RunPodUS-EastOn-demand
$0.79/ GPU / hr
Open provider
Cloud GPU · verified 2026-09-05
NVIDIA L40S
Ada Lovelace48 GB
Ada inference board: 48 GB GDDR6 ECC, 864 GB/s, no NVLink. The grown-up 4090 for people who need ECC and a datacenter slot. Not a trainer with a fabric.
VRAM48 GB
From /GPU/hr$0.79
ArchitectureAda Lovelace
VendorNVIDIA
Sponsored
Featured L40S offers
The four cheapest in-stock on-demand L40S hours in this sample. Each card is a sponsored placement.
Vast.aiUS-EastOn-demand
$0.85/ GPU / hr
Open provider
FluidstackEU-WestOn-demand
$0.89/ GPU / hr
Open provider
LambdaUS-WestOn-demand
$0.99/ GPU / hr
Open providerSponsored does not change ranking. Commissions never move a row, a GPU, or a displayed price.
In brief
What to know before you rent
- 48 GB ECC is the pitch. Diffusion, inference, mixed graphics+AI.
- 350 W PCIe. No NVLink. Do not build an 8-GPU NCCL fantasy.
- FP8 exists on Ada Tensor cores, still not Hopper HBM.
Catalog listings
L40S Pricing and Availability
Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.
ProvidersRegionBilling typeInterconnect/GPU/hr
Our reading
Who this GPU is for
I rent L40S for inference and 48 GB fine-tunes that should not live on a 4090. I do not rent it as “almost an A100.” GDDR6 and no NVLink are the product limits, not footnotes.
Rent it if Serving, image models, video encode adjacent to inference, single-GPU 48 GB work with ECC.
Skip it if NVLink training, jobs that need HBM bandwidth, anyone who actually needed 80 GB.
How we compare hoursDatasheet
Technical specifications
NVIDIA L40S · Per GPU · PCIe
Compute
- FP32
- 91.6 TFLOPS
- FP16 Tensor
- 362 TFLOPS733 with sparsity
- FP8 Tensor
- 733 TFLOPS1,466 with sparsity
- CUDA cores
- 18,176
- Tensor cores
- 5684th generation
- Precision support
- FP8, FP16, BF16, TF32, FP32, INT8
Memory
- Capacity
- 48 GB GDDR6ECC
- Bandwidth
- 864 GB/s
- Bus width
- 384-bit
- ECC
- Yes
Silicon
- Architecture
- Ada Lovelace
- Class
- Datacenter PCIe
Fabric and host
- GPU interconnect
- None
- Host interface
- PCIe 4.0 x16
Power
- Board power
- 350 W
L40S is a PCIe inference/pro card. If the listing says L40 (no S), memory and clocks differ. Read the S.
Trade-offs
Pros and cons of L40S
What this GPU does well, and when the hour is the wrong buy.
Strengths
- 48 GB ECC without Hopper rent
- Strong FP32/graphics path vs A100
- Datacenter board, not a gaming cooler lottery
- Often in stock when SXM Hopper is not
Limits
- 864 GB/s is not 2 TB/s
- No NVLink
- 48 GB still loses to 70B full states
- Easy to overpay vs a verified 4090 if you do not need ECC
Inference first
NVIDIA sold L40S to people doing inference, rendering, and “AI on a workstation server.” That is still the honest use. Training that wants HBM will feel the bus. I use it when 48 GB ECC is the constraint and Hopper is a vanity SKU.
How I actually rent it
- 01
Need ECC?
If no, price a 4090. If yes, stay.
- 02
Need 80 GB?
If yes, A100/H100. 48 GB is not close.
- 03
Need NVLink?
If yes, leave.
FAQ
L40S questions
Who has the cheapest L40S?+
As of September 5, 2026, the cheapest verified in-stock on-demand L40S tracked by GPUBeacon is $0.79 per GPU per hour from RunPod.
How much does the L40S cost per month?+
As of September 5, 2026, at 720 hours per month, one L40S costs an estimated $641 at the median on-demand price in this sample. Cheapest verified in stock: $569 per month on-demand.
Where can I rent an L40S?+
GPUBeacon currently tracks L40S configs from 5 providers. The cheapest verified in-stock on-demand configs come from RunPod, Vast.ai, Fluidstack and Lambda. See the full price comparison above for every provider and config.
Are L40S prices going up or down?+
GPUBeacon publishes a catalog sample, not a 90-day price index. As of September 5, 2026, the median on-demand L40S in this sample is $0.89 per GPU per hour. The cheapest verified in-stock hour is $0.79.
Why choose the L40S?+
Serving, image models, video encode adjacent to inference, single-GPU 48 GB work with ECC.
When is the L40S not a good fit?+
NVLink training, jobs that need HBM bandwidth, anyone who actually needed 80 GB.
L40S vs RTX 4090?+
L40S: 48 GB ECC, datacenter board. 4090: 24 GB, consumer, often cheaper. Buy ECC and VRAM on purpose.
L40S vs A100 80 GB?+
A100 wins HBM, NVLink, 80 GB. L40S wins some FP32/graphics and a simpler PCIe story. Not substitutes.
Keep comparing
Alternatives to NVIDIA L40S
Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.
NVIDIA L4
24 GB GDDR6 · 65% less bandwidth than the L40S
From$0.39/ GPU / hr
CompareNVIDIA RTX 4090
24 GB GDDR6X · 17% more bandwidth than the L40S
From$0.29/ GPU / hr
CompareNVIDIA A100
80 GB HBM2e · 2.4x the bandwidth of the L40S
From$1.19/ GPU / hr
CompareNVIDIA RTX PRO 6000
96 GB GDDR7 · 2.1x the bandwidth of the L40S
From$0.66/ GPU / hr
Read the verdictSources
Where these numbers come from
TFLOPS and TDP are from the vendor datasheet. Hourly prices are from the GPUBeacon catalog sample. Open the originals, then confirm the live SKU with the provider before you rent.