Marktechpost has launched GPU Leaderboard, a free tool for comparing cloud GPU prices for AI inference. It also released a Chrome extension called Deploy It to access the same data. The site pulls on-demand and spot prices directly from provider pages to answer two questions: what GPU is required and where it is cheapest.
What is GPU Leaderboard
The dashboard ranks rentable cloud GPUs by price per GPU-hour. It targets teams running inference workloads.
The first version offers four views:
- Inference pricing: a cost calculator for a complete run rather than a single hourly rate.
- Leaderboard: every offer ranked by cost per GPU-hour, filterable by GPU, tier, region and availability.
- Cheapest Inference: the lowest on-demand floor for each tracked GPU.
- Providers: a directory showing instance counts, regions, status-page incidents and price-check history.
How the GPU inference cost calculator works
Users select an accelerator, GPU count, duration and utilization. The calculator prices that exact run at every provider. An optional model field narrows the list.
The default example uses 8×H100 for 72 hours at 90% utilization. This totals 576 GPU-hours. Latitude offers the cheapest result at $967.68, with a rate of $1.68 per GPU-hour. The median across 15 providers is $1,716.48. CoreWeave is the most expensive at $3,545.28.
The tool also displays a real cost per used GPU-hour. Idle time is still billed, so the $1.68 rate at 90% utilization becomes $1.87. This is a figure most budget sheets overlook.
What the Deploy It Chrome extension does
GPU Leaderboard: Deploy It brings price data to where developers select models. According to its store listing, it functions in three places:
- Hugging Face model pages: detects model size and estimates VRAM at FP16, INT8 and INT4. It then shows the cheapest hourly GPUs that fit.
- GitHub repos: detects model references and provides a hardware estimate inline.
- Toolbar lookup: allows searching for any model or GPU from the browser toolbar.
Key takeaways
- GPU Leaderboard ranks 201 live GPU offers from 24 providers by cost per GPU-hour.
- Its calculator prices a full run, including idle time based on utilization.
- Deploy It shows VRAM needs and cheapest GPUs on Hugging Face and GitHub.
- An 8×H100, 72-hour run ranges from $967.68 to $3,545.28 across multiple providers.
- Coverage and daily refresh are the main gaps compared to older trackers.
The tool is available at gpuleaderboard.com. The Deploy It extension can be installed from the Chrome Web Store.
