Indigo 2 Power Supply Basics

[Elf]

As you may have seen, I previously did a writeup on Indy power supplies. I've recently examined the SGI Indigo 2 PSU, and while I won't be exploring it in the same level of detail as the Indy I wanted to share some of the basics.

If referencing or duplicating this information, I ask that you maintain a prominent link to this original forum post for further updates, as well as a credit by name.

Non-Impact
Power is supplied over two 18-pin single row 3.96mm connectors, with non-keyed male pin header on the motherboard side. I will refer to them as the Left Hand and Right Hand connectors from the perspective facing the power supply, as outlined in the picture below.

There is an additional AMP / TE Mate-N-Lock (a.k.a. disk drive "Molex connector") supplying +5/+12V for the disk backplane. The pin configuration on this is standard, as on modern PCs.

The Left Hand connector pins are as follows:

Pin #	Wire Color	Function
1	Black	0V
2	Black	0V
3	White	"MH" / Motherboard Higher (?) - raises 5V rail by 0.3V
4	White	"ML" / Motherboard Lower (?) - lowers 5V rail by 0.2V
5	Red / White	+5V standby
6	Red	+5V
7	Red	+5V
8	Red	+5V
9	Red	+5V
10		N/C
11	Red	+5V
12	Red	+5V
13	Red	+5V
14	Red	+5V
15	Black	0V
16	Black	0V
17	Black	0V
18	Black	0V

The Right Hand connector pins are as follows:

Pin #	Wire Color	Function
1	Black	0V
2	Black	0V
3	Black	0V
4	Black	0V
5	Yellow	+3.3V
6	Yellow	+3.3V
7	Yellow	+3.3V
8	Blue	-5V
9	White	Power good signal
10	Violet	-12V
11	White	Run (Ground to stop)
12	Orange	+12V
13	Orange	+12V
14		N/C
15	Black	0V
16	Black	0V
17	Black	0V
18	Black	0V

The model number under test is an ITT PowerSystems PEC4074, with the following output ratings:

Rail	Rating
+5V	40A
+12V	12A
-12V	1A
+3.3V	12A
-5V	1A
+5V standby	1A

The Indigo 2 power supply (like the Indy PSU) is Ground to Stop on pin 11 of the Right Hand connector. This means that properly functioning Indigo 2 (non-impact) power supplies will start up on their own if unplugged from the motherboard and given AC line power. However they will want at bare minimum load on their +5V rail to enter proper regulation. Be advised that testing an Indigo 2 power supply without load is unlikely to yield meaningful results.

The +5V rail can be increased in voltage (+0.3V) or decreased in voltage (-0.2V) by a register on the SGI HPC3 PBUS, interacting with the Left Hand pin 3 (MH) and pin 4 (ML). These are fixed voltage increments controlled fully by software, with no analog voltage sense feedback. See post #6 for more information.

The SGI Vault M is an external SCSI disk enclosure based on the Indigo 2 chassis, and uses the Indigo 2 PSU to power the disks inside. It achieves PSU stability by presenting 1 Ohm of resistance on the +5V rail, for 25W of power draw. This is the notorious "heater cage" in the enclosure, which is also present in the picture above.

Standby power usage was 5.5622W when switched soft-off via the Run line.

Like the Indy power supply it appears to have no power factor correction, and exhibited a dismal -0.57 (capacitive) Power Factor under load.

 

[Elf]

Next up will be to figure out the unknown functions of pins 3, 4 on the Left Hand connector, pin 9 on the Right Hand connector, and addition on the Impact supply. I will update the post accordingly.

Impact
The Indigo 2 Impact power supply is similar in terms of external interface to the non-Impact supply, but has substantially different internals. Covering the internal differences is beyond the scope of this article on PSU basics, but the externals are outlined below.

