

The new Framework Desktop with 192GB unified memory, specifically for LLM inference. I’ve looked at the numbers and there’s no scenario for me in which it’s economically sensible to upgrade from 20GB gpu vram + rent EU API inference for larger models. Even if the tokens became more expensive by factor of 10 overnight, by the time the hardware paid for itself it would already long be obsolete.
It’s an impressive machine, but I just can’t justify the purchase.




Finally someone who at least knows the difference between the WWW and the Internet…