28 Sep
1998
28 Sep
'98
8:25 p.m.
Questions came up last week about how much power a DECsystem-1090 CPU (KL10)
consumes, and how the power supply works. Tony Duell suggested that the KL10
might not consume any more power than a big PDP-11/45 system; I disagreed.
My previous description of the power system was not entirely accurate, so I've
dug out some technical documentation, including the manual "Power System
(1080/1090) Interface Description", part number EK-PWR1-ID-002, dated March
1977.
The KL10 CPU cabinet consists of three bays. In a 1080 or 1090, the middle
bay contains the main CPU card cage, which has 47 extended hex modules full
of ECL logic.
The left bay (as viewed from the front) contains an I/O card cage containing
up to 50 extended hex modules full of TTL for the RH20 Massbus channels, DTE20
front-end interfaces, DMA20 memory bus interface, and DIA20 I/O bus interface.
The right bay contains a PDP-11/40 front end processor, a TU56 dual DECtape
transport, and a TC11 DECtape control.
Note that this does not include any memory or peripherals necessary to
make a complete system. I am going to describe only the power requirements
of the CPU cabinet.
All KL10 models require three-phase power. The KL10-AA and KL10-BA models
operate on three phases of 90-132 VAC 60 Hz (208 VAC nominal phase-to-phase)
The KL10-AB and KL10-BB operate on three phases of 180-264 VAC 50 Hz (416V
nominal phase-to-phase). Total input power is specified as 12.6 kW typical
for a basic system; a real-world configuration would have more DTE20 and
RH20 units, so it would consume more power.
Needless to say, the power supplies in the KL10 are not power factor
corrected. The 863 three-phase power control has a 75A main circuit breaker.
Unfortunately I do not have a copy of the DECsystem-10 Site Preparation Guide,
so I don't know what the actual specifications for peak current per phase
are.
With the exception of the power supply for the main CPU cage and the cooling
fans for that bay, most of the machine actually operates on single phase
power. The various fans and power supplies are distributed between the
phases to balance the load.
The CPU cage fans operate between two phases (208V) in the -AA and -BA models,
and between one phase and neutral (240V) in the -AB and -BB models.
The power supply for the ECL consists of an H760A or H760B raw power supply
(depending on the line voltage), a 70-09475 capacitor assembly, and an H761
regulated series pass assembly.
The H760A/B contains three ferroresonant transformers with primaries wired in
a three-phase delta. The secondaries are center-tapped and wired with
half-wave bridge rectifiers. The output is -12V at 490A. The diodes are
mounted to a massive heat sink with its own fans.
The capacitor assembly consists of thirteen 0.3F capacitors in parallel for
a total capacitance of 3.9F. The output under full load is -11V+-10%.
In the event of a power failure, this is sufficient to allow a hold time
of 20 ms before the voltage drops below -7.5V.
The H761 regulated series pass assembly consists of 13 linear regulators.
Nine have -5.2V 35A outputs for the ECL main power, and four have -2V 35A
outputs for ECL termination. There is room for a 14th linear regulator
which is normally not installed.
The maximum load for the CPU card cage is thus (9*5.2V*35A)+(4*2.0V*35A), or
1918W. At that load, the input power required by the power supply is
over 5460W. This is less than 35% efficiency.
Of course, the ECL power supply is nowhere near the whole story.
The following additional power supplies are in the system:
quant model output use
----- ----- ------ --------------
1 H720 +5, -15 DECtape system
1 H725
1 H7420 AC power supply PDP-11/40
3 H744 +5
1 H745 -15
1 H754 +20, -5
1 H732 0-20VDC margin check
1 783-C +10, -15
1 H7420 AC power supply
1 H744 +5V CPU cage TTL supply
2 H744 +5V I/O cage DTE20 TTL supply
1 H744 +5V I/O cage RH20 TTL supply
1 H7420 AC power supply
4 H744 +5V I/O cage RH20 TTL supply
1 H7420 AC power supply
1 H770 +15V air flow sensors
1 H745 -15V I/O cage DIA20/DMA20
3 H744 +5 I/O cage DIA20/DMA20 TTL supply
I have not found any figures for typical current used by the various
subsystems. However, the ratings of the regulators are:
H720 5V 22A, -15V 10A, unregulated -22V 1A, unregulated +8V 1.5A
(unregulated outputs not used in KL10)
H732 0-20V 5A
H744 5V 25A
H745 -15V 10A
H754 +20V 8A and -5V 1A
H770 +15V 10A
DEC didn't waste any more money on regulators than necessary, so it is
a safe assumption that a fully configured system will use most of the
available DC power, which for just the +5V supplies alone is 1860W.
Fortunately the power supplies used with the H7420s are switchers, so they
are probably at least 65% efficient.
Eric
29 Sep
29 Sep
6:26 a.m.
A non-text attachment was scrubbed...
Name: not available
Type: text
Size: 3737 bytes
Desc: not available
Url :
http://www.classiccmp.org/pipermail/cctalk/attachments/19980929/6ac9e9a2/at…
10:30 a.m.
Tony Duell wrote:
What I meant was that although they used 25A regulators for all of the 5V
supplies, and I'm sure that none of them are actually run at close to the
current limit, they wired the machine with as few 25A regulators as they
reasonably could. They didn't have four 25A regulators each being used
for 5A, for instance, when a single regulator would do.
Agreed.
But I was arguing that it would be possible for an
enthusiast to run and
maintain a KL10. I stand by that statement...
Since I (half-)own a KL10, I'd like to believe so also. I'm just pointing
out that it's not going to be easy.
I was only considering the CPU rack. I was comparing
it against a pretty
full 11/45 system as there are a number of those owned by enthusiasts,
with a number of peripherals on them, which they keep going.