The main Left Hand / Right Hand motherboard power connectors maintain largely the same functionality, but there are some differences in wire color. The Left Hand connector pins are as follows:

Pin #	Wire Color	Function
1	Black	0V
2	Blue	0V (sense?)
3	White	"MH" / Motherboard Higher voltage request
4	White	"ML" / Motherboard Lower voltage request
5	Yellow	+5V standby
6	Red	+5V
7	Red	+5V
8	Red	+5V
9	Red	+5V
10		N/C
11	Red	+5V
12	Red	+5V
13	Red	+5V
14	Red	+5V
15	Black	0V
16	Black	0V
17	Black	0V
18	Black	0V

The Right Hand connector pins are as follows:

Pin #	Wire Color	Function
1	Black	0V
2	Black	0V
3	Black	0V
4	Black	0V
5	Yellow	+3.5V
6	Yellow	+3.5V
7	Yellow	+3.5V
8	Blue	-5V
9	Yellow	Power good (always high)
10	Purple	-12V
11	White	Run (high - run / low - stop)
12	Orange	+12V
13	Orange	+12V
14		N/C
15	Black	0V
16	Black	0V
17	Black	0V
18	Black	0V

The Indigo 2 Impact PSU maintains the secondary +5/+12 "disk drive" connector as before, but adds another +3.5V connector and sense connection for the graphics riser.

The Graphics Power connector comes on a blue, two-conductor "ribbon cable" (with extraordinarily wide flat conductors for power). It is terminated by a connector made by Amphenol, notated by sticker as AMP P/N 765950-1. However, this part number does not appear to be in any searchable AMP catalog. It is unclear whether the part number refers to the connector or to the cable assembly as a whole, and the connector does not look visually similar to any other connector in their current lineup. It is possible that the part has been long obsoleted (which seems more likely given the multi-pin to two-conductor arrangement), or was a custom design (which Amphenol offers as a service, but would likely result in a more form fit design).

The Graphics Power connector pins are as follows:

Pin #	Function
1-6	+3.5V
7-12	0V

The Graphics Power Sense connector is a standard keyed 0.1" pin header connector, with pins as follows:

Pin #	Wire Color	Function
1	Black	0V sense
2	Yellow	+3.5V sense

The model tested is a Zytec P/N 22928800 (SGI P/N 060-8001-001) The specifications of the power supply differ from the non-Impact supply:

Rail	Rating
+5V	40A
+12V	4.25A
-12V	0.5A
+3.5V	12A
+3.5V	36A
-5V	0.8A
+5V standby	0.1A

Note the significantly reduced +12V rating, and the significantly increased +3.5V rating, as well as a confirmed voltage increase from the previous +3.3V of the non-Impact supply.

Starting the Impact supply is different from the non-Impact supply. Rather than turning on by default, Pin 11 (Run) of the Right Hand connector must be brought logic-high via the +5V standby rail. Additionally, the +3.5V/0V Graphics Power Sense lines must be connected to the +3.5V/0V rails. This is performed either through the riser, or via a "termination" PCB attached to the exterior of the supply.

The "termination" PCB provides no active functionality and no load on the rail. It simply connects the power rails to the sense lines. Without this connection, the supply will shut down.



Lastly, the Power Good signal on the Zytec / Impact supply is always asserted, regardless of power rail regulation state, and even when the supply is in standby. Not very useful!

 

[Elf]

Pin 9 of the Right Hand connector was easy to figure out; it is the Power Good signal.


(C1 - Pgood, C2 - +5V, C3 - +12V)

Pin 3 and 4 of the Left Hand connector will take some more investigation. My best guess at the moment is that they are two distinct remote sense lines or some other sort of feedback or control signal. If loaded down they make the power supply quite unhappy, although it maintains rough +5/+12 regulation. They are not connected to each other although they float at the same voltage.

Edit / addendum:
The motherboard appears to have diodes from the +5V rail and 0V to pin 3/4 of the Left Hand connector like so:
+5 ⇥ Pin 3 ⇤ 0V
+5 ⇥ Pin 4 ⇤ 0V

This is just speculation due to behavior; I have not traced it out or located the actual diodes. Note that Pin 3 / 4 are not directly connected, even on the motherboard. Oddly that leaves the diodes reversed biased, and in a running Indigo 2 these pins still continue to float at ~14V.

