16GB (and in many cases 12GB) is the max vram most people will ever reasonably have

This sub is, needless to say very niche and skewed towards the high end. There are tons of extremely high end setups here with multiple gpu's etc.

Even 24GB is out of reach of most people financially, forget about the 3x3090 or 5090 or even higher setups. Macs/Strix Halo/dgspark etc are all similarly expensive. 16GB is pretty much the high end for most. And this completely changes in most of the rest of the world where even 12GB would be a luxury.

Things have changed recently (I think even last 6 months have been huge) and even agentic coding is now feasible on 16GB cards (eg with Qwen 27B quants).

I think/hope things will continue to improve. Of course there's going to be a hard limit on how much world knowledge these smaller models will have.

The holy grail is new architecture that supercedes the Transformer and new techniques that don't depend on vram/bandwidth.

and