NVIDIA · Ada Lovelace · 48 GB VRAM tier

Best AI Models for NVIDIA L40S (48 GB)

Runs 27–31B and 35B MoE models well; 70B barely fits and is slow (14 tok/s).

48 GB GDDR6864 GB/s362 TFLOPS FP16350 WAda Lovelace 2023

What LLMs can the L40S run?

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 153 t/s	153 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 184 t/s	184 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 136 t/s	136 tok/s	Q4_K_M · 22.0 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 153 t/s	153 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 184 t/s	184 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 153 t/s	153 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 184 t/s	184 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Math Score	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s		Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s		Q4_K_M · 17.9 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 136 t/s	136 tok/s		Q4_K_M · 22.0 GB
4	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	overall	Q4_K_M · 21.7 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 184 t/s	184 tok/s		Q4_K_M · 19.3 GB

#	Model	Speed ↓	Quant · VRAM
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 172 t/s	172 tok/s	Q4_K_M · 3.4 GB
2	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 184 t/s	184 tok/s	Q4_K_M · 19.3 GB
3	GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 169 t/s	169 tok/s	Q4_K_M · 13.1 GB
4	Qwen3 30B A3B Thinking 2507Alibaba Q4_K_M · 19.4 GB · 159 t/s	159 tok/s	Q4_K_M · 19.4 GB
5	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 153 t/s	153 tok/s	Q4_K_M · 16.5 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 32.2 t/s	32.2 tok/s	Q4_K_M · 17.9 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 27.5 t/s	27.5 tok/s	Q4_K_M · 21.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 153 t/s	153 tok/s	Q4_K_M · 16.5 GB
5	MiniCPM V 4.7 35B A3BOpenBMB Q4_K_M · 22.1 GB · 153 t/s	153 tok/s	Q4_K_M · 22.1 GB

See all 368 models on the L40S →Best quant that fits · speeds estimated on this card

About the L40S

The L40S runs 27–31B and 35B MoE models well. Ada Lovelace generation, released 2023.

GenerationAda Lovelacereleased 2023

L40S strengths and weaknesses

Strengths

Runs 35B MoE models: Qwen3.6 35B A3B at 136 tok/s in 22 GB
Room for long context: Qwen3.8 27B at Q4 uses 17.9 GB, leaving 30.1 GB free
Reads long prompts fast: A 4K-token prompt on Qwen3.8 27B starts answering in 1.4 s

Weaknesses

120B MoE models don't fit: GPT OSS 120B needs 70.8 GB at Q4 vs 48 GB here

L40S tokens per second by model size

Output speed depends on memory bandwidth (864 GB/s). Grey rows need more than 48 GB.

9BQwen3.5 9B · Q4_K_M · 6.6 GB
88.3tok/s
14BQwen3 14B · Q4_K_M · 11.9 GB
59.1tok/s
27–31BQwen3.8 27B · Q4_K_M · 17.9 GB
32.2tok/s
35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB
136tok/s
70BLlama 3.3 70B Instruct · IQ4_XS · 46.4 GB tight · IQ4_XS
13.5tok/s
120B MoEGPT OSS 120B · not at 4-bit
needs 70.8 GB

