Rent NVIDIA L40S GPU by the hour
The NVIDIA L40S is an Ada Lovelace datacenter GPU available for cloud rental on ONIE. It provides 45 GB of VRAM, about 73.3 TFLOPS of single-precision compute, and a memory bandwidth of about 626 GB/s within a 350 W power envelope. Supporting CUDA compute capability 8.9, this instance handles demanding AI training, inference, and graphics workloads with per-hour billing.
Specifications
VRAM per GPU
Architecture
Compute capability
FP32 compute
Memory bandwidth
Power limit
Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.
Available configurations
Live stock and prices, updated continuously. The final price is confirmed before you order.
Loading live prices…
What the L40S is good for
Large Language Model Inference
Deploy quantized or full-precision language models using ready templates like vLLM and Ollama within the 45 GB memory capacity.
Generative Image Workflows
Run complex diffusion pipelines and batch generation tasks in ComfyUI leveraging the Ada Lovelace architecture.
Deep Learning Fine-Tuning
Fine-tune neural networks and LoRA adapters in PyTorch/CUDA environments using about 73.3 TFLOPS of compute.
Datacenter Graphics and Rendering
Execute headless rendering and simulation tasks that benefit from a datacenter GPU with about 626 GB/s of memory bandwidth.
Is it the right GPU for you?
Choose the L40S when your workload requires 45 GB of VRAM and datacenter reliability, stepping beyond standard consumer cards. If your models fit into less memory, a smaller workstation or consumer GPU is more economical, whereas massive unquantized foundation models warrant a larger memory datacenter GPU.
Compare with similar GPUs
Frequently asked questions
Is the NVIDIA L40S suitable for LLM inference and fine-tuning?
Yes, with 45 GB of VRAM and CUDA compute capability 8.9, the L40S is well-suited for serving medium-to-large language models and running parameter-efficient fine-tuning. You can launch instances with pre-configured vLLM, Ollama, or PyTorch/CUDA templates.
How does the L40S compare to consumer GPUs or larger datacenter GPUs?
The L40S offers more VRAM than standard consumer Ada Lovelace cards, making it capable of holding larger model weights without offloading. However, it has lower memory bandwidth than high-memory compute-only datacenter GPUs, which are preferable for massive memory-bound training jobs.
How do I connect to and configure my L40S instance on ONIE?
You get full SSH access to your environment as soon as the instance boots. You can also select ready templates including PyTorch/CUDA, ComfyUI, vLLM, and Ollama to start working immediately.
How does billing work for the L40S?
ONIE uses per-hour billing for all L40S rentals. You only pay for the time your instance is allocated.
Rent the L40S by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent L40S