Pretty weird! I will continue to look further.

 

[Elf]

I am continuing to investigate the mystery of Left Hand pin 3 and 4, and it gets into the oddities of the oldschool PWM switching controller.

Both pin 3 and pin 4 make their way directly to the switching controller board, of the three boards in the I2 power supply: high voltage, low voltage, and controller. The pin headers on the low voltage side of the controller board are labelled with abbreviations in a semi-helpful way, and many of them make sense once you trace them back.

Pin 3 and 4 go to controller inputs labelled "MH" and "ML." It is not clear to me yet what these meanings are, but tracing things back through a lot of discrete components (ouch, '90s switching PSU design) they are pulled up via resistors to the switching controller's power. This power comes off a toroidal transformer on the high voltage board through very thin wires, and is zener regulated on the controller board, explaining why these lines float at a higher voltage than the +12V output rail.

They then go to a series of 2N7000 mosfets, one set up to pull down normally, and the other set to invert (not pull down). These then make their way through a series of resistors to one of the op-amps involved in the UC3825 regulator circuit.

Because of the mosfets in the path I no longer believe these are voltage feedback lines but perhaps some sort of "overvoltage" (MH) and "undervoltage" (ML) indicators? Or maybe a high power or low power mode for when more peripherals are added and they want to change something about how the PWM is done? Whatever it is, these are an on/off affair. I think I will have to pull a motherboard to really know.

 

[Elf]

So with another misspent evening I think I finally have the what, if not the why...

My previous guess about high and low voltage signaling from the motherboard appears to have been correct. I am going to assume that "MH" on the control board means "Motherboard Higher" and "ML" means "Motherboard Lower" (Voltage).

This one took some doing; I had to trace out parts of the 4 layer through hole (!) control board.

Partial schematic below:

(Resistor values are just measured approximate values; I did not take time to decode the bands)

As you can see, Left Hand Pin 3 (MH) is set up with two MOSFETs that invert; Q306 is normally off. If LH Pin 3 is pulled down to 0V by the motherboard, R320 will enter into the divider network, causing a 24k pulldown.

Left Hand Pin 4 (ML) is set up with one MOSFET, normally on, with R321 in the divider network causing a 27k pulldown. If Pin 4 is pulled down to 0V by the motherboard, R321 will drop out of the divider network. This is a binary arrangement on either line; the resistors are either in or out of the resistive divider.

The +5V rail output of the PSU is on the high side of the divider. The output of the divider network eventually makes its way to Pin 1 of the UC3825N PWM controller, which is at the heart of the switching operation of the power supply. Pin 1 is the negative feedback line of the Error Amplifier, which sets the PWM duty cycle and thus the regulation of the output voltage.

I put this through some rather invasive tests with a somewhat sickly extra I2 PSU I had on hand:


As you can see, the PSU appears to operate on a ~ 130kHz PWM frequency, with the duty cycle varying according to load:


The gist of it is, when the motherboard pulls LH Pin 3 / "MH" down, it is asking for a higher voltage. The voltage appears to increase by about 0.3V on the +5V rail.

When the motherboard pulls LH Pin 4 / "ML" down, it is asking for a lower voltage. The voltage appears to drop by 0.2V on the +5V rail.

Due to the design of the supply I expect the increases and decreases would be proportionate on the other rails, but I have not tested this. I also don't yet know what (if anything) on the motherboard actually controls these lines, and will need to further examine an Indigo 2 motherboard.

 

[Elf]

Well, I had a small adventure in tracing things out on a dead Indigo 2 motherboard and finally resolved the complete hardware mystery of the higher/lower voltage signals to the power supply.

