Skip to content
Run LLM Locally

gpt-oss-120b on the H100 PCIe

With 80 GB of VRAM, the H100 PCIe runs gpt-oss-120b at MXFP4 entirely on the GPU with an 8K context (about 63.2 GB needed).

ShareFacebookXWhatsAppLinkedInEmail

gpt-oss-120b

MXFP4 · 8K context

Excellent

The model needs about 63.2 GB and fits in the 77.6 GB available, with 14.4 GB to spare.

Your hardware
H100 PCIe 80 GB · 64 GB RAM
Estimated memory
~63.2 GB estimate
VRAM (usable memory)
77.6 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
175–488 tok/s estimate
Largest context in fast memory
128K tokens
Confidence: medium

Based on: same model · same quantization · other hardware, scaled by memory bandwidth · 21 benchmark(s)

Memory breakdown

~63.2 GB (61.6 GB–64.8 GB) needed · estimate

Fits entirely in VRAM · 14.4 GB left

  • Model weights59 GB
  • KV cache0.3 GB
  • Runtime overhead0.9 GB
  • Safety margin3 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
  • scaled from other hardware

Other quantizations on the H100 PCIe

gpt-oss-120b on H100 PCIe + 64 GB RAM · 8K context
gpt-oss-120b MXFP48K~63.2 GB · 175–488 tok/s estimateExcellent

Alternatives

GPUs that run gpt-oss-120b MXFP4 entirely in VRAM

Other models that fit the H100 PCIe

Hardware that runs gpt-oss-120b MXFP4 entirely in memory

Prices and links follow the market you choose.

Radeon PRO W7900

AMD

ExcellentRuns gpt-oss-120b MXFP4 entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

NVIDIA DGX Spark (GB10)

Unified memory PC

ExcellentRuns gpt-oss-120b 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 H100 PCIe run gpt-oss-120b?
The model needs about 63.2 GB and fits in the 77.6 GB available, with 14.4 GB to spare.
Which quantization should I use on the H100 PCIe?
MXFP4 is the highest-precision quantization that fits entirely in VRAM at 8K context. ~63.2 GB / 77.6 GB.
What is the largest context on the H100 PCIe?
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-120b (H100 PCIe)?
We estimate 175–488 tok/s (estimate, confidence medium). Based on: same model, same quantization, other hardware, scaled by memory bandwidth, 21 benchmark(s).