You Can't Escape the AI Tax

Decided on DDR4 for a new local AI server, works well as I try to keep models in VRAM anyway.

For browsers and general apps, devs have blown up memory usage like crazy the past two decades.. there's so much low hanging fruit in optimizing for reduced RAM usage.

Like many things it was cheaper to just use more memory, now it may become worth it to spend some time thinking really hard how to get your Electron message app using a few GB less..