I recently have had the (dis)pleasure of working on a PSU for an Silicon Graphics Indigo2 Impact workstation. The PSU was manufactured by Zytec, with an SGI part number of 060-8002-001, revision A according to the external sticker. It provides the following rails:

(Total power shall not exceed 385 W)

This power supply differs from the regular Indigo2 supply, as it has an extra 3.5 V supply rail to run the Impact graphics. This rail is broken out via a blue high-current ribbon cable. There is also a small black and yellow twisted sense cable, which must be connected near the power cable on the graphics riser board. If the Impact graphics option is not installed, the termination PCB (mounted on the side of the power supply) must be used to allow the PSU to operate correctly. Without voltage feedback for the Impact rail, it will not work as expected (I have not tested what it does. The rail either will not regulate correctly, or the supply will shut down).

Pin-out (and lots more valuable information) taken from here. Thanks, Elf.

This connector has the same pin-out as the usual PC “Molex” hard drive connector.

<insert pinout here> (3.5 V and Ground)

(Note: This is far from complete. It is based off my limited experience repairing one of these supplies, as well as my own research. I will attempt to cite any sources I used.)

This power supply can be divided into two sections (conveniently, the primary and secondary are each on distinct PCBs), as with most line powered switch-mode power supplies, and is of a forward converter topology. Additionally, it has a smaller (self-oscillating flyback?) supply which is used to provide the supply's (unregulated) 18 V standby rail.

The primary side of this power supply was designed around a UC3845 current-mode controller from Texas Instruments. This controller is used to drive the primary-side switching device for this power supply. However, it is being used in an unusual configuration, as it does not perform any kind of voltage regulation. There are two optocouplers (MOC8101) on the primary-side board. One of these is used to control the supply. When the secondary-side board wants to start the supply, a control signal is allowed to float. When this happens, the supply starts oscillating and powers up. The second optocoupler appears to be used for some kind of shutdown signal from the primary board to the secondary, but has not been investigated further. Additionally, while this supply is not power factor corrected, it does have automatic line voltage selection. This functionality is based around an AVS1AC automatic voltage selection switch. This device configures (via a TRIAC) the primary filters as a voltage doubler when used on 120 V, or as a regular rectifier when used on 220 V.

There are several connections between the primary and secondary PCBs in this power supply, split across two connectors and four cables. First, there are the two sets of high-current cables which exit the forward converter transformer directly. These are the square-wave AC outputs, and are (likely, I have not specifically confirmed) used to derive the 5 V and 3.5 V rails of the supply. These connections are soldered directly to the secondary-side board, with a fusible link at the end.

Next, there is a four-pin Molex Mini-Fit Jr. connector. This connector also carries AC square-wave output, at the lower currents required for the +12 V, -12 V, and -5 V rails. It has a connector on the secondary-side PCB end of the cable.

Finally, there is a six-pin control cable. It has a reddish-brown connector on both ends. It carries the standby voltage (+18 V, red wire) and two control signals (yellow wires) from the primary-side to the secondary-side PCB. There are also three grounds (black). One yellow wire provides an isolated power-on signal, and the other is a control signal from the primary-side to the secondary-side of unknown purpose, as previously described.

This board is where most of the complication of this supply lies. As the supply is a forward converter topology, there are additional magnetics on the secondary side, and the supply is entirely secondary-side regulation.

The simplest rail of the supply is the 5 V Standby supply. It is regulated from the 18 V supply via a LM340T-5.0 regulator. The 5 V from the regulator output is provided to the computer, but is also used for operating the control logic on the power supply itself.

The negative voltage rails are the next simplest rails. They are provided using a pair of MC34166 inverting step-down converters. These are self-contained and fairly straightforward to troubleshoot. The circuit appears to be very similar to the typical application found in the datasheet.

