Seems like it is more oriented for LLM inference where it has ~80% higher memory bandwidth and ~30% higher memory and tensor core performance, and 100% increase in PCIe bandwidth.
and a 25% increase in cost
yeah so they basically only improved power density, nothing else :D
Performance dont scale linearly especially in the same node. If you could get a top range Intel / AMD / Apple CPU to get 20% increase in performance for a 30% incase in power it is a massive win.
Not sure why this is downvoted. But then somewhat not surprised given the hardware level on HN.