7 Mar
2005
7 Mar
'05
3:13 p.m.
Tom Jennings wrote:
I'm still impressed by a million ops/second. Imagine! One million!
I'm still working getting a homebrew computer with a million ops/sec.
The problem is a REAL instruction requires TWO memory cycles
at least so that is 500 ns memory cycle and normal ( read vintage
parts since they don't make new I/O chips) I/O parts just meet that speed.
I feel like one of those jungle tribe people who count
"one, two,
three... many".
Better than most sales-droids. :-)
Most of the TTL computers are more-or-less one million
-- the mini
era. It's like 100 KHz for tubes, there were plenty faster but
straightforward design ignoring transmission line issues seems to
be these two numbers.
How ever most TTL computers still used core memory so the
split cycle was very matched for ISZ/DSZ type instructions.
That requires a whole new cycle generaly since most modern
memory does not suport read/modify/write cycles. The 2167
16kx1 does have data in so this may be a useful chip for
modem memory in old computers.
They always seem crisp with a 9600 bits/sec console
:-)
Gasp I hope to get 2400 baud. -:)
More seriously, most people grossly overrate the importance of CPU
speed within some reasonably local framework. We've got students
at UCI doing 3D image rendering, OK, CPU matters there (especially
with bloatware), but for general purpose computing the CPU is
usually waiting for you to type something.
Or read the stuff.
Ben alias woodelf.