Can GPT-OSS 120B run on NVIDIA H800 80GB?

Q: What should I upgrade first if GPT-OSS 120B feels slow on NVIDIA H800 80GB?

Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

YES — With Offload

A80Great

Estimated from fit model

GPT-OSS 120B needs ~85.2 GB VRAM. NVIDIA H800 80GB has 80.0 GB. With Q4_K_M quantization, expect ~26 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: HighStack: StandardBottleneck: Host offload

Operating mode

Choose the run profile you care about

Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 85.2 GB, 28.5 tok/s, Runs with offload (needs ~4.3 GB host RAM)

85.2 GB required80.0 GB available

107% VRAM needed

5.2 GB over capacity — needs offload or smaller quantization

Fit status

Runs with offload (needs ~4.3 GB host RAM)

Decode

28.5 tok/s

TTFT

6795 ms

Safe context

Memory

85.2 GB / 80.0 GB

Offload

10%

Memory breakdown

Weights71.4 GB

KV Cache4.9 GB

Runtime0.9 GB

Headroom8.0 GB

See how fast it feels

See how fast it feelsGPT-OSS 120B on NVIDIA H800 80GB
1st promptCold start — includes initialization
>Write a Python function to sort a list
Here's a Python function that sorts a list using quicksort:

def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

# Example usage
numbers = [3, 6, 8, 10, 1, 2, 1]
print(quicksort(numbers))  # [1, 1, 2, 3, 6, 8, 10]

The algorithm picks a pivot, partitions the list into smaller, equal, and larger elements, then recursively sorts each partition and concatenates the results.
2nd promptWarm — model ready, no init overhead
>Now add type hints and a docstring
def quicksort(arr: list[int]) -> list[int]:
    """Sort a list using the quicksort algorithm.

    Args:
        arr: List of integers to sort.

    Returns:
        A new sorted list.
    """
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

Added type hints for the input and return type, plus a Google-style docstring.
Estimated: 28.5 tok/s decode · 6.8s TTFT (warm) · 71 tok/s prefill

What limits this setup

It fits through host-memory offload, and offload is the main reason performance drops.

CPU or host-memory offload is active

About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Best improvement path

Remove offload with more accelerator memory

Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Increase host RAM if you keep offloading

This setup may need roughly {ram} GB of extra host RAM just for the offloaded portion, before OS and other tools.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	S	Runs with offload	27.5 tok/s	3842 ms	4K
Coding	A	Runs with offload	26.2 tok/s	7390 ms	4K
Agentic Coding	A	Very compromised	23.9 tok/s	11784 ms	4K
Reasoning	A	Runs with offload	26.2 tok/s	8733 ms	4K
RAG	A	Very compromised	23.9 tok/s	14731 ms	4K

Quantization options

How GPT-OSS 120B (117B params) fits at each quantization level on NVIDIA H800 80GB (80.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	45.6 GB	Low	S88
Q3_K_SBest for your GPU	3	57.3 GB	Low	S88

Get started

Copy-paste commands to run GPT-OSS 120B on your machine.

Run

ollama run gpt-oss:120b

Your hardware

More models your NVIDIA H800 80GB can run

Model	Params	Grade	Decode	Capabilities
Devstral 2 123B Instruct	123B	A	25.1 tok/s
Qwen 3.5 122B A10B	122B	S

Frequently asked questions

Can NVIDIA H800 80GB run GPT-OSS 120B?

Yes, NVIDIA H800 80GB can run GPT-OSS 120B with a A grade (Runs with offload). Expected decode speed: 26.2 tok/s.

How much VRAM does GPT-OSS 120B need?

GPT-OSS 120B (117B parameters) requires approximately 85.2 GB of memory with Q4_K_M quantization.

What is the best quantization for GPT-OSS 120B?

The recommended quantization for GPT-OSS 120B is Q4_K_M, which balances quality and memory efficiency.

What speed will GPT-OSS 120B run at on NVIDIA H800 80GB?

On NVIDIA H800 80GB, GPT-OSS 120B achieves approximately 26.2 tokens per second decode speed with a time-to-first-token of 7390ms using Q4_K_M quantization.

Can NVIDIA H800 80GB run GPT-OSS 120B for coding?

For coding workloads, GPT-OSS 120B on NVIDIA H800 80GB receives a A grade with 26.2 tok/s and 4K context.

What context window can GPT-OSS 120B use on NVIDIA H800 80GB?

On NVIDIA H800 80GB, GPT-OSS 120B can safely use up to 4K tokens of context. The model's official context limit is 131K, but available memory constrains the safe maximum.

What should I upgrade first if GPT-OSS 120B feels slow on NVIDIA H800 80GB?

See all results for NVIDIA H800 80GB See all hardware for GPT-OSS 120B

Embed this result▼

Paste this snippet into any page to show a live fit card.

<iframe src="https://willitrunai.com/embed/gpt-oss-120b-on-h800-80gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: