SGI Octane Troubleshooting Guide

An Octane with the bezel door open. The power and reset buttons and the lightbar sit behind the door (afrank99, CC BY-SA 2.0)

Fault-finding on the SGI Octane, from the lightbar diagnostic charts and test procedures in SGI's Octane Workstation Owner's Guide (007-3435-004) and the Octane_ide Reference Guide (108-0217-002), with repair reports from owners on the IRIXNet forum. SGI's procedures stop at the replaceable module: power supply, system module, CPU module, DIMMs, XIO and PCI boards, frontplane. Everything below that level comes from community repairs.

Order of work

1. Note three things at power-on: the colour and behaviour of the lightbar, whether the fans run, and whether the boot tune plays. The charts below start from these.
2. Connect a terminal to serial port 1 at 9600 baud, 8 data bits, no parity, 1 stop bit, through a null-modem cable. The PROM prints memory errors and other start-up messages there even when nothing reaches the monitor.
3. Strip the machine down. Remove the PCI module and every XIO option board, and test with one graphics board. One owner has seen an Octane refuse to boot with a damaged video option board fitted.
4. Swap modules with a known-good machine where you can. The power supply, system module, XIO module and frontplane are each self-contained, and owners on IRIXNet isolate faults by swapping them between machines.

Lightbar diagnostic charts

SGI's three charts, combined into one table. Each chart ends by telling the owner to contact an SGI service provider with the system serial number to hand.

Lightbar symptoms (SGI owner's guide, Figures 8-1 to 8-3)

Symptom	SGI's possible causes	SGI's action
No LED, no fan, no sound	No power to the system; power supply failure	Check the power connections and the LEDs on the small power supply; replace the power supply
No LED, fan running, no sound	System module not seated; CPU failure; frontplane failure	Reseat the system module; replace the system module or CPU
Solid red LED	XIO module or graphics board not seated; graphics board failure; monitor cable unseated	Reseat the XIO module, XIO board and monitor cable; replace the graphics board
Solid red LED, no system drive	System drive not fully inserted; system drive failure	Check that the drive handle is locked down; record any error message; replace the system drive
Blinking red LED	Memory diagnostic failure; failure of the second CPU on a dual module	Check that DIMM sockets 1 and 2 are filled, reseat the DIMMs, check that each pair matches in density and type; replace the DIMMs or the dual processor module
No boot tune	Headphones plugged in; speakers unplugged or faulty	Unplug the headphones and connect the speakers; replace the system board, chassis or speakers
Solid white LED	Diagnostics passed	Any message on screen names the part to replace
Solid white LED, no display	Monitor off, unplugged, in power saving or turned down; cable, monitor, system board or graphics board failure	Check the monitor, its cable and its LED; replace the monitor cable, monitor or system board
Solid white LED, keyboard or mouse prompt on screen	Keyboard or mouse not connected or faulty; system board failure	Check the connection; replace the keyboard or mouse, then the system board
White LED for more than 2 seconds, then red flashing for about 10 seconds, then white flashing for about 10 seconds	The system is recovering its flash PROM	If it recovers, the flash PROM has been reset automatically. If the red LED keeps flashing slowly at 2-second intervals the flash PROM has failed: power off and replace the system module

The guide does not explain what "the LEDs on the small power supply" are, and owners on IRIXNet asked the same question.

Power faults

Nothing happens at all

No click, no fan, no light when the power button is pressed. Erik Baigar spent many hours on a Cherokee supply that turned out to be sound; the fault was on the system board. His explanation: while the cord is connected, a standby converter in the supply feeds 3.3 V through the frontplane to the Dallas DS1687 real-time clock on the system board. Pressing the power button makes the DS1687 pull the supply's INHIBIT line low, through two MOSFETs, and the main converters start. His checks, in order:

1. With the supply out, check its outputs and the machine's power rails for shorts.
2. Run the supply on the bench from the mains, through an isolating transformer if you have one. The standby (kick-start) voltage and the INHIBIT line should both sit close to their standby levels: 3.3 V standby at the system board and about 5 V on INHIBIT. If not, the supply is at fault.
3. Refit the supply and check that the standby voltage still reaches the system board. If it does not, check capacitors C1053 and C177 on the system board. In his machine C1053, a 10 µF 16 V part next to the DS1687, had shorted the standby rail.
4. If the standby voltage is present, see whether INHIBIT goes low when the power button is pressed. If it does, the supply is at fault. If it does not, check the DS1687's power output (pin 22); if that goes low, suspect the MOSFETs (one is Q7) and their pull-up resistors.

Another owner on the same thread found the supply's standby output at 3.1 V and sagging to 2.6 V over a few minutes, and concluded his supply was faulty.

Click of death

A relay clicks when the cord goes in, perhaps with a twitch of the fans, and nothing else. weblacky, who rebuilds SGI supplies, describes this as the usual sign of a dead Octane supply and takes it to be the supply's over-current protection acting. Before condemning the supply:

• Try a different wall socket, on a different circuit, with a known-good cord and no extension lead. At least two owners on the forum traced start-up failures to a voltage drop in a bad outlet or extension cable.
• Remove the drives to lighten the load.
• Try a known-good supply. Owners report that the Lucent and Cherokee supplies are interchangeable except with VPro graphics, which needs the Cherokee.

megaimg cured a click-of-death Lucent 623 W supply (060-0028-003): recapping alone did not help, but replacing the TOPSwitch converter on the small AC/DC control board in the middle board and a 67F095 thermostat did.

Starts for a moment, then shuts off

• The supply clicks on, powers the machine for a split second and stops. weblacky suspects protection reacting to low input or output voltage; over-current protection trips too quickly to let the machine start at all. Check the mains supply as above.
• Both fans start for a second, the relay clicks off and the machine buzzes faintly. If the fans start, weblacky suspects a short in a module, not the supply. Take the supply out and look for scorch marks on its connector, then check the system board, the frontplane and the XIO module for shorted rails with a meter in diode and resistance modes, starting with the system board.

