NPU was a miss. Limited programming model and a vendor specific API. Too much dev effort to support all different NPUs.
Today everybody has a GPU with tensor cores. Apple M5/M6 has them, New Intel/AMD GPUs have them. Nvidia Turing (RTX 20) had them already 8 years ago.
Roadmaps change all the time. You would need to specify *which* roadmap, at what point in time, remains intact. That statement is true for today, but roadmap could've changed yesterday. (That new one would then still be intact today.)
MASSIVE DELAY: Just 3 months after Jensen demoed Kyber NVL144 at GTC, it has faced major setbacks and has been delayed by more than 12 months, pushing it back to 2028. Below, we explain why Kyber has faced massive delays and why NVIDIA’s NVL72x2 back-to-back rack architecture was
