SGI Onyx 300 Troubleshooting Guide

Onyx 300 compute module rear panel: LINK (NUMAlink) and XIO connectors at left, heartbeat LEDs, serial, RT interrupt and L1 console ports, external SCSI, USB L1 port, USB, Ethernet, two PCI slots and the power inlet (SGI User's Guide, Figure 1-17)

Fault-finding on the SGI Onyx 300, from SGI's SGI Onyx 300 Graphics System User's Guide (007-4509-002) and owners' reports. The compute module is shared in design with the SGI Origin 300, and its own faults, with the L1 console settings and command list, are covered on SGI Origin 300 Troubleshooting Guide; the system-wide checks and graphics faults are below.

SGI's troubleshooting chart

Troubleshooting chart (SGI User's Guide, Table 4-1)

Problem	Action
The system will not power on	Check that the PDU cord is seated and the PDU circuit breaker is on.
A NUMAlink, PCI expansion or graphics module will not power on	Check that its rear power switch is on, read its L1 display, check that its L1 is running, and check its connection to its power source.
Service-required or failure LED lit on a module	Read that module's L1 display and look up the message below.
Green or yellow LED of a NUMAlink port not lit	Reseat the NUMAlink cable at both ends.
Green LED of a power bay supply not lit	Reseat its AC cord at the rear of the power bay.
PWR LED of a populated PCI slot not lit	Reseat the card.
Fault LED of a populated PCI slot lit	Reseat the card; replace it if the LED stays on.
Amber LED on a disk drive	Replace the drive.

SGI's chart sends everything else, including a system that will not boot IRIX, to its own engineers.

Lights

• Compute module front: power button LED green when powered; service-required LED yellow when something has failed but the module still runs; failure LED red when the module is down.
• Graphics module front: On/Off LED green; service-required LED orange; failure LED red.
• Compute module rear: four heartbeat LEDs, one per processor, show processor activity.
• NUMAlink module ports: green when a cable is connected properly; yellow when the L1 has detected the "Remote Power OK" signal from the module at the other end.

L1 messages

SGI counts a voltage more than 10% from nominal as a warning and more than 20% as a fault.

L1 controller messages (SGI User's Guide, Table 4-2)

Message	Meaning
ATTN: x.xV high (or low) fault limit reached @ x.xxV	A voltage is 20% out; the module starts a 30-second power-off
ATTN: x.xV high (or low) warning limit reached @ x.xxV	A voltage is 10% out
ATTN: x.xV level stabilized @ x.xV	The voltage is back within limits
ATTN: FAN # x fault limit reached @ xx RPM	A fan has reached its maximum speed; the room may be too hot, or a fan has failed
ATTN: FAN # x warning limit reached @ xx RPM	A fan has speeded up; check the room temperature and whether the fan settles
ATTN: FAN # x stabilized @ xx RPM	The fan speed is back to normal
ATTN: TEMP # advisory / critical / fault temperature reached	Inlet air above 30, 35 or 40 °C (27, 31 or 35 °C at the high-altitude setting)
Auto power down in xx seconds	The L1 has registered a fault and is shutting the module down; repeated every five seconds
Server appears to have been powered down	The L1 has shut the module down after a fault

SGI's Onyx 3000 guide describes the fan messages differently, in terms of a fan failing to reach its minimum speed; see SGI Onyx 3000 Troubleshooting Guide.

A module will not power on

1. Check that every L1 display reads L1 running.
2. At the L2 prompt, type cfg. A module that is not in the list will not power on; check that its L1 is running and that its USB cable to the L2, NUMAlink module or USB hub is connected.
3. The graphics module plugs into its own 220 V outlet, not the PDU, and its L1 starts only when its rear switch is on.
4. In a base system without a NUMAlink module, the upper compute module reaches the L2 through its NUMAlink cable to the lower one, so check that cable too.
5. When powering on with the buttons, follow SGI's order: PCI expansion and graphics modules, then the NUMAlink module, then the compute modules from the lowest one up.

No L1 comm

jpstewart reports that No L1 comm on the L1 display is exactly what an Origin 300 shows when its Dallas battery has died. jwhat, who has tested drop-in replacements for the Dallas DS1742W clock and NVRAM chip used by the L1 in SGI's later machines, notes that other owners have tested one in an Origin 300. Check the battery-backed chips before suspecting the board.

Graphics faults

• Each pipe needs its own compute module, cabled from the module's XIO port to the Ktown2 board: the top Ktown2 connector serves the right-hand pipe (seen from the rear) and the bottom connector the left.
• Wrong colours on a monitor from another system: set its RGB inputs to 75 ohms.
• Blank screens when adding monitors: SGI gives about 300 Mpixel/s as a pipe's total output, so mix lower and higher resolutions to stay within it.
• Before plugging in a monitor, check that its rating label matches the outlet voltage.

Before calling it a hardware fault

Racks joined by NUMAlink or Xtown2 cables must not differ in ground potential by more than 0.25 V; feed the graphics modules and the PDU from the same breaker box.

Related pages

• SGI Onyx 300
• SGI Onyx 300 General Maintenance
• SGI Onyx 300 Capacitor Replacement Guide
• SGI Onyx 300 Documentation
• SGI Origin 300 Troubleshooting Guide
• SGI Onyx 3000 Troubleshooting Guide