NVIDIA L40 GPUs, by the hour
48 GB each, as ready-to-use Endpoint: A web address (HTTPS) your app sends requests to. The supplier runs the GPU behind it; you never manage a server. in the US and Europe. Live within 24 hours, Pay as you go: You pay for the GPU-hours you run, from a balance you top up. No term; stop any time., no term.
if it runs all month · $0.79 per GPU-hour: One GPU running for one hour: the unit compute is priced in. A server of 8 GPUs for a day is 192 GPU-hours.
13% under the public on-demand median of $0.91 (8 providers, 2 Nov 2026) · all L40 prices
From 100 GPUs the price is a quote
Order L40 →What runs on it
- · LLMs up to ~30B parameters at FP8 (~14B at BF16), and mixture-of-experts models to ~40 GB at FP8
- · FLUX- and SDXL-class image models
- · Video generation (H3, Wan-class)
- · Embeddings, rerankers and Whisper-class speech
Ways to run it: Your container · An open model we deploy for you · Hosted video API (H3).
What we measured
- 5 Oct 2026 · 24 video jobs at once: 23 ran in parallel, all 24 delivered
- 7 Oct 2026 · Qwen3-30B-A3B at FP8 on 2 GPUs: 161-179 tokens/s on one stream, 2,754 tokens/s across 64 streams, no errors
- 14 Oct 2026 · Re-checked: video and LLM endpoints both serving, no errors
What you get
- · Ready-to-use endpoint: Your model or container runs behind its own web address (HTTPS) and key. No servers to manage, and no SSH.
- · Data centers in the US and Europe; placement is automatic.
- · Live within 24 hours of your container (or the model you name) and the first top-up.
- · Per GPU: 48 GB GPU memory, about 50 GB system memory, about 175 GB local disk (wiped on restart).
- · No SSH, virtual machines, bare metal or root, and no job larger than one GPU.
Evidence level: How much of what a supplier says we have checked ourselves, from Declared (the supplier says so) to Delivered (a rental here ran its term).
Declared. The supplier says it has this capacity and has passed our company checks. We have not measured it ourselves.
How you pay
- · Pay-as-you-go from a prepaid balance, drawn down by the GPU-hour while GPUs are allocated. No term: scale up or down any time.
- · The compute partner invoices the prepaid balance and is paid directly. Silicon Liquid never holds your money.
- · Nothing is charged when you order. We confirm the GPUs first.
- · You pay the price shown and nothing to us: the supplier pays our commission.
Order NVIDIA L40
Four things, under a minute. Nothing is charged now. Next, our desk confirms with the supplier, usually the same business day. Then the supplier emails you its invoice.
Not sure, or want a person? Arpit Dhamija runs the Silicon Liquid desk and replies the same business day: arpit@siliconliquid.com.