At this week’s HotChips 2022, UntetherAI has announced its second generation at-memory architecture to accelerate AI interference. SpeedAI (codenamed Boqueria) improves efficiency by up to 100x, compared to the company’s first generation architecture (runAI200) and has 80 times more throughput to accelerate neural networking. The second generation architecture also quadruples performance achieving 2Petaflops with a ...
The post Untether AI moves to floating point for SpeedAI architecture appeared first on Electronics Weekly.
from | Electronics Weekly https://ift.tt/OmN1bfP
via Yuichun
沒有留言:
張貼留言