Can T4 16GB run Llama-4-Scout-17B-16E-Instruct?

NVIDIA · 16 GB GDDR6 · 320 GB/s bandwidth · Meta · 109.0B (active 17.0B) · MoE · model ctx up to 10240K

No — Llama-4-Scout-17B-16E-Instruct at Q4 needs ≈ 62.65 GB even at 4K context, beyond the T4 16GB's 16 GB GDDR6. It would only run with heavy CPU/disk offload. FP16 needs ≈ 222.4 GB — does not fit on this card. Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.

Fit & speed by quant and context

Quant	Context	VRAM needed	Fits?	tok/s (decode)
Q4	4K	62.65 GB	No	2.2 est.
Q4	8K default	64.35 GB	No	—
Q4	32K	74.55 GB	No	—
FP16	4K	220.7 GB	No	—
FP16	8K default	222.4 GB	No	—
FP16	32K	232.6 GB	No	—

Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s means the model is far beyond this card (offload-only territory).

VRAM breakdown at 8K context

Quant	Weights	KV cache	Overhead	Total needed	T4 16GB VRAM
Q4	59.95 GB	3.4 GB	1 GB	64.35 GB	16 GB
FP16	218.0 GB	3.4 GB	1 GB	222.4 GB	16 GB

Other GPUs that run Llama-4-Scout-17B-16E-Instruct

• B200 192GB SXM — ≈299.5 tok/s Q4 est.
• MI300X 192GB — ≈198.4 tok/s Q4 est.
• H200 141GB SXM5 — ≈179.7 tok/s Q4 est.
• H100 80GB SXM5 — ≈125.4 tok/s Q4 est.
• Gaudi 3 128GB — ≈122.6 tok/s Q4 est.
• A100 80GB SXM4 — ≈76.3 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — ≈30.7 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — ≈29.9 tok/s Q4 est.

All GPUs for Llama-4-Scout-17B-16E-Instruct →

Other models for the T4 16GB

• MiniCPM5-1B — ≈188.6 tok/s Q4 est.
• Qwen3.5-2B — ≈101.8 tok/s Q4 est.
• Llama-3.2-3B-Instruct — ≈63.6 tok/s Q4 est.
• gpt-oss-20b — ≈56.6 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — ≈50.9 tok/s Q4 est.
• Qwen3-4B-Thinking-2507 — ≈50.9 tok/s Q4 est.
• NVIDIA-Nemotron-3-Nano-4B-BF16 — ≈50.9 tok/s Q4 est.
• gemma-4-E2B-it — ≈39.9 tok/s Q4 est.

All models on the T4 16GB →