Random power-offs

Two owners had Octanes that shut down at random, with XIO errors logged beforehand. In both, the large coils inside the supply had cracked solder joints; resoldering them cured the fault.

Lucent supply damage

Inside dead Lucent supplies owners have found burnt 3 W load resistors on the outputs, a sooty diode, and small electrolytic capacitors that had leaked near the burnt area. One repaired supply exploded at its first power-up and tripped the house breaker; power a repaired supply through a current-limiting lamp the first time. Charring on the primary side has to be cleaned off down to the copper before power is applied, because carbon conducts at mains voltage. See SGI Octane Capacitor Replacement Guide.

Memory faults

• A blinking red lightbar is SGI's memory symptom (see the table above).
• Over the serial console the PROM names the socket: one Octane2 reported "Bad or missing DIMM in bank 0, DIMM S1", mismatched DIMMs in bank 1, read errors in banks 2 and 3, and finally "No usable memory found. Make sure you have at least one full bank (2 DIMMs)."
• That machine's DIMMs and sockets were very dusty. After cleaning, moving a good pair into bank 1 let it boot with half its memory.
• The IDE memory test calls out the DIMM that failed. Swap the pair between banks and run it again to see whether the fault moves with the DIMM.
• The original 030-0887 system board takes 256 MB modules only in the stacked form, not the later single-board 256 MB DIMM.

System module will not seat or will not start

• SGI's first step for a machine that does not boot after module work is to reseat the system module.
• One owner found that small hex screws on the actuators driving the module's locking hooks had worked loose. One hook lagged behind the other and the module never fully engaged the frontplane; with the screws tightened it seated and booted.
• Do not clean the compression connectors with alcohol or anything other than dry gas. See SGI Octane General Maintenance.

Graphics and XIO faults

• A solid red lightbar points at the XIO module or graphics board (see the table).
• One owner's ESSI (SSE) board caused kernel panics until it was replaced. On the failed board a chip (U8) had burst and capacitors C234 and C79A were damaged.
• VPro boards need the VPro-rated supply, IRIX 6.5.10 or later and PROM 4.0 (V6, V8) or 4.5 (V10, V12). The V10 and V12 also need frontplane Xbow 1.4. The command monitor's system command shows the Xbow revision.
• Owners report that the Personal Video, Digital Video and Octane Compression options work only with the earlier IMPACT-class graphics. The Personal Video board connects to the graphics board by three ribbon cables, and VPro boards have no headers for them.

PCI module

On one PCI card cage, part 013-2304-001, tantalum capacitor C25 exploded. weblacky identified it as 22 µF 35 V from his own cage, a different board revision with the same values in that area. The replacement did not cure the machine, which then booted only without the cage, so look for a shorted part nearby before fitting a new capacitor.

Diagnostics

Power-on, confidence and IDE tests

Power-on tests run every time the machine starts and turn the lightbar red on a fault. Confidence tests for the mouse, keyboard, monitor, audio and external SCSI devices are in the Toolchest under System. The Interactive Diagnostic Environment (IDE) is more thorough and takes 30 to 45 minutes: switch off, switch on, click Stop for Maintenance or press Esc at the start-up notifier, and choose Run Diagnostics (option 3). Esc stops the tests. If the System Maintenance menu cannot be reached at all, SGI treats the system as faulty.

Octane_ide

SGI's field service tool Octane_ide boots from /stand on the system disk with boot dksc(x,y,z)/stand/Octane_ide. Set setenv ge_ucode_from_disk 1 first or the graphics test fails, and type list before running any test. Run the PCI shoebox test only on a machine with a PCI module; otherwise Octane_ide stops with a fatal exception.

Octane_ide tests and SGI's action on failure

Command	Tests	On failure
ip30	NIC, HEART, BRIDGE, XBOW, memory, RAD audio, IOC3, real-time clock, DUART	HEART, BRIDGE, RAD, IOC3, RTC or DUART: replace the IP30 board. XBOW: replace the frontplane. Memory: replace the DIMM named
pm	CPU module NIC, TLB, caches, FPU, lpackd	The guide lists the tests but gives no action
fp	Frontplane and XBOW	Replace the frontplane
gfx, mg_test_tram	IMPACT-class graphics board ASICs, NIC, buses, DAC and video timing, framebuffer; texture RAM	Replace the graphics board. A texture RAM failure always names the graphics board, but the fault may be in a TMEZZ texture card on it, so reseat or replace either
shoebox	PCI module	Replace the PCI module

The NIC test checks that no Number-in-a-Can is missing; there is one on the CPU module, the system board, the frontplane and the power supply. gfx does not test VPro boards; SGI's IRIX-based odydiags does that.

NMI button

For a machine that hangs without reporting anything, the recessed NMI button behind the bezel, below the power and reset buttons, forces a non-maskable interrupt. Press it with a straightened paper clip. The system writes its state to files in /var/adm/crash and powers down, losing any unsaved work.

Forgotten PROM password

Remove the system module. Near the CPU is a three-pin jumper inside a white rectangle marked "Disable passwd". Move the jumper from the middle and right pins to the left and middle pins to disable the password; move it back to enable it again.

System does not power off

Press the power button again; this cuts power at once without a clean shutdown. If that fails, press the reset button with a pen, and if the machine still stays on, unplug the cord.

Related pages

• SGI Octane
• SGI Octane General Maintenance
• SGI Octane Capacitor Replacement Guide
• SGI Octane Documentation
• SGI Octane2 Troubleshooting Guide
• Capacitor Failure Symptoms