Question 1

What AI models can run on L40S?

Accepted Answer

The L40S (48 GB VRAM) can run models including Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8, Llama-3.2-1B-Instruct-QLORA_INT4_EO8, Llama-3.2-3B-Instruct-QLORA_INT4_EO8, Llama-2-7b, Llama-3.2-3B-Instruct-SpinQuant_INT4_EO8 and more — 10 compatible models indexed so far. Compatible models are determined by whether the model's memory footprint fits within the GPU's 48 GB VRAM, potentially using quantization (Q8, Q6, or Q4) to reduce memory usage.

Question 2

What is the cheapest cloud option for L40S?

Accepted Answer

The cheapest on-demand option for the L40S is Vast.ai at $0.60/hr (1x-L40S-32-65536). Spot/preemptible pricing is available at $0.00/hr. We track pricing across 78 providers to help you find the best deal.

Question 3

Is L40S good for fine-tuning?

Accepted Answer

Yes, the L40S is well-suited for fine-tuning. With 48 GB of VRAM, it can handle LoRA/QLoRA fine-tuning of models up to ~30B parameters and full fine-tuning of smaller models (up to ~7B parameters). For larger models, consider multi-GPU setups.

Question 4

How much VRAM does L40S have?

Accepted Answer

The L40S has 48 GB of VRAM (GDDR6). It also offers 864 GB/s memory bandwidth, 366.1 TFLOPS of FP16 compute, FP8 support for efficient inference.

Question 5

What quantization formats work on L40S?

Accepted Answer

The L40S supports the following quantization formats: FP16 (full precision), FP8, Q8 (8-bit), Q6 (6-bit), Q4 (4-bit). FP16 works for models up to ~7B parameters. FP8 halves memory vs FP16 with minimal quality loss. Q4 quantization lets models up to ~163B parameters (by VRAM estimate) fit on the L40S.

Provider	On-demand	Spot	Actions
Vast.ai	$0.600/hr	$0.000/hr	Rent
RunPod	$0.990/hr	$0.860/hr	Rent Get $5–$500 free
Nebius	$1.548/hr	$0.749/hr	—
Amazon Web Services	$1.861/hr	$0.535/hr	—

L40S

Specifications

Cloud pricing

Compatible models

Compare this GPU

Frequently asked questions

VRAM	48 GB
Memory type	GDDR6
Bandwidth	864 GB/s
FP16 TFLOPS	366.1
FP8 support	Yes