I don't quite get the complexity/performance graph's X axis. Around the time of the early microprocessors in the 1970s, vastly greater performance was obtainab

I don't quite get the complexity/performance graph's X axis. Around the time of the early microprocessors in the 1970s, vastly greater performance was obtainable via discrete parts, for the simple reason that more parallel stuff could be built than the LSI of the day could accommodate (e.g. 32-bit datapath). He even has a specific X axis point with the MSI 74181 bit slice ALU chip - which was in all the high-performance discrete part CPUs of the day, either that one or the AM2901 which also contained registers. So X axis should be in net complexity in transistors or gates, regardless of implementation. That being said, the device presented still gets an enormous amount of compute per TTL chip, especially considering no dedicated ALU is involved.