Making something high-spec like a graphics card is expensive, and communicating with the CPU via PCI is structurally disadvantageous, so going with a CPU+iGPU format seems like the right answer. Is this similar to using an M chip in a Mac mini?
Honestly, the price is still very high. ㄷㄷㄷㄷ I wish AMD and Intel would hurry up and put in some effort.
It wouldn't be bad if a dockable device like a smartphone with Linux and Android came out, specifically for AI.
Right now, RAM prices are high and infrastructure investment is a priority. For the market to stabilize, it seems like it will take 6 years at the earliest, or realistically around 10 years.
Of course, devices like the Mac mini with an M chip are still expensive, even though they exist now. The performance is a bit disappointing too...
I hope that the day will come soon when smartphones become popularized and every manufacturer starts producing them.
Until then, it seems like the best solution is to give up on finding cheap 32GB graphics cards anywhere and just use free frontier model APIs.
But I can't easily give up on the romance of running something locally. ㅜㅜ
