A100 80GB vs L40S: Which GPU Should You Choose?

This page settles when A100 80GB or L40S is the better choice using current provider pricing and manufacturer specifications.

Updated 2026-08-23 · refreshed daily

Choose L40S for workloads within 48GB at an on-demand minimum of $0.80 versus $0.60, and choose A100 80GB when its 80GB capacity avoids extra GPUs, except for FP8.

Run the same small representative job once on each GPU at the live table’s on-demand minimums of $0.60 and $0.80, then compare the cost per completed job.

Does your workload require native FP8 hardware support? If yes, choose L40S: it supports FP8, BF16, or FP32 workloads within 48GB VRAM, while A100 80GB is suited to BF16 or FP32 workloads within 80GB VRAM.

Is finishing sooner worth the difference in per-job cost? Choose L40S when the deadline matters; A100 80GB has an effective cost-per-job ratio of 0.87 relative to it.

Can your workload tolerate interruptions? Start with a short test rental on the exact listing before moving the full workload.

Today's prices

USD/hr · 1× GPU · lowest on-demand price per provider

Provider	A100 80GB	L40S
Vast.ai	$0.60	$0.80
Hyperstack	$1.35 · Not reported	Not offered
RunPod	$1.39	$0.99
Jarvislabs	$1.49 · Not reported	Not offered
DataCrunch	$1.79	$1.37
Crusoe	$2.00 · Not reported	$1.50 · Not reported

Prices refresh daily. Marlin always books the cheapest available. Try Marlin →

0.75×

A100 80GB’s on-demand price premium is small enough for memory fit to decide.

0.87×

L40S delivers better BF16 throughput value at on-demand rates.

$438/mo

Baseline monthly budget for continuously running one A100 80GB at its on-demand minimum.

What your workload needs

Find your parameter count on the model's Hugging Face card. Serving maps to the inference rows, training to the fine-tuning rows.

Workload	VRAM needed	A100 80GB	L40S	Cheapest today
70B Q4 inference	42 GB	1× $0.60/hr	1× $0.80/hr	A100 80GB ×1 → $0.6/hr
70B FP16 inference	168 GB	3× $0.60/hr	4× $0.80/hr	A100 80GB ×3 → $1.8/hr (est.)
70B QLoRA	46 GB	1× $0.60/hr	1× $0.80/hr	A100 80GB ×1 → $0.6/hr
7B full fine-tune	112 GB	2× $0.60/hr	3× $0.80/hr	A100 80GB ×2 → $1.2/hr (est.)
70B FP8 training	154 GB	unsupported	4× $0.80/hr	L40S ×4 → $3.2/hr (est.)
Downshift recommendation	42 GB	1× $0.60/hr	1× $0.80/hr	A100 80GB ×1 → $0.6/hr

What the numbers say

A100 80GB has an on-demand price ratio of 0.75 relative to L40S, so choose L40S when both fit the workload and choose A100 80GB when its larger memory avoids extra GPUs.

L40S is favored by the datasheet-based comparison: A100 80GB has a performance ratio of 0.86 and an on-demand effective cost-per-job ratio of 0.87 relative to it. Treat these as directional estimates from vendor peak specifications, not measured workload throughput.

A100 80GB is suited to BF16 or FP32 workloads within 80GB VRAM, while L40S supports FP8, BF16, or FP32 workloads within 48GB VRAM. Choose L40S when FP8 is required or the workload fits its single-GPU limit; choose A100 80GB when its larger memory avoids extra GPUs after L40S reaches its memory ceiling.

Before you rent

Data egress fees are omitted from headline GPU rates; verify them on the provider’s network pricing page before renting.

Exceeding one GPU’s VRAM forces a multi-GPU configuration, increasing the number of billed accelerators.

Interruptible spot rentals can stop mid-job, causing non-checkpointed work to restart and consume additional GPU time.

Billing granularity can make short experiments cost more than their runtime suggests; each provider’s pricing page lists the minimum charge that determines the billed amount.

Marlin matches your workload requirements to the lowest-priced suitable GPU across supported CSP and GPU-cloud providers.

Coverage is limited to the listed provider pricing and manufacturer specifications, with freshness shown separately; rates and availability can change. Workload costs and performance comparisons are modeled from VRAM assumptions, listed prices, and vendor peak specifications rather than measured end-to-end benchmarks. https://vast.ai/pricing · https://www.paperspace.com/pricing · https://datacrunch.io/pricing · https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf · https://resources.nvidia.com/en-us-l40s/l40s-datasheet-28413 · https://www.runpod.io/pricing · https://lambda.ai/service/gpu-cloud · https://www.hyperstack.cloud/gpu-pricing · https://jarvislabs.ai/pricing

Stop comparing. Start running.

Marlin matches your workload to the lowest-priced suitable GPU across supported providers.

Get started with Marlin