Can L40S 48GB run gemma-4-31B-it?

NVIDIA · 48 GB GDDR6 · 864 GB/s bandwidth · Google · 30.7B · model ctx up to 256K

Yes — the L40S 48GB (48 GB GDDR6) can run gemma-4-31B-it. At Q4 with the default 8K context it needs ≈ 24.03 GB VRAM (weights 16.89 GB + KV cache 6.14 GB + 1 GB runtime overhead), which fits within 48 GB. Decode at Q4/8K: ≈ 17.9 tok/s (estimated, batch 1). FP16 needs ≈ 68.54 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	20.96 GB	Yes	17.9 est.
Q4	8K default	24.03 GB	Yes	17.9 est.
Q4	32K	42.45 GB	Yes	17.9 est.
FP16	4K	65.47 GB	No	7.6 est.
FP16	8K default	68.54 GB	No	6.9 est.
FP16	32K	86.96 GB	No	4.3 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	16.89 GB	6.14 GB	1 GB	24.03 GB	48 GB
FP16	61.4 GB	6.14 GB	1 GB	68.54 GB	48 GB

Other GPUs that run gemma-4-31B-it

• B200 192GB SXM — ≈165.8 tok/s Q4 est.
• MI300X 192GB — ≈109.9 tok/s Q4 est.
• H200 141GB SXM5 — ≈99.5 tok/s Q4 est.
• H100 80GB SXM5 — ≈69.4 tok/s Q4 est.
• Gaudi 3 128GB — ≈67.9 tok/s Q4 est.
• A100 80GB SXM4 — ≈42.3 tok/s Q4 est.
• RTX 5090 32GB — ≈37.1 tok/s Q4 est.
• RTX 6000 Ada 48GB — ≈19.9 tok/s Q4 est.

All GPUs for gemma-4-31B-it →

Other models for the L40S 48GB

• MiniCPM5-1B — ≈509.1 tok/s Q4 est.
• Qwen3.5-2B — ≈274.9 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.
• Qwen3-Coder-Next — ≈183.3 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈183.3 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈183.3 tok/s Q4 est.
• Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 — ≈183.3 tok/s Q4 est.

All models on the L40S 48GB →