discrete transistors

From: Eric Smith <eric_at_brouhaha.com>
Date: Mon Oct 19 16:32:03 1998

Allison J Parent wrote:
> I disagree that a 286 had more raw cpu performance than some of the big
> transistor machines. I still remember the BOCES/LIRICS KA10 running some
> 300 users. I've never seen a 286 run more than 4. Same applies to IBM
> 2060s.

I never said that there were NO discrete transistor machines that could
outperform a 286. I cited an example of a specific very large
discrete-transistor machine made in 1962 which the 286 can outperform on
many tasks. My point was that the larger a discrete-transistor machine
is, the more difficult it is to make it run fast.

Number of users supported is not a reasonable benchmark of raw CPU
performance.

The KA10 was nearly two orders of magnitude simpler than the Stretch CPU. Yet
it was faster than the Stretch for some things. However, this is not too
surprising since the Stretch was IBM's first attempt at designing a
transistor-based computer. The 7090 and several other IBM transistor-based
computers shipped before Stretch, but those designs were started later and
were based on the circuit and packaging technology developed for Stretch.

My entire point was that it is essentially impossible to build a
discrete-transistor machine compatible with a modern high-end single-chip
microprocessor (such as a Pentium II) and achieve better performance than the
single-chip implementation. In fact, even using SSI and MSI integrated
circuits will not get you there.

Eric
Received on Mon Oct 19 1998 - 16:32:03 BST