GBL Data Book

like,

I had a job offer to go work for Tektronix making electro-optic ADCs around about then. They were going to use lithium niobate Mach-Zehnder interferometers, which have a nearly sinusoidal response to the E field. The idea was that you make a little interferometer for the MSB, and double the path difference in each succeding bit. The digitizing was done by GaAs comparators, iirc, and the whole thing went about 2 Gs/s, which was pretty good for 1987.

Would probably have worked OK except that ISTR they never managed to get it linear enough for an 8-bit converter, and management made that an absolute requirement. A pity nobody knew how to make interferometers that produced Gray code. ;)

Cheers

Phil Hobbs

like,

They were doing 7 GHz distributed-deflection scan converter tubes back then! My friend Bernard still sells them.

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

I read somewhere that one of the Cray mainframes was built with GaAs NMOS. Pch simply doesn't exist in GaAs, and without that, no CMOS. Water cooling, y'think?

They could've just waited ten years and thrown a bucket of 80386s at it for the same clock rate and processing power. Price of progress...

Tim

The entire Cray-1 CPU was immersed in chilled liquid freon! I think it was MECL.

All n-channel logic is messy, like this gaas stuff and like older depletion-load nmos logic (ie MC6800.) MECL is all npn bipolar, so is a similar power hog.

Cray-1 was 80 MFLOPS!

It was expensive because the GBL never managed to get their production process to the point where the yield of working chips per wafer was anywhere near the industry standard. It was no more power-hoggy than 100k ECL - current steering logic is fast and offers clean power rails, but does so by drawing much the same current all the time.

ling, y'think?

As with the Cray-2 the processing modules were immersed in in liquid fluorinert

Or twenty years and a bunch of Pentiums/ARM processors. It's not an approach that find much favour with the marketing department, who want new product yesterday.

I've got a couple Z80-CPUs on foam around here somewhere. Dug one out, once upon a time, to play with. Works fine -- nice instruction set, a bit slow by today's standards. The 40PDIP doesn't get warm, but I can't help but wonder who thought 200mA was low power! :)

Tim

NMOS.

I guess that one must have been the one with the Freon "waterfall"

And at about 25 years later i have a laptop Dell 17R-5270 (4 execution units each with FP and HT, 8 GiB RAM, 1 TB disk; about 5 GOP + 3 GFLOP) under $1000 with comparable resources to a 1990 IBM 3090-600J (4 execution units + one vector unit, 1.5 GiB RAM, 2 TB disk; about 2 GOP + 1 GFLOP) baby supercomputer almost $ 1 Million.

Ain't Moore's law a bitch.

?-)

The 600J had six CPUs, 2MP X 3AP (MP = had its own channels and memory, AP = attached - shared channels and memory). It also went for a long shot north of $1M (like you're missing a zero), and that was without wheels.

Indeed.