Observed arrival · 2026-11-04
LocoFósil asks what old GPUs can still do
A research project measuring whether older Nvidia GPUs can run useful local AI inference—and at what speed, power use, and cost.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- People testing local AI on older, low-cost Nvidia GPUs
- Worth noticing
- Its hardware shelf spans a 2 GB K620 to a 12 GB K6000, including cards that no longer get modern CUDA.
Field notes
The project’s hardware scope runs from Kepler through Pascal, with a listed range of 2 GB K620 to 12 GB K6000 cards. Its planned comparison includes an Optiplex/K1200 reference node and a lightweight harness for a 2B model; the homepage says Nvidia is the initial focus, with AMD and Intel results left to community contributions once that matrix is solid.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue