Marktechpost 推出 GPU 排行榜与 Deploy It Chrome 扩展
Marktechpost 推出了 GPU Leaderboard,一个面向 AI 推理的免费云 GPU 价格比较工具。它还发布了 GPU Leaderboard: Deploy It,一个提供同样数据的 Chrome 扩展程序。该网站从各家服务商自己的定价页面读取按需和竞价价格。目标很简单:在一个地方回答“我需要什么 GPU,哪里最便宜?”
GPU Leaderboard 是一个定价仪表盘,按每 GPU 小时价格对可租用的云 GPU 进行排名。它是为运行推理的团队打造的。
v1 版本应用有 4 个视图:
你选择加速卡、GPU 数量、时长和利用率。随后计算器会按每个服务商为这次具体运行给出价格。可选的模型字段可以缩小列表范围。
默认示例是 8×H100 运行 72 小时、利用率 90%。这计费 576 GPU 小时。最便宜的结果是 Latitude 的 967.68 美元,即每 GPU 小时 1.68 美元。15 家服务商的中位数为 1,716.48 美元。最贵的是 CoreWeave 的 3,545.28 美元。
计算器还会显示“实际”的每已用 GPU 小时成本。空闲时间仍会计费,因此利用率 90% 时的 1.68 美元变成 1.87 美元。这是大多数预算表都会漏掉的数字。
GPU Leaderboard: Deploy It 把价格数据带到开发者挑选模型的地方。根据其商店页面说明,它可在 3 个场景中使用:
可在 gpuleaderboard.com 试用,并从 Chrome 网上应用店安装 Deploy It 扩展程序。
《Marktechpost AI 发布 GPU 排行榜并推出 Deploy It Chrome 扩展程序》一文最先出现在 MarkTechPost 上。
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(Marktechpost AI Launches GPU Leaderboard and the Deploy It Chrome Extension)
Marktechpost has launched GPU Leaderboard, a free cloud GPU price comparison for AI inference. It also shipped GPU Leaderboard: Deploy It, a Chrome extension for the same data. The site reads on-demand and spot prices from each provider’s own pricing page. The goal is simple: answer “what GPU do I need, and where is it cheapest?” in one place.
What is GPU Leaderboard?
GPU Leaderboard is a pricing dashboard that ranks rentable cloud GPUs by price per GPU-hour. It is built for teams running inference.
The v1 application has 4 views:
- Inference pricing: a cost calculator for a full run, not a single hourly rate.
- Leaderboard: every offer ranked by $/GPU-hr, filterable by GPU, tier, region and availability.
- Cheapest Inference: the lowest on-demand floor for each tracked GPU.
- Providers: a directory with instance counts, regions, status-page incidents and price-check history.
How does the GPU inference cost calculator work?
You pick an accelerator, GPU count, duration and utilization. The calculator then prices that exact run at every provider. An optional model field narrows the list.
The default example is 8×H100 for 72 hours at 90% utilization. That bills 576 GPU-hours. The cheapest result is $967.68 at Latitude, at $1.68 per GPU-hour. The median across 15 providers is $1,716.48. The most expensive is $3,545.28 at CoreWeave.
The calculator also shows a “real” cost per used GPU-hour. Idle time is still billed, so $1.68 at 90% utilization becomes $1.87. This is the number most budget sheets miss.
What does the Deploy It Chrome extension do?
GPU Leaderboard: Deploy It brings the price data to where developers pick models. Per its store listing, it works in 3 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 gives a hardware estimate in place.
- Toolbar lookup: search any model or GPU from the browser toolbar.
Key Takeaways
- GPU Leaderboard ranks 201 live GPU offers from 24 providers by $/GPU-hr.
- Its calculator prices a full run, including idle time at your utilization.
- Deploy It shows VRAM needs and cheapest GPUs on Hugging Face and GitHub.
- An 8×H100, 72-hour run spans $967.68 to $3,545.28 across multiple providers.
- Coverage and daily refresh are the main gaps versus older trackers.
Try it at gpuleaderboard.com and install the Deploy It extension from the Chrome Web Store.
The post Marktechpost AI Launches GPU Leaderboard and the Deploy It Chrome Extension appeared first on MarkTechPost.
正文保留英文原文,标题与摘要由机器翻译。
原文发布于 MarkTechPost(Marktechpost AI Launches GPU Leaderboard and the Deploy It Chrome Extension),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。