Unlike most conventional switch mode supplies, this supply is entirely regulated on the secondary side. Additionally, it is done in a rather unconventional manner, using magnetic amplifiers. Magnetic amplifiers use a special kind of inductor core as a switch by saturating and unsaturating it. This can be done with a (relatively) small switching transistor, unlike a conventional buck regulator designed for the large currents of this power supply. The circuit for the 12 V rail appears to be the simplest, and follows a design very similar to this application note. While the circuitry is similar, each regulator is built around a discrete LM358 op-amp, diodes, and switching transistor.

While each voltage rail is individually regulated using a mag-amp, all rails are monitored and faults are latched to shut down the supply. A UC2903 is used to monitor the voltage rails, and shuts down the supply if over or under-voltage conditions are detected. Faults appear to be latched using an overengineered mess of NAND gates, RS latches, and comparators.

NOTE: This power supply has deceptive soft power circuitry, and does not work like a normal PC power supply. To turn on the supply, the power on pin must be driven logic high (5 V), and floating appears to shut down the supply correctly. However, floating the pin does not properly reset the circuit and allow the supply to operate. When bench testing, you must use a strong pull-down (100 Ω works) or drive the pin with a push-pull output to get the supply to start. Many, many hours were lost to diagnosing this “issue”.

As this unit was received, it would not attempt to power up the machine or start. Measuring the standby rail with the unit removed from the computer showed that it measured 5 V as expected. Attempting to jumper the power-on pin to the 5 V supply would cause the fan in the supply to twitch briefly, but the supply would not start, with or without load.

End view of supply, showing internal arrangement of PCBs.

To disassemble the supply, the external screws were removed. With the screws removed, the supply separated into two halves, one with the primary PCB attached and one with the secondary PCB attached. These boards can be removed from the halves of the chassis by removing the 5 screws that hold them in place. There are also two screws attaching the power connector and filter assembly to the back of the chassis. They must be removed to detach the primary PCB from its part of the chassis.

Disassembly showed significant signs of corrosion, primarily around the large, low voltage electrolytic capacitors used for filtering the high current rails. In this supply, these were Panasonic FA series capacitors. Removing these showed that they were all leaking.

Replacing all of these capacitors did not restore operation to the supply. It was determined that one of the operational amplifiers for the mag-amp that regulated the 12 V rail was missing entirely. This chip was located directly underneath a large, leaky electrolytic capacitor, and likely fell off the board at some point during disassembly, though it was never located. Additionally, a resistor in the same area was also missing. The LM358 op-amp and the resistor were replaced using photos from this thread, which shows corrosion in a similar location, though not the same components.

Low-voltage board on partially disassembled supply.

After replacing these components (and clearing an accidental solder bridge) along with all electrolytic capacitors in the supply (the large FA series capacitors were the only ones obviously defective, but all should be replaced), the supply would power up and all rails were within appropriate specification when jumpered to power up and plugged in. However, the supply did not seem to start if shut down and reconnected. It was at this point much time was wasted determining that this was expected behavior, as the computer likely drives the power control input with a push-pull output.

It was also determined that the supply can be forced to power up by removing the yellow wire closest to the red wire from the control cable. This disables the power-down circuitry, causing the primary side to start switching. This results in the supply running with no protection of any kind. Before the mag-amp circuitry was rebuilt, this resulted in approximately 22 V being present on the 12 V rail. This resulted in the failure of a diode in another portion of the control circuitry. It is not recommended to operate the supply in this manner, especially not connected to anything other than dummy loads. This over-voltage protection is likely why the supply would attempt but fail to start before repair. It is likely that the rail voltages began to rise before the 12 V rail would trip the over-voltage protection of the UC2903, and the supply would shut down.

It should also be noted that this power supply is difficult to rework, especially around the section with the primary low-voltage filters. This section of the board has several large power planes, and the four-layer board requires a lot of heat to clear the holes from the capacitors. This problem is exacerbated by corrosion of the solder joints. Multiple pads were damaged during the repair process. Avoid repeated rework if possible.

Main low-voltage capacitors removed, revealing corrosion.

(TODO: add capacitor dimensions, values for primary filters)