Indy Power Supply information

Elf

Storybook
Feb 4, 2019
56
8
8
Hi all,
It is probably no secret that I am designing a replacement Indy PSU using off the shelf DC-DC converters. While that is a matter for another series of posts, along the way I have collected some basic information about the functioning of the Indy PSU.

If referencing or duplicating this information, I ask that you maintain a link to this original forum post for further updates.

To start that off, here is the power supply pinout from the motherboard's perspective:
3

Power is supplied on a 20-pin Molex Mini-Fit Jr. type connector, with the motherboard side being the male connector and the power supply side being female. The pin numbers are as per the mechanical drawing.

The +5V standby is supplied when the power supply is connected to AC line power but has main power turned off via the motherboard. I am guessing the -12V rail is for audio amplifiers.

Pin #Wire ColorFunction
1White+3.3V
2White+3.3V
3Red+5V
4Red+5V
5Red+5V
6Red+5V
7Red+5V
8Red+5V
9Green+5V Standby
10Light Blue-12V
11Black0V
12Black0V
13Black0V
14Black0V
15Black0V
16Black0V
17Yellow+12V
18Black0V
19Black0V
20Black0V

4


Auxiliary connections are supplied on a 20 position 0.1" spacing (2x10) pin header, with the motherboard side being male and the power supply side being female. Noting the position of the connector key, pin numbering on these connectors alternates between the bottom and top row, starting from the bottom left.

The power supply is controlled by pin 3 (Run), which is "ground to stop." It naturally floats at +5V (standby), meaning that disconnected from the motherboard, the power supply runs and supplies regular power. The motherboard grounds out this pin to stop the power supply.

The temperature sensor appears to be a thermistor. I have not yet characterized it but will do so in a later follow up. It appears to control the fan speed, dramatically so for Sony models.

Many people also seem to be unaware that the Indy does have a reset / reboot switch. This switch is the small nub beneath the power switch.

Pin #Wire ColorFunction
1N/C
2BrownTemperature Sensor Control Voltage (0-5V)
3Red / WhiteRun (Ground to stop)
4OrangePower good signal
5N/C
6N/C
7N/C
8N/C
9GrayPower switch (other leg to Switch Common)
10WhiteVolume Up switch (other leg to Switch Common)
11Blue / WhiteVolume Down switch (other leg to Switch Common)
12PeriwinkleReset switch (other leg to Switch Common)
13BlackSwitch Common
14RedLED Common (Anode)
15VioletRed Status LED (Cathode)
16Violet / WhiteGreen Status LED (Cathode)
17Orange / WhiteSpeaker (terminal 1)
18Green / WhiteSpeaker (terminal 2)
19N/C
20N/C

To aid in Indy power supply testing, I have made a small breakout board. I may release or sell this design (if there is interest) in a further update. More Indy PSU information to follow!
 
Last edited:

Elf

Storybook
Feb 4, 2019
56
8
8
Corrected the part about the reset switch being the system status indicator LED. It is actually the small nub below the power switch. Whoops! Not sure why I remembered it that way. o_O
 

Elf

Storybook
Feb 4, 2019
56
8
8
Here is some information about the performance of Indy power supplies, based on an arbitrary sample of one Sony and one Nidec. However, just based on previous observations these seem to be fairly representative of their kinds. I will break this up amongst multiple posts due to attachment limits.

First, some specifications on the Sony, and then some measurement information:
RailRating
+3.3V7A
+5V25A
+5V standby0.02A
+12V4.5A
-12V0.75A

Electrically the Sony seems to be the better supply versus the Nidec, but it is still not great by modern standards as can be seen by its performance under various load conditions:
Load / StatePower FactorWatts (True power)3V3 rail
volts/ripple
5V rail
volts/ripple
12V rail
volts/ripple
-12V rail
volts/ripple
Soft power-off-0.48543.267W0/00/00/00/0
No-load-0.46086.174W3.328V, 124mVpp5.28758V, 84mVpp12.5171V, 80mVpp~ -12V, 70mVpp
5V @ 2A load-0.505317.807W3.316V, 156mVpp5.13511V, 114mVpp12.5067V, 158mVpp~ -12V, 134mVpp
5V @ 10A load-0.611264.69W3.271V, 274mVpp4.88171V, 128mVpp12.4618V, 196mVpp~ -12V, 164mVpp

After some previous experiences loading down all rails with resistors, or trying to load down the 3.3 or 12V rails first, I settled on a testing methodology of just using a programmable DC load to load down the 5V rail in constant current mode. This seems to keep the supply happy, as loading down other rails without loading down 5V seems to stress it. Additionally it was too cumbersome to keep attaching power resistors on the other rails and I only have one programmable load. So, while this is not an optimum methodology and likely leads to higher ripple figures on the unloaded rails, it still seems to give a reasonable representation of power supply performance.

