The industry’s response is to split compute, memory, and I/O across dies, XPU chiplets are pushing toward the reticle limit, and stitch it all together with high‑bandwidth, energy‑efficient die‑to‑die ...

Kalinagaswamy: Both have their own challenges. Scale-up may have hundreds of GPUs or accelerators, and they have to work in a scale-up network. Scale-out is point-to-point. They cannot make that many ...