The long-held skepticism around wafer-scale architectures is deep and goes back decades. Few have tried and all have failed, either for business or technology reasons, including the venerable Gene Amdahl. But perhaps what was missing was the right timing in equal addition to a suitably established semiconductor technology base.

If you ask Andrew Feldman why a wafer-scale approach isn’t more prevalent, his answer is simple: His team at Cerebras Systems are the only ones that has figured out how to actually do it. And in his view, no one can or will try—at least not in the near term.

“I don’t think anybody can do it. It’s taken us five years and we have a huge patent portfolio to work around. This workload, AI, will be about one-third of the total compute. If you look at Google, the canary in the compute coalmine, they’ve formed most of their work to look like AI workloads. Further, the compute being used to solve AI problems is growing exponentially, it’s astounding. Over the next three to five years, more of the work in datacenters will be AI or AI-like, more of the hard problems will be around how to find insight in that data.”

Visit the link for more.

https://www.nextplatform.com/2020/10/23/is-there-a-wafer-scale-revolution-on-the-horizon/