By CPU rack, do you mean the entire 3-bay cabinet (which still does not
include peripherals), or do you mean just the CPU card cage in the center
bay?
The ECL requirements (-5V @ 400A say total) are not as
bad as I'd have
thought. That PSU could be replaced (temporarily) with a switcher to keep the
machine running.
My plan is to replace it with Vicor power supply modules, which include
DC-DC converters and AC-DC front ends.
The VI-HAM front ends accept a wide range of AC input voltages, and produce
about 300VDC for input to the DC-DC converters. They are power-factor
corrected which is vital to minimizing the necessary circuit amperage.
Each VI-HAM handles 600W.
The DC-DC converters are available in 2V 40A and 5V 40A models; 5.2V is well
within the adjustment range. They have remote sensing. It is important to
the KL10 power layout to maintain separate remote sensing power supplies for
each section of the backplane, hence it is *NOT* a good idea to replace the
original nine 5.2V 35A supplies with a single 5.2V 315A supply.
The DC-DC converters are rated at a minimum efficiency of 80%, so the
maximum load of 1918W may require 2397.5W from the front end modules.
In principle four of the 600W VI-HAM modules should be sufficient for this,
but I'm inclined to go with five VI-HAM modules and distribute the loads in
order to not run them quite so close to the maximum rating.
The VI-HAMs are rated at a minimum efficiency of 90%, and a power factor of
0.99. So the AC input will have to supply 2664W, which will be under 12A
at 240VAC. This is about 72% efficiency (minimum), almost twice the
efficiency of the original linear supplies.
Assuming that the rest of the KL10 power supplies (which are not power factor
corrected) don't end up requiring more than 18A at 240VAC, I should be able
to operate the KL10 from a common 30A circuit.
Of course, we haven't even discussed memory, peripherals, and cooling yet.
The Ampex 4 MW memory box requires a few KW. RP06s and TU78s use hefty
amounts as well.
DEC didn't
waste any more money on regulators than necessary, so it is
Actually, that's
not always true! In many cases on 11's, the PSUs are
overrated by quite high margins. The 5V 25A regulator was a standard part a safe
assumption that a fully configured system will use most of the
available DC power, which for just the +5V supplies alone is 1860W.
OK, call that 2kW. Erring on the safe side, lets say we need 3kW input to
the (modern) SMPSU that we'll use to replace the ECL supply. Say 1kW for
the other lines. That's 6kW. Not at all impossible to find. In fact I
think I've had that sort of total power in my machine room at once already.
No, I expect that the rest of the KL10 will require at least another 2kW
or so, not the 1kW you suggest.
However, your total of 6kW suggests that perhaps you really meant 3kW for
all the non-ECL power.
Cheers,
Eric
11:46 a.m.
A non-text attachment was scrubbed...
Name: not available
Type: text
Size: 3811 bytes
Desc: not available
Url :
http://www.classiccmp.org/pipermail/cctalk/attachments/19980930/9c896658/at…
12:50 p.m.
Tony Duell wrote:
Seems reasonable. I'd keep the original PSU,
though, and make as few mods
as possible. Just in case you want to go back to the original config at
some time.
Absolutely!
None of the changes we intend to make to the KL10 involve any permanent
alteration. Aside from the power supply, we want to change two other
things:
Memory - We have an Ampex 4 megaword ECC memory box for the KI10-style memory
bus (which is interfaced to the KL10 through the DMA20 memory bus interface in
the I/O bay). However, it consumes massive amounts of power, and is
physically huge. The main CPU card cage has two "S bus" memory interfaces
(not to be confused with Sun's S-bus which came many years later). For each S
bus there is an ECL<-->TTL translator board, and a slot for cabling to the
DMA20. I've done preliminary design work on a board to plug into the S bus
cable slot and implement the 4 megawords of memory using six 8M*8 SDRAM
chips. I'd use a DIMM, but it won't fit mechanically due to the 12.7 mm
board spacing. Over 55% of the memory bits will go to waste!
Disk and tape I/O - Several companies made Massbus-to-SCSI adapters; most
of them only supported 16-bit Massbus transfers for use on VAXen and
PDP-11en, but a few of them supported 18-bit Massbus needed for the
DECsystem-10 and DECSYSTEM-20. I think I've talked someone who once
worked on these things to lend or give me one of the leftover prototype
units. I don't have it yet, though, so this part could still fall through.
We'll still need to have the capability of using the RP06 (and perhaps the
TU78) on occasion in order to deal with standard media.
30 Sep
30 Sep
8:21 a.m.
A non-text attachment was scrubbed...
Name: not available
Type: text
Size: 3445 bytes
Desc: not available
Url :
http://www.classiccmp.org/pipermail/cctalk/attachments/19980930/ffd98be7/at…
11:26 a.m.
I wrote:
Memory - We have an Ampex 4 megaword ECC memory box
for the KI10-style memory
bus
...
I've done preliminary design work on a board to
plug into the S bus
cable slot and implement the 4 megawords of memory using six 8M*8 SDRAM
chips.
Tony Duell replied:
If it was my machine, I'd probably try to get the
memory box working,
even if it eats power (what? 3kW or so?). On the other hand, a new memory
subsystem is certainly a reversable mod, and it's something worthwhile...
If it was a DEC memory box, I'd be more interested. We're not going to
throw it away, but I don't really see any point in trying to use it. If
it was core memory, that would be interesting, but it's just a humungous
box of 16K DRAMs, which aren't exciting at all. And it's certainly not
representative of the type of memory that was available when the machine
was introduced.
Cheers,
Eric
9580
days inactive
9581
days old
6 comments
2 participants
participants (2)
-
ard＠p850ug1.demon.co.uk -
eric＠brouhaha.com