5090?
At MXFP4, about 128K tokens fit entirely in VRAM with an f16 KV cache. A q8_0 KV cache roughly doubles that.
How fast is gpt-oss-20b (RTX 5090)?
We estimate 264–291 tok/s (measured, confidence medium). Based on: same hardware, same model, same quantization, 3 benchmark(s).