Re: The Seymour Cray Era of Supercomputers

On 5/19/2025 3:11 PM, Michael S wrote:

On Mon, 19 May 2025 18:14:17 +0000
mitchalsup@aol.com (MitchAlsup1) wrote:
 

On Mon, 19 May 2025 13:35:37 +0000, Michael S wrote:
>

On Sun, 18 May 2025 22:01:19 +0000
mitchalsup@aol.com (MitchAlsup1) wrote:
 

On Sun, 18 May 2025 8:33:30 +0000, Michael S wrote:
 

On Sat, 17 May 2025 21:27:04 +0000
mitchalsup@aol.com (MitchAlsup1) wrote:
 

Did the book relate the story of why CRAY-1 presented a DC-load
to the power supply:: that is, the ECL gates were all of the form
where they would switch 20 ma into either the true or the
complement out- put and thus have no AC energy at the power
supply level ??
>
During the CDC 7600 reign, when performing vector calculations,
(even though CDC 7600 was not a vector machine, it could stream
calculations through its execution window at impressive rates);
Certain data bit-patterns in CDC 7600 would cause more Gnd bounce
and Vdd drop than the gates cols accommodate and the machine
would take a data-dependent hard crash.
 

>
Which voltage, current and frequency are we talking about?

>
Vdd and Gnd which fed the integrated logic gates.
 

>
My question was about absolute numbers. Volts, amperes,
nanoseconds.

>
CDC 6600 was built with (effectively) RTL logic using individual parts
{transistors, resistors, capacitors, ...} well documented in "design
of a computer" Thornton.
>
CDC 7600 was built with some kind of integrated circuits, but not
TTL or ECL. I don't remember which (its been too long).
>

 I was wondering why the problem was not solved by improvement of
decoupling capacitors.
 

Yeah, it is a mystery...
Or, depending on swing or frequency, maybe using some batteries?...
Could absorb larger load swings than capacitors, but with worse frequency response (capacitors will readily pass AC transients but block DC; batteries would maintain a roughly consistent voltage but wouldn't necessarily react fast enough to deal with transients).
Would likely need both batteries and capacitors in parallel if going this route.
Though, depending, not sure if there were batteries with the correct properties.
NiMH would probably work, if the rail voltage is less than the nominal voltage (say, for a 5V rail, using 5 cells each held at around 1.0V). Cells should be stable if used this way.
Lead-Acid would not likely work as well. Lead-Acid cells need to be kept slightly above the nominal voltage, otherwise the cells sulfate (but, not so high that the water breaks down). Even in ideal conditions, Lead-Acid has a short lifespan. Would likely be best to use plain lead-metal plates if going this route.
NiCd would likely not work, cells would likely quickly develop memory effect and become useless if held near a constant voltage below nominal, or overcharge and fail if above nominal (unlike Lead-Acid, the cells do not stabilize once full charge is reached).
NiFe should probably work (would likely be used similar to NiMH). Possibly NiFe might be ideal for this, as it should be the most stable.
Modern cells (would not have existed at the time):
Maybe LiFePO4 at 2.5V per cell, roughly 0% charge, but should be "safe" (may or may not fail due to dendrite formation);
Na-Ion cells may be better (no dendrite issues), but are newer tech.
Though, these are likely a poor fit for the use-case (likely to be less stable and more prone to spontaneous failure vs NiMH or NiFe cells)
Would likely also need to disconnect the batteries from the DC rail during power-down, possibly using a relay or similar.
But, dunno...

 

 

So, Cray got rid of the problem by presenting a DC-load to the
power supply.