About VRAMGlass
VRAMGlass tracks used GPU prices and the VRAM local LLMs need, turning what this card can run and whether it is worth the price into one sortable table.
What VRAMGlass is
Every 6 hours the site pulls live listings from Xianyu and eBay, computes a median used price per device and divides it by VRAM, bandwidth and INT8 compute to get three unit prices. On the model side it syncs structure parameters and quantized sizes for popular open-weight models from Hugging Face every week and derives the VRAM each version needs. Cross the two and you can answer what runs in 16GB and what the cheapest way to run a 32B model costs.
Who makes it
VRAMGlass is maintained by the VRAMGlass editorial team. Specs and formulas are checked by hand, guides are written by people with AI only helping gather material, and every table on the site renders live from the database rather than from hand-copied numbers.
Where the data comes from
Price definitions, cleaning rules, the fit and speed formulas and the exchange-rate source are all written up on the methodology page. The coefficients are public, so you can recompute everything and tell us where we are wrong. Methodology
Affiliate disclosure
The buy buttons point at marketplace search pages, and some of those links may carry affiliate tags that pay us a small commission at no extra cost to you. Affiliate relationships do not affect the ordering: rankings come from prices and specs alone, and no placement is for sale. Amazon prices are not shown until we have PA-API access, so those buttons only say view on Amazon.
Privacy policy
There are no accounts, no logins, no personal data collected and no third-party analytics or ad-tracking scripts. The browser stores exactly two things, your theme and your display currency, both in local storage, and neither ever leaves your device.
Contact
Corrections, partnerships and data licensing: write to hello@vramglass.com.