Supercomputers, fishing for information

Jon Elson elson at pico-systems.com
Mon Nov 7 12:25:16 CST 2016


On 11/07/2016 10:59 AM, ethan at 757.org wrote:
>> All 370's ran off 415 Hz 3-phase power. The mid-sized 
>> ones had the motor-generator set built into the back of 
>> the machine.  The 415 Hz (regulated) power was 
>> transformed to low voltage and run through an 
>> inductor-input filter and then series pass regulated.  
>> They had a circuit they called an "electronic capacitor" 
>> that pulled extra current through the inductor during the 
>> voltage peaks, so the inductor carried enough current 
>> during the voltage valleys.  This reduced the ripple 
>> current on the capacitor banks.
>> The 360s ran off single-phase 120 V 2500 Hz power, 
>> produced by a "converter-inverter" unit in the back, that 
>> converted utility 3-phase power to DC, then inverted it 
>> with an SCR inverter. This made the DC power supplies in 
>> the machine quite small. Since they ran off regulated 
>> 2500 Hz power, they dropped only a very small voltage 
>> across the series-pass transistor.
>> Jon
>
> Am I getting this right?
>
> So they take AC turn it to DC then turn it back to high 
> frequency AC then turn it back to DC to drop the need for 
> larger capacitors to keep the DC clean?
>
> Pretty wild.
>
Yes, that's how the 360's (at least mid-range) were set up.  
You could tell, the converter-inverter was INSANELY loud, at 
a massively piercing audio frequency.  The key was that this 
one unit gave regulated AC power to all critical power 
supplies in the system, so they could skimp on regulation on 
the individual power supplies.
It actually didn't work so well, if you had crummy power, 
which we had a bad case on at Washington University in the 
70's - 80's.  They had 4160 V cables buried all over campus, 
fed from one transformer at the power plant.  These cables 
were in concrete "conduits" which had water leaks.  So the 
wires, with 4160 V on them, were essentially sitting in 
muddy water all the time, and corona discharges broke down 
the insulation.  We had one of these cables fail every 
couple months, with a massive power dip for a few seconds 
before the fuses cleared.  But, for weeks before the big 
POP, they would be arcing and flashing, which drove the 360 
crazy.  A small dip would cause things like control store 
parity errors, main store parity errors or just system 
power-on resets.  They got a Dranetz analyzer, and could 
correlate more than half the machine crashes with a power 
glitch.  They got some big Digital Power Systems 
ultra-isolation transformers, but I really don't think they 
helped much, as the problem was momentary dips.  (Later, 
shielded cables came in, these have a layer of copper mesh 
over the main insulation, so there is no AC field present on 
the outside of the cable.  As far as I know, the original 
shielded cables are still in there, once they got all the 
old unshielded stuff out, these problems just stopped 
happening.)
> Anyone run any of this stuff at home / light commercial 
> clubhouses?
>
We tried to get a 370/145 running at a guy's house.  That 
had the 17 KVA motor generator set in the back (WAY more 
than a 145 needed, but they apparently used one MG set for a 
range of machines).  But, he only had 60 A 240V single-phase 
service, and we couldn't even spin up the MG set with no 
load.  We built a static phase converter, but the imaginary 
current was over 60 A.  Well, NO SURPRISE, if we'd just read 
the nameplate we would have known it was a fool's errand. 
The 3-phase line current on the thing was about 55 A, per 
3-PHASE LINE, so running it off SINGLE-PHASE, the line 
current would HAVE to be 1.7 X that much, wouldn't it?  OF 
course!

We should have just scrapped the original 415 Hz power 
supplies and got our hands on a bank of 5V supplies and 
adapted them.  The thing ran off +1.2 and -3 V supplies at 
390 A, although I think that was for the max configuration, 
this one had minimal internal memory, so should have been less.

Jon


More information about the cctech mailing list