Framer AI AgentsDesign and publish professional sites with AI
Promoted
Launching today
VRAMGlass
Find the right GPU for running LLMs locally
0 followers
Find the right GPU for running LLMs locally
0 followers
Pick an open-weight model and a quantization, see how much memory it needs, and find the GPUs that fit. Then check what those cards actually list for: used on eBay (US) and Xianyu (China), new on Amazon Japan, with price history, sample counts and freshness labels. Plus RAM prices, a GPU price index, plain-English guides, and an embeddable live price card with an MIT-licensed SDK. Free, no sign-up, 4 languages.
Hey Product Hunt 👋 I'm voltwake, the solo maker of VRAMGlass.
Planning a machine for local LLMs comes down to two questions: will this model actually fit on this card, and what does that card really cost right now? Answering them usually means a dozen tabs: spec sheets, quantization file sizes on Hugging Face, marketplace searches. And the same card is priced very differently in the US, China and Japan.
VRAMGlass puts that in one place:
• Pick a model and quantization, see the memory it needs and which GPUs fit
• Compare GPUs by VRAM, bandwidth, power and software ecosystem
• Listing prices from eBay (US), Xianyu (China) and Amazon Japan, tracked over time, with sample counts so you can judge how far to trust a number
• A GPU price index, RAM prices, and guides on VRAM, KV cache, quantization and offloading
• An embeddable live price card for your own blog or docs (the SDK is MIT on GitHub)
It's free, with no account. Two honest caveats: prices are asking prices, not completed sales, and fit estimates are for planning, not benchmarks. The methodology page explains how every number is made.
What was the most confusing part of speccing your own local AI box? And which GPU or model am I missing?