Quote:
For starters, there's no less than four times as many traces to place and route. That means increased board-related expenses, and really, THAT ultimately is what determines product price.
As one who has laid out many very complex boards (up to 12 layers and 1500 holes for component leads), I can tell you that what affects the cost is generally size, the number of layers, width of trace and space (down to .006" is standard cost, and cost goes up as you get into narrower ones), board quantity, and whether or not you have:
soldermask (and what kind)
legend (and whether it's on one side or both)
controlled impedances
burried capacitance
blind and/or burried vias
secondary operations (like milling a slot, or drilling a hole after the others are thru-plated)
gold plating on contacts
bare-board testing at the board house
(the majority of these not being necessary for our kind of work).
The number of traces does not affect the cost directly. If you need more layers to get it all in, or use super-narrow trace and space (like .002"!), then the price increases, but OTOH, these may result in decreased board size for a given complexity too.
Quote:
4. You don't need to reload all 32-bits of an address bus. What are the odds that you will even use half of that address space?
IOW, 31 bits instead of 32? Even if I never use more than 24 bits (16M addresses), having all 32 bits may make the address decoding easier and faster.
Quote:
Remember, the original 68000 was limited to 16MB, and the 68008 to 1MB. Lots of useful systems were built with both, each with significantly less than their complete memory capacity installed.
I don't think I want to be limited to 16MB (or 16megawords) since I might want to do digital recording (and not necessarily just audio). I won't generate 16MB of tight code in my lifetime, but recording at possibly hundreds of thousands of samples per second fills up 16MB pretty quickly. I want to implement a couple of MB of
look-up tables for very fast, accurate math too, in addition to program and data space.
Quote:
since you have the load the buggers each time.
I haven't paid much attention, but aren't they mostly self-loading from an external 8-pin serial EEPROM? It's quick and cheap.
Quote:
By all accounts, WDC has only one significant customer, Winbond. And, by significant, I mean, if Winbond stopped using the 65xx lineup tomorrow, WDC would find themselves desperately seeking new clientele to keep cash flow, or risk burning cash reserves.
It could make things interesting. On one hand, it could be a threat to the business (which would be bad for us), and OTOH, it could make them do something wonderful, like start putting a real effort into selling their own hardware, or license Microchip to put out a line of PICs with 6502's at the core instead of their decrepit PIC RISC processor.
Quote:
Overall, if you're looking for anything resembling performance, you'll want either a discrete component CPU (using actual TTL parts) or an FPGA solution.
I wish I could put the time into it. One thing I was trying to avoid however was a large number of parts, unless I could lay out a very small board that would serve as a building block and get quite a few of them made and plug them into each other as we talked about briefly on the other recent topic.
Quote:
However, you might want to consider building a stack-architecture CPU if you go this route. I think you'll find Phil Koopman's text on the subject, "Stack Computers: The New Wave" (
http://www.ece.cmu.edu/~koopman/stack_c ... index.html) most interesting.
That's another book I've been interested in but wasn't going to get until I could put the time into it. There again though, I might be into making my own stack processor, if I want stability. Ones I was interested in before, like the Harris RTX-2000, were not on the market for long.
(Edited 6/25/12 to add link to math tables web pages)