Can L40S 48GB run gemma-3-4b-it?

NVIDIA · 48 GB GDDR6 · 864 GB/s bandwidth · Google · 4.3B · model ctx up to 128K

Yes — the L40S 48GB (48 GB GDDR6) can run gemma-3-4b-it. At Q4 with the default 8K context it needs ≈ 4.15 GB VRAM (weights 2.37 GB + KV cache 0.78 GB + 1 GB runtime overhead), which fits within 48 GB. Decode at Q4/8K: ≈ 141.0 tok/s (estimated, batch 1). FP16 (≈ 10.38 GB) also fits, at ≈ 38.8 tok/s (estimated). 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	3.76 GB	Yes	141.0 est.
Q4	8K default	4.15 GB	Yes	141.0 est.
Q4	32K	6.49 GB	Yes	141.0 est.
FP16	4K	9.99 GB	Yes	38.8 est.
FP16	8K default	10.38 GB	Yes	38.8 est.
FP16	32K	12.72 GB	Yes	38.8 est.

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	L40S 48GB VRAM
Q4	2.37 GB	0.78 GB	1 GB	4.15 GB	48 GB
FP16	8.6 GB	0.78 GB	1 GB	10.38 GB	48 GB

Other GPUs that run gemma-3-4b-it

• B200 192GB SXM — ≈1305.4 tok/s Q4 est.
• MI300X 192GB — ≈864.8 tok/s Q4 est.
• H200 141GB SXM5 — ≈783.2 tok/s Q4 est.
• H100 80GB SXM5 — ≈546.6 tok/s Q4 est.
• Gaudi 3 128GB — ≈534.5 tok/s Q4 est.
• A100 80GB SXM4 — ≈332.7 tok/s Q4 est.
• RTX 5090 32GB — ≈292.4 tok/s Q4 est.
• RTX 4090 24GB — ≈164.5 tok/s Q4 est.

All GPUs for gemma-3-4b-it →

Other models for the L40S 48GB

• Qwen3-0.6B — ≈687.3 tok/s Q4 est.
• gemma-3-1b-it — ≈549.8 tok/s Q4 est.
• Llama-3.2-1B-Instruct — ≈458.2 tok/s Q4 est.
• Qwen3-1.7B — ≈274.9 tok/s Q4 est.
• Qwen3-30B-A3B — ≈183.3 tok/s Q4 est.
• Qwen3-Coder-30B-A3B-Instruct — ≈183.3 tok/s Q4 est.
• GLM-4.7-Flash — ≈183.3 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈183.3 tok/s Q4 est.

All models on the L40S 48GB →