Hacker News new | ask | show | jobs
by wg0 772 days ago
Yield would be low. No?
3 comments

Not necessarily. It's not like an image sensor where you throw it out if theres a single dead pixel. To increase yield of CPUs, they design them to have parallel redundancy and deactivate cores or memory chunks and sell it as a lower tier model. These AI chips are way more parallel than a CPU. They had decades of statistics of wafer flaw distributions before they began the design process, so they would design in just enough redundancy to get the desired yield for the process. I wouldn't be surprised if each processor has hundreds of things disabled (ALUs,memory units, whatever theyre using), and thousands to spare.
Traditionally, that has been the issue with wafer scale processors. Cerebras is supposedly selling these things to production customers for obscene pricing, probably offsetting the low yields. Who knows if it’s profitable though or still burning cash.
Cerebras has a way of bypassing bad transistors. They over provision the WSE by 1.5%, and have 100% yields that way. https://www.anandtech.com/show/16626/cerebras-unveils-wafer-...
So it sounded like the wafer was being used as an interconnect kind of thing that might be much easier to fab.

The actual processors sound like they’re being made on other wafers in a standard way and then cut out and attached to the interconnect wafer.

So it’s not like the entire part was built only on one wafer and a single mistake would cause the whole thing to have to be thrown out.

At least that’s how I understood the article + educated guessing.