2026-09-26 21:06 UTC
Replies (3)
-
@decay@gts.todayiwilllaunchmyinfantsonintoorbit.com 2026-09-26 21:08
@mcc@mastodon.social @riley@toot.cat @MaddieM4@raphus.social @ireneista@adhd.irenes.space @amarioguy@social.treehouse.systems that's the sort of thing I was thinking. An open GPU doesn't need the kind of economic drivers Nvidia has. That's the compromise but "we have to extend an ISA or possibly eventually support different flavors of vector unit ala the Amiga" seems a much more palatable one than "No SIMD" in the John Darnielle voice and all
-
@MaddieM4@raphus.social 2026-09-26 21:13
@mcc@mastodon.social @decay@gts.todayiwilllaunchmyinfantsonintoorbit.com @riley@toot.cat @ireneista@adhd.irenes.space @amarioguy@social.treehouse.systems Architecture revision is a very reasonable good faith assumption to make, which GPU companies IMHO do not even slightly deserve. It's just an IP protection thing. SPIR-V would not exist, in an extremely successful way, if GPU ISAs were a bad idea for technical reasons. It exists nestled into Vulkan as a compromise with IP protection.
-
@riley@toot.cat 2026-09-27 15:33
@mcc@mastodon.social GPUs get their high density from not doing full CPU-like infrastructure per each vector unit. There might be a feasible Transputer-like compromise, but considering how different CPU hardware optimisation and GPU hardware optimisation are these days, I'm not inclined to hold my breath. @decay@gts.todayiwilllaunchmyinfantsonintoorbit.com @MaddieM4@raphus.social @ireneista@adhd.irenes.space @amarioguy@social.treehouse.systems