Very interesting! I was aware of the 5nm advancements and the limitations of chip sizes approaching the physical limitations of the material but I had been assuming since we worked around the single core issue a similar innovation would appear for this bottleneck. It seems like the focus instead was turned towards integrating AI into the gpu architecture and cranking up the power consumption for marginal gains in performance instead of working towards a paradigm shift. Thanks for the in depth explanation though, I always appreciate an opportunity to learn more about this type of stuff!