Can a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB) run Llama 4 Scout (MoE)?

Yes — Llama 4 Scout (MoE) runs at FP16 (215.6 GB) at about 17–26 tok/s on a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB).

How much memory does Llama 4 Scout (MoE) need?

Llama 4 Scout (MoE) needs about 41.2 GB at 4K context at its smallest quantization. A NVIDIA GH200 Grace Hopper (144GB HBM + 480GB) has 144.0 GB.

Yes — Llama 4 Scout (MoE) runs on a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB) at FP16

Yes — it runs great.

Runs great

Best fit: Llama 4 Scout (MoE) at FP16 · 215.6 GB weights · 17–26 tok/s on a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB).

Arena Text Overall · style-controlled

1,322FP16 untested

llama-4-scout-17b-16e-instruct · published configuration reference · FP16 not tested

Arena Text Overall (style-controlled), exact published entry llama-4-scout-17b-16e-instruct, 2026-10-08: reference rating 1,321.5. Published confidence bounds 1,316.8–1,326.3. 29,836 votes. Published llama-4-scout-17b-16e-instruct configuration reference. Reasoning mode is not disclosed by the Arena dataset. Tested precision, runtime and sampling settings are not disclosed in this dataset. Selected FP16 has not been measured by this Arena entry. No quantization quality adjustment is applied.

Snapshot published 2026-10-08 · snapshot imported 2026-10-11 UTC · validated last-good data · manual refresh

Arena leaderboard dataset · Arena / lmarena-ai, Arena Leaderboard Dataset (CC BY 4.0). Selected Text Overall style-controlled rows; field names adapted, ratings unchanged. · dataset CC BY 4.0 · revision e16f90232986

Total memory at 4K context: 217.2 GB; Weights + KV cache + runtime overhead
Estimated generation: 17–26 tok/s
Estimated first token: ~0.8s

Every quantization on this machine

On small screens, each quant lists every metric below its name. On wider screens, scroll sideways for total memory, speed and fit. Keyboard: focus the table, then use the arrow keys. Quant names stay visible.

Llama 4 Scout (MoE) quantizations on NVIDIA GH200 Grace Hopper (144GB HBM + 480GB), estimated at 4K context. Weights and total memory are shown separately.
Quant	Weights	Total memory	Speed	Fit
Q2_K	39.6 GB	41.2 GB	214–321 tok/s	Runs great
Q3_K_M	51.8 GB	53.4 GB	192–288 tok/s	Runs great
Q4_K_M	65.4 GB	67.0 GB	173–259 tok/s	Runs great
Q5_K_M	76.6 GB	78.2 GB	159–239 tok/s	Runs great
Q6_K	88.4 GB	90.0 GB	147–221 tok/s	Runs great
Q8_0	114.5 GB	116.1 GB	126–189 tok/s	Runs great
FP16	215.6 GB	217.2 GB	17–26 tok/s	Runs great

Open the full calculator →Everything a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB) runs Llama 4 Scout (MoE) on other machines What to buy to run it →

Ran Llama 4 Scout (MoE) on a NVIDIA GH200 Grace Hopper (144GB HBM + 480GB)? Add your real numbers →