Rent L40S
Compare live on-demand and spot rental prices across 97+ cloud providers. Best Ada Lovelace data center GPU. High FP32, 48GB GDDR6. Strong inference-per-dollar.
Choosing the right billing model for L40S
Provision and terminate at any time. Ideal for development, short experiments, and workloads with unpredictable duration.
Instances can be reclaimed when demand spikes. Best for fault-tolerant batch jobs, training with checkpointing, and preprocessing.
Lock in a rate for 1–3 months. Right for sustained production inference or long training runs where cost predictability matters.
Best use cases
- Inference serving
- Generative AI
- Mixed compute + graphics
L40S — Specs & Benchmarks
Performance bars, compute tiers (FP32/FP16/BF16/FP8/INT8), memory specs, LLM model size guidance, and related GPU comparisons.
L40S Rental Guide
The L40S is the most cost-effective option for inference serving of models in the 7B–30B parameter range. At $0.80–$1.80/hr on-demand — roughly half the cost of an A100 80GB — it delivers comparable throughput for models that fit in 48GB. For Llama 3.1 8B, Mistral 7B, and similar models, the L40S often provides better cost-per-token than any HBM-based alternative. Its 91.6 TFLOPS FP32 also makes it the top choice for Stable Diffusion and image generation workloads.
Spot instances for L40S are available at 35–55% below on-demand rates. The L40S is increasingly available across providers including Lambda Labs, CoreWeave, and Vast.ai. For batch inference jobs and image generation pipelines, spot L40S instances are often the best cost-per-output option available. On-demand is appropriate for production serving where uptime matters.
The main limitation to plan around is the 48GB GDDR6 ceiling and 864 GB/s bandwidth. For models above 30B parameters at FP16, you will need either quantization (INT8/INT4) or multiple cards — but without NVLink, multi-card VRAM pooling is not supported. If your model requires more than 48GB, consider the A100 80GB or H100 80GB instead. Use the GPU cost calculator to compare total cost across L40S, A100, and H100 for your specific model size and throughput requirements.
Frequently Asked Questions
How much does it cost to rent a L40S?
L40S on-demand rental prices vary by provider and region. On-demand rates typically range based on availability and provider margins — use the comparison table above to see current live rates across all providers. Spot instances are generally 40–70% cheaper than on-demand but can be interrupted. Monthly cost estimates (hourly rate × 730 hours) are shown in the table for sustained workloads.
Which cloud provider has the cheapest L40S?
The cheapest L40S provider changes as providers update their pricing. The comparison table above shows live rates sorted by price, so the cheapest option is always at the top. Factors beyond headline price include region (latency to your users), availability (high/medium/low), and billing granularity (per-second vs per-hour minimums).
What can I run on a L40S?
With 48GB of GDDR6, the L40S can run LLM models up to approximately 24B parameters at FP16, 48B at INT8, or 96B at INT4/GGUF quantization. Common workloads include: Inference serving, Generative AI, Mixed compute + graphics. Best Ada Lovelace data center GPU. High FP32, 48GB GDDR6. Strong inference-per-dollar.
Should I use on-demand or spot pricing for L40S?
Spot instances save 40–70% vs on-demand but can be interrupted when the provider needs capacity back. Use spot for: batch inference jobs, training runs with checkpointing, preprocessing pipelines, and any workload that can tolerate interruption and restart. Use on-demand for: production inference serving, interactive workloads, and jobs that cannot be interrupted. Most providers bill per second, so short on-demand jobs are not penalized by hourly minimums.
How does the L40S compare to the H100 for cloud rental?
The H100 80GB delivers 1,979 TFLOPS FP16 with 3,350 GB/s HBM3 bandwidth, compared to the L40S's 183 TFLOPS FP16 and 864 GB/s bandwidth. The H100 is significantly more expensive — typically $2.50–$5.00/hr vs lower rates for the L40S. For workloads that fit within 48GB and don't require FP8 precision, the L40S often delivers better cost-per-token than the H100.
What is the memory bandwidth of the L40S and why does it matter?
The L40S has 864 GB/s of memory bandwidth. For LLM inference, memory bandwidth is often more important than raw TFLOPS — each autoregressive token generation reads the full model weight matrix from VRAM, so bandwidth directly determines tokens-per-second throughput. Higher bandwidth means faster inference for the same model at the same batch size. For batch inference (processing many requests simultaneously), compute throughput becomes more important.
Can I use the L40S for Stable Diffusion or image generation?
Yes — the L40S is well-suited for Stable Diffusion and image generation workloads. Image generation is primarily FP32 and FP16 compute-bound, and the L40S's 91.6 TFLOPS FP32 throughput determines images-per-second. The 48GB VRAM fits SDXL (requires ~6GB) and most ControlNet pipelines. For high-throughput image generation at scale, compare cost-per-image across providers using the GPU cost calculator.