Natalia Vassilieva & Kirk Bresniker | HPE Discover 2017
summary
Shows a working prototype of The Machine, a memory-driven architecture built around Memristor and a shared memory pool that lets all compute nodes address the same persistent memory so training no longer bottlenecks on node-to-node communication. Claims the prototype is active and cuts training time dramatically, enabling model retraining on live data with microsecond measure‑analyze‑act cycles at the edge, cites Gen‑Z fabric and resistive RAM partnerships and estimates brain‑scale models need ~10^26 FLOPS.
AI-written summary · about
