Most deployedCloud GPU
L40S
L40S is available through OneInfer GPU marketplace with live provider pricing, regions, VRAM, and deployment options from the API.
VRAM
48GB
Best price
$0.55/gpu/hr
Providers
6 available
Regions
16 regions
VRAM
48GB
CUDA cores
18,176
TDP
350W
Bandwidth
N/A
Process node
5nm
Launch date
N/A
Available from 6 providers
Sorted by price · click price to re-sort| Provider | Regions | Config | Price / hr ▾ | Availability | |
|---|---|---|---|---|---|
 novita | AF-ZA-01, AS-AE-01, AS-IN-01, AS-SG-02, EU-GB-01, EU-DE-02, EU-IS-01, JP-TYO-02, OC-AU-01, SA-BR-01, US-CA-06 | 1x 48GB | $0.55 | Limited | |
 vastai | AS-TW-01, US-TX-01 | 1x 48GB | $0.74 | Limited | |
 runpod | US-TX-04, OC-AU-01 | 1x 48GB | $1.09 | Limited | |
 e2e_networks | IN-DL-01 | 1x 48GB | $1.20 | Limited | |
 nebius_ai | EU-FI-01 | 1x 48GB | $1.35 | Limited | |
| vastai | AS-TW-01 | 2x 48GB | $1.60 | In stock |
Full specifications
Data from GPU specs and provider APIsAI TOPS733
FP8 TFLOPS1,466
FP16 TFLOPS362
BF16 TFLOPS362
FP32 TFLOPS91.61
INT8 TOPS1,466
Sparse multiplier2
Benchmarks
L40S vs comparable GPUs on the platformInference throughput
7B model · tokens/sec
B200 SXM15,000
H200 SXM8,500
H100620
A100500
GeForce RTX 5090180
Training performance
Overall training multiplier
B200 SXM30x
H200 SXM18x
H10015x
A1003.4x
L40S2.8x
GeekBench OpenCL
Compute benchmark score
GeForce RTX 5090367,740
H200 SXM343,598
L40S337,706
H100336,474
B200 SXM333,768
Best for
LLM fine-tuningHigh-throughput inferenceMulti-modal model trainingLarge-batch distributed trainingRAG pipelines at scale
Similar GPUs
Ready to deploy L40S?
Spin up an instance in minutes from your choice of 6 providers.