level1
patreon
AI and Unified Memory Architecture: Where is it? Is it Long on Promise, Short on Delivery?
🕑 Added 2024-06-20 18:29:09 +0000 UTC
Comments
MyCookie
This whole year sometimes feels like deja vu from 2012/2014. Back then there was a very (hand-wavy) similar approach from both AMD and Nvidia. AMD introduced hUMA with their Kaveri APU, where they did a lot of work throughout the stack to build a unified memory arch APU. Whereas Nvidia made CUDA that included a software abstraction to present a unified memory arch completely in software.
