In einer eMail vom 31.01.1999 23:58:22, schreiben Sie: << I wonder if bit-serial would be simpler (less data path, more control logic, basically). A lot of the valved machines were bit-serial for this reason. [...] I simply find bit-serial AWFUL - so much time wasted !!! Anything should be designed to get the max. performance out of given technology/budget, in my opinion; bit-serial certainly then would be a good solution for a LONG word, like 30-60 bit, doing floating point, if and only if budget limits do not allow paralell, but not for smaller things. THAT is the reason for the historic machines - they wanted floating point, therefore long words, therefore serial because of budget/size/etc constraints. This doesn't sound out of line. The CPU of the PDP8/e on my desk is 3 quad cards of TTL, mostly simple gates. Perhapes 250chips total. OK, some of them are more complex (like full adder circuits), but there's nothing that big in there. If you're going to allow anything in the TTL data book, then you can trivially make a CPU in 200 chips. You have to cheat slightly (74181 ALUs, small PROMs, etc), but it can be done. It has been done - many times. -tony The big difference is in the basic AND-OR-INVERT tube gate used: only one half tube (the INVERT, one triode section out of a dual triode tube) is used, but a lot of diodes doing the ANDs and ORs, so that one tube section is equivalent to MORE than a standard SSI 74xx IC. Typically, I would say one AND-OR-INVERT, i.e. one tube section plus associated diodes and resistors is equivalent to about one entire 74xx package, so one tube is equivalent to about two packages, on the average. John G. Zabolitzky