How much memory does gpt-oss-120b need at Q4?

The Q4 figure used here is 63.1 GB. GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.

What is the cheapest catalog device that runs gpt-oss-120b at Q4?

Framework Desktop 128GB, list price $3,449.

How fast is gpt-oss-120b on the GeForce RTX 5090, as a bandwidth estimate?

About 654 tok/s at Q4. That is device bandwidth divided by active weight bytes, not a measured run.

No. gpt-oss-120b does not fit on the GeForce RTX 5090 at Q4. The estimate is 63.1 GB and usable memory is 30.4 GB.

Can the GeForce RTX 5090 run gpt-oss-120b?

Data updated 16 October 2026. Last reviewed the same day, when this catalog snapshot was written.

Usable memory on this 32 GB VRAM config is 30.4 GB (95% of the sticker figure). A result under 2 GB of headroom is labeled tight.

Quant ladder

Estimated memory for gpt-oss-120b on GeForce RTX 5090
Quant	Estimate	Headroom	Est. tok/s	Verdict
Q4	63.1 GB	-32.7 GB	654 tok/s	No
Q8	63.7 GB	-33.3 GB	648 tok/s	No
FP16	280 GB	-250.0 GB	176 tok/s	No

GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.
Formula estimate: parameter count × bytes per weight × 1.20. No GGUF file size is stored for this quant.
Tokens/sec is a bandwidth roofline: GB/s divided by active weight GB. It is not a measured generation speed.

KV cache at 8K and 32K

FP16 KV-cache estimate: 0.3 GB at 8,192 tokens and 1.2 GB at 32,768 tokens. GQA counts two tensors (K and V) for the stored KV heads and head dimension. config.json also sets a sliding window of 128 tokens. The figures above assume every counted layer stores the full context, so they can overstate cache on sliding layers.

Offload

About 32.4 GB of the Q4 weights sit outside usable VRAM. A partial offload would need at least that much free system RAM. This catalog does not record system RAM, so offload is not counted as a fit.

Ollama

This command pulls the Q4_K_M file from unsloth/gpt-oss-120b-GGUF, the Hub repo whose file size is used above. It is not an official Ollama library tag.

ollama run hf.co/unsloth/gpt-oss-120b-GGUF:Q4_K_M

What fits instead

Cheapest catalog device at Q4: Framework Desktop 128GB
Closest catalog model that still fits at Q4: Mixtral 8x7B.

All hardware for gpt-oss-120b · All models for the GeForce RTX 5090

Questions

How much memory does gpt-oss-120b need at Q4?: The Q4 figure used here is 63.1 GB. GGUF file size plus an FP16 KV-cache estimate at 8,192 tokens. The cache counts only the layers and heads named in the method on the About page.
Can the GeForce RTX 5090 run gpt-oss-120b?: No. gpt-oss-120b does not fit on the GeForce RTX 5090 at Q4. The estimate is 63.1 GB and usable memory is 30.4 GB.
What is the cheapest catalog device that runs gpt-oss-120b at Q4?: Framework Desktop 128GB, list price $3,449.
How fast is gpt-oss-120b on the GeForce RTX 5090, as a bandwidth estimate?: About 654 tok/s at Q4. That is device bandwidth divided by active weight bytes, not a measured run.