> On 2015, Oct 2, at 11:42 AM, John Hearns <john.hea...@xma.co.uk> wrote:
> 
> http://www.theregister.co.uk/2015/10/02/seymour_cray_90_anniversary/ 
> <http://www.theregister.co.uk/2015/10/02/seymour_cray_90_anniversary/>
>  
>  
> I remember learning about Gallium Arsenide and Emitter Coupled Logic when I 
> was but a little graduate student.
> I don’t suppose anyone else will remember what FASTBUS was – but it was fast 
> because it used ECL!
>  
>  
>  

The good old days!  I heard Seymour Cray speak at Stanford in the late 70’s I 
suppose.  Later at Digital
I got a chance to design with ECL for part of the first Alpha machine.  The 
Alpha EV3 and EV4 chips had truly weird IO pads, and could speak either TTL or 
ECL on the outside.  For the Alpha Demonstration Unit we chose 100K ECL because 
power and cost were not that important.  As the Register article notes, ECL has 
just beautiful signaling properties.  If you followed the transmission line 
rules exactly there just weren’t noise problems.  And fast!  That project is 
when I first had to consider picoseconds.  The jellybean gates we used were 
about 200-300 pS.

The stuff did suck power though.  IIRC we had 4-0 welding cable to connect the 
power supplies (400 Amp at -4.5 V?)
to the chassis.  The welding shop that fabricated the cables asked “what sort 
of welder uses 18 inch cables?

The other point the article makes is that Cray designed simple machines and 
back then more logic meant longer wires, which were slower.  So a simple 
machine was a fast machine.

In recent computers logic is basically free, but the resultant complexity of 
hardware and software is not so good.

-Larry

_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit 
http://www.beowulf.org/mailman/listinfo/beowulf

Reply via email to