1. Left Hand Pin 3 (MH) / 4 (ML) off the power supply go to U16 (7406 inverter) - MH to inverter 5, ML to inverter 6. This is fairly close to the PSU inlet to the motherboard and was easy to find.
2. Inverter 5/6 from U16 pop out all the way across the motherboard to U98 (74FCT273 octal flip-flop register) - MH to bit / flip-flop 7, ML to 6. This is in a cluster of similarly used registers as well as the burned PROM for the system. This was hard to find!
3. The bits on these registers are set by U97, an Intel AWARD NB242WB which I can not find a datasheet for, or even a description of what it does. My guess is some sort of system management bus style controller.
4. The U97 Intel AWARD chip appears to be connected to... the PBUS of the SGI HPC3!

See the attached PDF page 24 for a general description of the PBUS, although it does not mention any of the Indigo 2 specific peripherals.

What this means is that these lines (to increase +5V rail voltage by 0.3 or decrease by 0.2) are fully software controlled, either by something in the PROM code or in IRIX. This is a fully digital on/off affair, it is not analog voltage sense feedback. Weird!

The whole arrangement looks something like this:

 

[Elf]

Obtained a spare Impact PSU for test, and so added a section on the Zytec Indigo 2 Impact power supply (edit to post #2 in the thread).

As a side note, the Indigo 2 Impact supply is more annoying than usual to take apart. Rather than having HV, LV, and controller boards, the Zytec Impact supply only has high voltage and low voltage boards, with the controller logic on the LV side. While physically annoying, the separation between HV and LV seems pretty clean and may bode well for it being somewhat repairable?

It also relies on unusually thick secondary coils (more like straps) from the magnetics of the HV side that are soldered directly to the LV side:


As described above, the startup mechanism is different. A +5V logic-high (from the +5V standby power rail) must be applied to the Run pin to start the supply.

It also appears to have either a combined 36 or 48 amps of +3.5V power (this bears further load testing)! The +3.5V rails supplied to the motherboard do not appear to be distinct in any way from those on the Graphics Power connector. That is to say, they share the same switching regulator.

Pin 2 of the Left Hand connector has been repurposed from a regular 0V return to what appears to be a 0V sense connection, and is now colored blue. Although it is connected internally to the 0V rail, it is by a distinct and much thinner connection on the board, and part of it seems to disappear elsewhere into the controller circuitry.

The Impact supply also appears to take the same "life is always good" philosophy as the Sony Indy power supply, and asserts its Power Good signal even when the supply is in standby, and regardless of the regulation state of any of the rails!

 

[Elf]

All I want are Amps for Christmas!

I performed some load testing on the Graphics Power +3.5V rail of the Indigo 2 Impact PSU. As said before, this 36A output shares its regulator with the 12A output on the regular motherboard power connector. They are just different physical connections to the same rail.

As can be seen below, it sustains output up to 40A (140W). The lower voltage displayed on the load is due to wire losses. At the Graphics Power Sense connector (on the "termination" board), it is maintaining 3.5V regulation as confirmed via multimeter. During these tests the +5V rail was loaded to 5A (25W) with a power resistor. I really need to get a second programmable load!

At 45A of load the supply went into soft shutdown due to overload (but did not blow any fuses), so it appears that the capacities between the two outputs (36A and 12A) are not fully additive, but close.

 

[XeTK]

Hi Elf,

Trying to diagnose a faulty Indigo2 Impact PSU and just wondering if you would happen to know what the voltage is that is sent from the HV side to the LV side.

This would be very helpful to us as we can then work out which side the issue lies on as we can substitute the HV side from a bench PSU.

Cheers

 

[Elf]

Certainly. Given the secondary windings of that transformer, with terminals as noted below, on an Indigo 2 Impact supply with 2.8 A being drawn from the 5 V rail only (no 3.5V or other load):

• L1 - L2: 28.62 Vac RMS
• L1 - L3: negligible voltage
• L3 - L4: 28.842 Vac RMS
• L2 - L4: 23.29 Vac RMS

I believe L1 - L2 are one secondary winding and L3 - L4 are another, but that is just a guess as I haven't pulled the transformer out of circuit or pulled it apart. As such the L2 - L4 voltage may not be very meaningful? But I could be wrong about that, so the meaning of these various voltages and the configuration of the transformer is ultimately an exercise left to the reader.

L1 - L2

L3 - L4

L2 - L4