Best home rig for gpt-oss 120B

Will it fit my GPU? Best value GPU per euro →

Real, measured generation speeds for gpt-oss 120B on home hardware, fastest first. Different quants/engines aren't directly comparable.

Build	Quant	Engine	tok/s	Cost
RTX Pro 6000 96GB workstation card	MXFP4	llama.cpp	196.31		source
Mac Studio M2 Ultra 76-core GPU, 192 GB	MXFP4	llama.cpp	79.68		source
Minisforum MS-S1 MAX (Ryzen AI Max+ PRO 495, 192 GB): gpt-oss 120B at 55.8 tok/s	MXFP4	llama.cpp Vulkan	55.8	$7,399 (≈€6,602)	source
Minisforum MS-S1 MAX (Ryzen AI Max+ PRO 495, 192 GB): gpt-oss 120B at 55.8 tok/s	MXFP4, 32K context	llama.cpp Vulkan	45.1	$7,399 (≈€6,602)	source
HP ZBook Ultra G1a (Strix Halo laptop, 128 GB): Qwen3.6 35B-A3B at 59.4 tok/s, gpt-oss 120B at 44.7	MXFP4	llama.cpp b11213 Vulkan	44.7		source
DGX Spark (GB10, 128 GB): 23 models on vLLM at 128K context, Qwen3.6 35B-A3B up to 102 tok/s		vLLM 0.31.0, 128K ctx, 1 user	44.4		source

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Compare: DGX Spark vs Mac Studio vs Strix Halo: measured tok/s and tok/s per watt

Memory math: how much VRAM / RAM does gpt-oss 120B need?

gpt-oss 120B has 117B parameters, about 5.1B active per token (source: OpenAI model card: 117B total, 5.1B active). Weight size = parameters × bits-per-weight ÷ 8, before context (KV cache) and runtime overhead.

Format	Bits/weight	Weights	Smallest common memory size that holds the weights alone
Native release format	–	≈65 GB	96 GB
Q8_0	8.50	124.3 GB	128 GB
Q6_K	6.56	96.0 GB	96 GB
Q5_K_M	5.70	83.4 GB	96 GB
Q4_K_M	4.89	71.6 GB	96 GB
Q3_K_M	4.00	58.4 GB	64 GB
Q2_K	3.16	46.2 GB	48 GB

Bits-per-weight for GGUF quants are llama.cpp's own figures (llama.cpp quantize README). "As measured" rows apply the source's stated bits-per-weight to the parameter count; real files differ slightly because some layers stay at higher precision. Try other combinations in the VRAM calculator.

Why it can still run on smaller GPUs: it is a mixture-of-experts model, so only about 5.1B of its 117B parameters are used for each generated token. Runtimes such as llama.cpp (--n-cpu-moe), ExLlamaV3 or Project Maya can keep most expert weights in system RAM and only the shared layers and hot experts in VRAM. Then you need VRAM + system RAM ≥ weights, and speed is limited mainly by RAM bandwidth.

FAQ

How much VRAM does gpt-oss 120B need?

The weights are about 65 GB in its native format (OpenAI model card: 117B total, 5.1B active). Add memory for context (KV cache) and the runtime. The smallest-memory setup with a measured result here, fully in memory, is RTX Pro 6000 (Blackwell, 96 GB) (96 GB) at 196.31 tok/s.

What is the fastest measured hardware for gpt-oss 120B?

RTX Pro 6000 (Blackwell, 96 GB): 196.31 tok/s generation (llama.cpp, MXFP4), from https://github.com/ggml-org/llama.cpp/discussions/15396.

Which GPUs have measured results for gpt-oss 120B?

Apple M2 Ultra, DGX Spark (GB10), RTX Pro 6000 (Blackwell, 96 GB), Strix Halo (Radeon 8060S).

Can it run on… Apple M2 Ultra · DGX Spark (GB10) · RTX Pro 6000 (Blackwell, 96 GB) · Strix Halo (Radeon 8060S)