Sony: Test rig & Soft power-off state
56

Sony: No-load
7811

Sony: +5V @ 2A state
91012
 
Last edited:

Elf

Storybook
Feb 4, 2019
56
8
8
Now, some specifications on the Nidec, followed by the same type of measurements.

RailRating
+3.3V7A
+5V25A
+5V standby0.001 A
+12V4.5A
-12V0.75A

As you can see, the power supply rails are all rated the same between the Sony and the Nidec, except for +5Vsb standby power. This (along with the low rating) implies to me that nothing substantial is powered on the motherboard during standby. Most likely some extremely minimal logic in combination with the front panel power button, but certainly not any part of the processor or likely not even a microcontroller.

Load / StatePower factorWatts (True power)3V3 rail
volts/ripple
5V rail
volts/ripple
12V rail
volts/ripple
-12V rail
volts/ripple
Soft power-off-0.42363.559W0/00/00/00/0
No-load-0.663317.309W2.806V, 202mVpp5.05336V, 144mVpp8.3252V, 308mVpp~ -12V, 224mVpp
5V @ 2A load-0.694334.324W3.350V, 200mVpp5.03092V, 162mVpp11.9587V, 154mVpp~ -12V, 194mVpp
5V @ 10A load-0.720286.26W3.309V, 266mVpp4.93535V, 340mVpp11.9165V, 374mVpp~ -12V, 452mVpp

Nidec: Test rig & Soft power-off state
1819

Nidec: No-load
212022

Nidec: +5V @ 2A state
232425
 
Last edited:

Elf

Storybook
Feb 4, 2019
56
8
8
Nidec: +5V @ 10A state
262728

Now we go back to the Sony for some rail sequencing information:

Sony: No-load Rail Sequencing
29

Sony: 5V @ 2A loaded Rail Sequencing
30

I thought about trying to grab rail sequencing information with full load (or with the supply attached to a working Indy) however as you can see, putting a load on the +5V rail did not affect much.

The power supply turns on relatively slowly in the following order:
Start orderFinish orderRailRamp time
14+5V16.81576ms
23-12V10.78516ms
313.3V8.13464ms
42+12V9.1609ms

As you can see, at least for these particular samples, the Nidec holds regulation better under load versus the Sony but is also a much noisier supply. Is it because the capacitors are going? No idea, but it is not the only Nidec I have that behaves that way. The no-load ripple of the 12V rail certainly stands out!

True vs. reactive vs. apparent power is a subject outside the scope of this post, but in simple terms the closer the power factor is to 1 (ideal), the fewer amps it takes to deliver the same amount of real work (watts). Power factor can be considered a secondary factor in terms of efficiency, affecting things like the amount of loss in power transmission from the utility company to you, how many things you can put on a circuit breaker even though it may appear under capacity, etc. The closer something is to a purely resistive load (e.g. heater or incandescent lightbulb), the closer to 1 its power factor will be.

A power factor of 0.8 to 0.9 is pretty good for a switching power supply, but 0.6 is not so hot. The negative sign on the power factor as read by the power meter in the tables above means that the current leads the voltage (capacitive), so a power factor of -0.4854 should be read as "0.4854, capacitive" (rather than inductive like a motor load). Don't mistake negative readings as being further away from 1, just ignore the negative sign.

On the subject of efficiency in general, the supplies are pretty bad:
Power supplyStateWatts / VA inWatts outReal power efficiencyApparent power efficiency
NidecSoft power-off3.559W, 8.402VA0W0%0%
NidecNo-load17.309W, 26.10VA0W0%0%
Nidec5V @ 2A34.324W, 49.44VA10.06W29.31%20.35%
Nidec5V @ 10A86.26W, 119.8VA49.35W57.21%41.19%
SonySoft power-off3.267W, 6.731VA0W0%0%
SonyNo-load6.174W, 13.40VA0W0%0%
Sony5V @ 2A17.807W, 35.24VA10.27W57.67%29.14%
Sony5V @ 10A64.69W, 105.8VA48.82W75.47%46.14%

The Nidec has a slightly better power factor than the Sony, but both are poor overall. The Sony is more efficient than the Nidec and seems to do better with both lighter and heavier loads, but again neither are great and from certain angles they are really quite bad! Efficiency and power quality alone are enough reason to develop a new Indy supply, even if reliability weren't an issue. With issues like the Sony supply's fan, heat buildup in the Indy can be problematic and an inefficient power supply that just throws away 25-70% of power away as heat doesn't help.

How attached are you to your Indy power supply? I am not very attached to mine after this!

Both draw a little over 3W when the machine is powered off but plugged in.
 
Last edited:

Elf

Storybook
Feb 4, 2019
56
8
8
I put in a bit of work today to solve the last remaining Indy PSU mystery: the temperature sensor. It is not quite what I thought before, but not far off either.