Estimated speeds: Q4_K_M, 4K context, one user. How we estimate

All 1216 models on the L40S

Featured only

Best quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with NVIDIA L40S — ranked by Score, recency and Fit
Model	Quant	VRAM	Speed	Context	Fit
Gemma 1.1 2B IT2.5BChat Q4_K_M·340.4 t/s	Q4_K_M	3.5 GB7%	340.4 t/s	—	S100, Runs great
Deepseek Coder 7B Instruct V1.56.9BChatCode Q4_K_M·103.0 t/s	Q4_K_M	6.5 GB13%	103.0 t/s	4K	S92, Runs great
Mistral Small Instruct 240922.2BChatFunctions Q4_K_M·39.8 t/s	Q4_K_M	17.4 GB36%	39.8 t/s	33K	A75, Runs well
Codegemma 2B2.5BChatCode Q4_K_M·340.4 t/s	Q4_K_M	3.5 GB7%	340.4 t/s	—	S100, Runs great
Meta Llama 3 8B Instruct8.0BChat Q4_K_M·104.2 t/s	Q4_K_M	6.2 GB13%	104.2 t/s	8K	S92, Runs great
Qwen2.5 Math 1.5B1.5BChatMathFunctions Q4_K_M·438.8 t/s	Q4_K_M	1.3 GB3%	438.8 t/s	4K	S100, Runs great
Baichuan2 13B Base13BChat Q4_K_M·65.5 t/s	Q4_K_M	11.5 GB24%	65.5 t/s	—	S85, Runs great
QwQ 32B Preview32.8BChatReasoning Q4_K_M·27.4 t/s	Q4_K_M	24.3 GB51%	27.4 t/s	33K	A68, Runs well
Qwen 1 8B1.8BChat Q4_K_M·464.1 t/s	Q4_K_M	1.9 GB4%	464.1 t/s	8K	S100, Runs great
Llama 2 7B HF6.7BChat Q4_K_M·103.6 t/s	Q4_K_M	6.5 GB14%	103.6 t/s	4K	S92, Runs great
SmolLM2 1.7B1.7BChat Q4_K_M·324.6 t/s	Q4_K_M	2.9 GB6%	324.6 t/s	8K	S100, Runs great
Llama 3.1 70B Instruct70.6BChatFunctions IQ4_XS·13.5 t/s	IQ4_XS	46.4 GB97%	13.5 t/s	131K	D26, Barely runs
MagicPrompt Stable Diffusion137MChat Q4_K_M·6240.0 t/s	Q4_K_M	0.1 GB0%	6240.0 t/s	1K	S100, Runs great
Yi 34B34.4BChat Q4_K_M·26.2 t/s	Q4_K_M	21.9 GB46%	26.2 t/s	4K	A67, Runs well
Baichuan 13B Base13BChat Q4_K_M·65.5 t/s	Q4_K_M	11.5 GB24%	65.5 t/s	—	S85, Runs great

Rent NVIDIA L40S in the cloud

Hourly rate on demand for one L40S.

Vast.aiMarketplace
from$0.77/hr
Deploy

Cost per million tokens

Output-token cost on an L40S rented at $0.77/hr, one request at a time. Batch several users on the same card and the cost per token drops several-fold.

Model	Quant	Speed	Per 1M output
NVIDIA Nemotron 3.5 Lightning 30B A3B BF16	Q4_K_M	184 tok/s	$1.17$0.0062 input
Gemma 4 26B A4B IT	Q4_K_M	153 tok/s	$1.41$0.012 input
Qwen3.6 35B A3B	Q4_K_M	136 tok/s	$1.58$0.013 input
Qwen3.8 27B	Q4_K_M	32 tok/s	$6.68$0.073 input
Muse Glimmer 30B	Q4_K_M	31 tok/s	$7.03$0.079 input

L40S specs, power and PSU

Specs

Architecture: Ada Lovelace
VRAM: 48 GB GDDR6
Memory bandwidth: 864.0 GB/s
FP16 (dense): 362.05 TFLOPS
CUDA cores: 18,176
Tensor cores: 568
Compute capability: 8.9
TDP: 350 W
Released: 2023-08-08

L40S FAQ

What is the largest model the L40S can run?

Qwen2.5 72B Instruct (72.7B) is the largest at 4-bit or better: at Q4_K_S it uses 46.6 GB of 48 GB, and runs at about 13.4 tok/s.

Can the L40S run Qwen3.8 27B?

Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 48 GB, and it is the best model that runs well on this card.

Can the L40S run 120B MoE models?

No. GPT OSS 120B needs 70.8 GB at Q4, more than the 48 GB on this card.

How many tokens per second does the L40S give?

About 32.2 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 1.4 s on a 4K-token prompt.

What power supply does the L40S need?

Plan a 700 W PSU (2× the 350 W TDP).

Is 48 GB enough for coding and agents?

Yes for coding: Qwen3.8 27B, the best of the 10 coding models that fit, runs at about 32 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.

Best coding LLMs 48 GB VRAM GPUs with 48 GB or more