AMD acquires Taalas to boost inference performance by etching models in silicon

Re:AMD acquires Taalas to boost inference performance by etc... by briansm on Saturday August 08, 2026 @01:31AM (#49219133)

Cerebras have special techniques to work around etching errors / bad cores on their wafers. This is possible since their wafers are effectively hundreds of identical copies of redundant cores. Can't do that for a globally unique model.

Etching failure in that situation would be like brain-damage in a human, all sorts of weird effects would start appearing.

[ Parent | on HN ]

Re:AMD acquires Taalas to boost inference performance by etc... by fulafel on Saturday August 08, 2026 @01:40AM (#49219160)

There's several ways to engineer around that as the errors are detectable. There's a big literature on how to trade off speed or transistors for error correction. [1]

(Is Cerebras doing something novel? CPUs and memory blocks have been doing those things for a long time too, since the error rate is otherwise too high for normal size chips as well)

[1] see eg https://www.vlsimentor.com/dft/redundancy-bisr to get some basic concepts

[ Parent | on HN ]

Re:AMD acquires Taalas to boost inference performance by etc... by phonon on Saturday August 08, 2026 @04:18PM (#49225500)

A few hundred bad bits/transistors in a trillion+ parameter model would compromise its abilities not one iota...the models are inherently lossy and resistant to "brain damage"...

[ Parent | on HN ]