Part numbers and analysis come from a sample of one CPU, SGI P/N 030-8201-001. The temperature sensor itself is a glass package (looks like a diode w/out a band marking) NTC thermistor labelled "U1" residing on the Indy CPU module. The thermistor is involved in a voltage divider bridge feeding a set of opamps (LM358AN, U6) which, minus a few capacitors (used for stabilizing the output), looks like:
32

The output of of U6:1 feeds through the top (closest to board edge) CPU connector on the 3rd pin from the left of the "41" marking, then goes through the motherboard directly to the "Temp" pin (2) on the Aux connector to the power supply. This outputs what is essentially a 0-5V signal that controls fan speed.

I was able to quantify this with some testing. First, of the temperature sensor itself and then of the output of the circuit, using hot air and a thermocouple probe:
333536

Indy CPU temperature sensor characteristics:
3437

As you can see, maximum output is reached around 180F (82C), and minimum around 60F (15.5C), from the motherboard to the power supply. All in all, that is a fairly sensible range for a control signal used in setting fan speeds.

What the power supplies do with it, though, is a little different between Nidec and Sony. The next test consists of feeding a control voltage in from a lab power supply to both types of PSUs to see what they do with the fan.

Sony under test:
3839

Neither one provides a PWM signal to the fan; both vary the DC voltage supplied to the fan to set its speed. The fan intakes air from the Indy through the side of the power supply and blows it out the rear of the power supply near the area where AC line power comes in.

42


The difference between the Nidec and Sony fan control inside the power supply (both of which receive the same control signal voltage from the motherboard) is quite substantial.

The Nidec acts linearly, with the fan spinning slowly even at 0V of control signal (60F, 15.5C) and ramping up to maximum speed at 5V of control signal (180F, 82.5C) which puts the fan at 11.87V maximum.

The Sony does not act linearly, with a 3rd order polynomial fit shown on graph. The voltage output by the Sony fan controller does not allow it enough torque to start spinning at all -- and only minimally -- until 2.17V of control signal, or in other words 113.1F / 45C in the vicinity of the CPU! It will need to be much hotter than that (e.g. 145F / 62.8C) before the fan really starts to blow. Interestingly the Sony fan has a higher maximum voltage, with a 5V control signal topping out at 15.76V to the fan, which at that point is going pretty fast.

This is almost certainly what leads to the myth that Sony Indy PSUs don't use their fan. I have even read one person say that the Sony PSU has no fan, which is visibly untrue. It just doesn't start moving any appreciable amount of air until your Indy is already way too hot! Why they designed it this way, and continued to produce them without fixing what seems like an obvious defect, is a mystery to me.

So there you have it; with that last mystery solved this should tell you all you need to know about Indy PSUs from an external perspective. I will likely not put any more time into reverse engineering them as I think a replacement is not only an easier, but also a much better choice. With the existing options, you can either have:
  • Nidec - An electrically noisy, extremely inefficient PSU that keeps the Indy cool, or...
  • Sony - A slightly more efficient, less noisy PSU that will cook the machine
While the Nidec is known for killing capacitors and diodes and the Sonys are supposedly reliable, more than one recent report (and an experience on my side) has the Sony PSUs dying with shorted switching MOSFETs which take out a lot of other components as well. Lots of folklore out there about the Indy PSUs and much of it untrue, it would seem!
 
Last edited:

Elf

Storybook
Feb 4, 2019
56
8
8
A bit of an addendum regarding the "Power Good" signal:

Nidec Pgood vs. loaded 5V rail:
5253

Sony Pgood vs. loaded 5V rail:
54

Sony Pgood loaded w/ 680Ω vs. loaded 5V rail:
55

Once again another difference between the Nidecs and the Sonys. The Nidec asserts Pgood around 3.3V, long before the 5V rail stabilizes. The Sony apparently thinks it is always good and ties Pgood via a significantly high resistance path to +5Vsb. As can be seen, loading it with just 680Ω drops the voltage down significantly. This behavior is not present on the Nidec.

Between this and the fan, it always seems like the Sony design doesn't quite hit the original spec.

The Nidec Pgood is extremely noisy! It might need capacitor replacements?
 

Str1kernaut

New member
Jun 19, 2019
1
1
3
I have a question: the new PSU for the Indy, will be compliant with ATX specs for ripple and regulation? A PSU with an high ripple output can cause damage to the hardware
 

onre

Administrator
Feb 8, 2019
59
18
8
Getting within (and well beyond) the ATX spec with modern DC-DCs is trivial.
 

Elf

Storybook
Feb 4, 2019
56
8
8
@Str1kernaut it should exceed them, but prior to releasing it I will put it under full load and post how it performs. Note however none of the original power supplies do too well :p
 
Last edited:

callahan

New member
Jul 20, 2019
7
3
3
Elf -- I've been making some improvements to the higher intellect wiki in areas where I have insight and suit my interests. Any heartburn if I incorporate your findings there, hopefully to preserve info in a readily accessible form?
 

callahan

New member
Jul 20, 2019
7
3
3
Of course. You can see some of my recent edits here:


 
Reactions: Elf