SGI Power Series Troubleshooting Guide

The POWERpath architecture: each CPU and FPU pair has its own caches on the shared MPlink bus (SGI, POWER Series Technical Report 1989, Figure 1)

Fault finding on the SGI Power Series. SGI's diagnostic documentation for the range is not available online, so the checks below draw on A. J. Corda's This Old SGI FAQ, Gerhard Lenerz's sgistuff.net and jan-jaap's restoration of a 4D/380 rack on the IRIX Network forums. Corda's warning applies throughout: some of his notes come from old Usenet posts, and he advises verifying everything with a meter or oscilloscope.

Order of work

1. Make the power supply safe and sound first; see SGI Power Series Capacitor Replacement Guide. jan-jaap tests a rebuilt supply with an electronic load or a bank of power resistors before connecting it to the backplane.
2. jan-jaap's next step was to power the backplane with no boards fitted. On his 4D/380 this is when a tantalum capacitor near the keyboard port on the I/O bulkhead burst into flames, so inspect the bulkhead and backplane capacitors before this test.
3. Boot a minimal system: one I/O board, one CPU board and one MC2 memory board, with a serial terminal as the console. jan-jaap's 4D/380 first came back in this form, with an IO3, one IP7 and one MC2.
4. Add the other boards one at a time, checking the status indicators after each.

Status indicators

Power Series status indicators

Where	Indication	Meaning
CPU board LEDs	Cycling	Power-on tests running
CPU board LEDs	Only the second LED from the bottom lit on CPU 0	Power-on tests passed (primary CPU)
CPU board LEDs	All LEDs of the other CPUs blinking	Power-on tests passed; system in the PROM monitor
MC2 board	Four red LEDs	Should be off once the power-on tests have passed
MC2 board	Eight green LEDs	SIMM groups fitted and accessible
Twin tower one-digit display	F	Error
Twin tower one-digit display	Cycling 1 and 2	In the PROM monitor
Twin tower one-digit display	Cycling 0 and 1	Running
Single tower yellow fault LED	On, then off	Lit with the green power LED at switch-on; goes out when the power-on tests pass

On dual-processor boards the lower set of eight LEDs shows CPU 0 and the upper set CPU 1. Corda was unsure of the CPU LED meanings and found no documentation for the codes on the seven-segment status display.

Single-tower status panel

The single tower hides its status panel behind the small black panel at the top right of the front, which carries the power and reset switches. Remove the screw on its left edge (the drives in the top two bays may also have to come out) and lift the panel off; a circuit board attached to it is connected by a ribbon cable. Rest it somewhere safe and non-conductive if the system is powered with the panel off. The board carries:

• the seven-segment status display and the eight-position configuration switch;
• test points for +12 V, −12 V, +5 V and −5.2 V;
• over-voltage and under-voltage LEDs for each of those rails;
• an over-temperature LED.

Measure the rails at the test points before suspecting the boards.

Mode switches

The eight "Mode" DIP switches, near the power switch on the twin tower and on the status panel board of the single tower, are not covered in public SGI documentation. sgistuff.net quotes an old Usenet post for some settings, with all switches closed as the default:

Mode switch settings (from a Usenet post quoted by sgistuff.net)

Switches 8 to 1	Hex	Function
00000000	00	Default
10000001	81	PON mode: drops into the pon> prompt
10000010	82	SYMMON (function unknown)
10000100	84	NODIAG: skips the power-on tests
10001000	88	MPDEBUG (function unknown)

jan-jaap says the mode switches can be used to debug hardware problems.

CPU and PROM faults

• A dead primary CPU makes the whole system misbehave. jan-jaap's 4D/380 arrived with two faulty CPUs, one of them the primary, and with mismatched PROMs. Move a known-good board into the first CPU slot to test.
• All CPU boards should carry the same boot PROM version, and the PROMs on the IO board must match the CPU PROMs.
• An IP7 is a carrier for two processor modules, each with an R3000, an R3010 and cache. jan-jaap notes that modules and CPU chips can be swapped between boards, so one working IP7 can be built from two faulty ones.
• Each CPU board has a card-edge reset switch that resets the processors on that board.
• To boot from CD-ROM the machine needs the last PROM version, which was built from the IRIX 4.x code.

Power supply faults

• Most SGI Power-One SPM5 supplies jan-jaap has seen had a failed MOSFET in the bias supply, the small switcher built round a UC3844 controller that runs whenever the supply is connected. When he damaged his own while probing, the fuse had also blown although it looked intact; measure the fuse, do not judge it by eye.
• The SGI version of the SPM5 keeps its outputs off until the inhibit line on its 12-pin control connector is pulled to ground. jan-jaap fits a wire link to switch a supply on outside the system. An extra board in the SGI version appears, from his tracing, to switch the outputs off if the supply overheats.
• Switch-mode supplies often need some load before they regulate properly; test with a load, not open-circuit.
• This Old SGI records a fault on the LH Research supply of the Professional IRIS, which some twin-tower revisions may also use: disks gave errors while a multimeter showed correct voltages, and an oscilloscope showed the 5 V rail swinging between 4.5 and 5.3 V. The cause was one of a pair of 2900 µF 200 V capacitors on the input side.

For bench-testing a Power-One supply, see SGI Crimson Troubleshooting Guide#Bench-testing a Power-One supply.

Overheating

The system switches itself off when its thermal sensor reaches 140 °F (60 °C). It will not restart until it has cooled and the thermal reset switch, behind a small hole on the front, has been pressed. Check the fans or blowers and the air path before resetting it.

Memory faults

• An MC2 whose red LEDs stay on after the power-on tests has a fault or a SIMM problem.
• Old MC2 revisions, including board 030-0117-001, do not work with 8 MB SIMMs.
• SIMMs go in groups of four, starting at the edge nearest the status LEDs, and officially one MC2 holds only one SIMM size.
• On a 4D/210 the IP9's own SIMM sockets are only used once any MC2 fitted is full.

Graphics faults

The GTX, VGX and VGXT graphics sets have a DB9 serial port. With a terminal attached at 9600 baud, 8 data bits, no parity and 1 stop bit, the GTX reports its own self-test results and offers its own PROM monitor; Corda had not tried this on a VGX. jan-jaap notes that the VGX GM3 graphics manager has a Motorola 68000-family processor that can be reached through the DB9 on the front of the board. A VGX runs with one raster manager board (5-span) or two (10-span).

Clock and NVRAM

If the date resets or the PROM settings are lost at power-off, suspect the SmartWatch module on the IO2 or IO3 board, which has a built-in lithium cell. Corda notes that on the Professional IRIS the PROM monitor recognises corrupted NVRAM and still lets you enter the monitor to correct it; he does not say whether the Power Series PROM behaves the same way.

Operating system

IRIX 5.3 is the last release for the Power Series. RealityEngine graphics needs 4D1-4.0.5 or later, and later boards and graphics are not supported by the 4D1-3.x releases.

Related pages

• SGI Power Series
• SGI Power Series General Maintenance
• SGI Power Series Capacitor Replacement Guide
• SGI Crimson Troubleshooting Guide
• Capacitor Failure Symptoms