HanSolo wrote on 2023-08-03, 13:40:
The transistor count is not a measure for complexity but it is one indicator for it (unless the guys at Intel don't know anything about their job 😀 ).
In particular, the transistor count isn't the same number of transistors you'd use to build that piece of the circuit in discrete components on a breadboard. Due to the economics and physics of wafer production and coating and layering, the cost of a particular component to a circuit might be another layer deposited on the wafer, which has a cost in production time, and has a cost in failure rate, each process adds like 1% more likelihood of screw up or number of bad parts when they're cut up. So, in a lot of situations the solution is "MoAr transistors" on the layers they've already decided on, to work around the need for more layers. So basically, that 5 transistors over there might actually be a capacitor. However, then there are some gates that can be just single transistors very simply, and in that design work because they don't need much drive or to drive much themselves. Whereas a general purpose logic IC for that gate has to be designed more robustly, will demand a certain drive current and will supply a certain drive current. Thus then the single logic element of a programmable logic device, be it PAL/GAL or FPGA might either replace hundreds of redundant transistors, used because it was the best way to do it on "IC rules" for that process, time and budget, or replace only a mere handful of transistors, because that part of the circuit didn't have huge fanouts and loads.
So in practice, it's a bit like there's 3 different ways to make a machine, out of lego, out of erector set/meccano , and 3D print it, and despite aiming to do the same task and use the same principles, no version is going to match the other in exact details of operation and scale.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.