Observing someone else's design! Not mine! Just seeing how they designed their system.
Things I've added (only two things):
1) I fixed the single pixel GPU contention issue (not the GPU RAM reset issue)
2) I implemented my "fast execution" system in it (see a later version of the Femto-4 to see its origins, or see the Delay Introduction for its explanation)
This computer is far less intensive than my design in terms of simulation load, and can execute 256 instructions per clock cycle as opposed to the Femto-4's measly ~16. (Admittedly the Femto-4 does do more per instruction but not 11 times more. Also I suspect the Femto-4's graphics updates are partly to blame for this). This raises the execution speed from 10 instructions per second to 2560 instructions per second. It is so fast that it can run through all 65536 instructions the computer can store in 25.6 seconds. When running the Hello World Program it runs through the entire 84 line code 3 times, and then executes 4 more instructions, making it appear to only execute 4 instructions at a time.
Note that the execution speed could be doubled by using both edges of the clock to pulse the loop, increasing the instructions per second to 5120, making it actually capable of running decent programs. With this it probably would be feasible to drive decently fast graphics.
Latest Femto-4 Update: