Yeah, but that's SGI doing SGI things basically. They were used to daylight robbery systems they had ultimate control over. This is nothing out of ordinary from their thinking. The weird thinking was more if you're doing a PC do a PC then, not this.. but maybe they initially didn't want to - doing an SGI workstation with PC components is more like it. They'd be shown by the market ultimately it's not what it wants. Primary cool thing about it in my opinion was cobalt and cpu aharing RAM, but what was weird about it was distribution of how much CPU gets and how much GPU was not dynamic but rather static which you had to set up manually before boot. Dynamic sharing is what only now Apple is doing. Something AMD also explored if you had their vertical (cpu, mobo, gpu) but only for fast path movement of data. I'd like to see more of that.
>>Keyfra+(OP)
The shared memory architecture came directly from the SGI O2 in 1996. The O2 had dynamic sharing, but it was impossible to make it work in Windows.
O2 dynamic memory sharing allowed things impossible on all other machines with its infinite texture memory, like mapping several videos seamlessly on moving 3D objects (also thanks to the built-in MJPEG encoding/decoding hardware).