Jump to content

SGI Onyx 4 Troubleshooting Guide

From RetroTechCollection
SGI Onyx4 G2-brick and G2N-brick rear panel (User's Guide Figure 1-2)
Rear panel of the G2-brick and G2N-brick: AC input, L1 console and USB ports, XIO and NUMAlink connectors with their link LEDs, and two graphics boards with two DVI-I outputs each (SGI User's Guide, Figure 1-2)

Fault-finding on the SGI Onyx 4 graphics bricks, from SGI's Silicon Graphics Onyx4 UltimateVision User's Guide (007-4634-002) and owners' reports. For the host compute modules, which are those of the SGI Origin 350, SGI refers owners to the host system's own guide; see SGI Origin 350 Troubleshooting Guide and SGI Onyx 350 Troubleshooting Guide.[1]

Order of work

[edit | edit source]

SGI asks for three checks before anything else:[1]

  1. Every graphics brick is connected to AC power. The bricks take mains directly, usually from the PDU at the rear of the rack, unlike the 48 V bricks of the Origin 3000 family.[2]
  2. Every graphics brick is powered on.
  3. The host can see every graphics brick: run hinv and count the lines reading Graphics board: SG2. There should be two for each brick.[3]

Then read the L1 display on each brick and check the power supply LED and the link LEDs described below.

SGI's troubleshooting chart

[edit | edit source]
Troubleshooting chart (SGI User's Guide, Table A-1)
Problem Action
The system will not power on Check that the PDU cord is seated and the PDU circuit breaker is on.
One brick will not power on Check the power cord between the PDU and the brick at both ends; read the L1 display. If the L1 is not running, SGI refers you to its engineers.
Service-required LED (yellow) or failure LED (red) lit Read that brick's L1 display and look up the message.
Green or yellow NUMAlink LED not lit Reseat the NUMAlink cable at both ends.
The system will not boot IRIX SGI refers you to its engineers.

[4] The yellow service-required LED means something has failed but the system is still running, for example a fan that has stopped; the red failure LED means the system is down.[5]

Power supply LED

[edit | edit source]

Each brick has a single power supply, with a status LED near the top of its front face.[6]

Power supply LED (SGI User's Guide, Table A-3)
LED Meaning
Off No AC reaching the supply: the brick is unplugged or a fuse has blown
Amber Fault: a voltage, temperature or current limit has been exceeded
Blinking green AC present, main DC output not yet on
Green Working normally

[6] The supply is changed from the front without opening the brick; see SGI Onyx 4 General Maintenance.

[edit | edit source]

A G2-brick connects to the host over its XIO connector; a G2N-brick uses its NUMAlink connector, and its XIO connector is not used. Each connector has a yellow and a green LED beside it.[5]

  • Yellow on: the graphics brick and the host brick at the other end of the cable are both powered.
  • Green on: the host has established a link to the graphics brick.

[5] A brick that shows yellow but not green has power at both ends but no link; SGI's chart says to reseat the cable at both ends.[4]

L1 messages

[edit | edit source]

A voltage more than 10% from nominal produces a warning, and more than 20% a fault, which starts a 30-second power-off of the brick. SGI points to the L1 env command for the limits themselves.[7]

L1 controller messages (SGI User's Guide, Table A-2)
Message Meaning
ATTN: x.xV high fault limit reached @ x.xxV
ATTN: x.xV low fault limit reached @ x.xxV		 A rail is more than 20% out; the brick powers off after 30 seconds
ATTN: x.xV high warning limit reached @ x.xxV
ATTN: x.xV low warning limit reached @ x.xxV		 A rail is more than 10% out
ATTN: x.xV level stabilized @ x.xV The rail 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 another fan may have 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 temperature reached @ xxC xxF
ATTN: TEMP # critical temperature reached @ xxC xxF
ATTN: TEMP # fault temperature reached @ xxC xxF		 The air at the brick's inlet has passed the advisory, critical or fault limit
ATTN: TEMP # stabilized @ xxC/xxF The inlet temperature is back to an acceptable level
Auto power down in xx seconds The L1 has registered a fault and is shutting the brick down; repeated every 5 seconds
Base module appears to have been powered down The L1 has registered a fault and has shut the brick down

[7] The Onyx4 guide gives no temperature figures for these levels; read them with env on the brick's L1. SGI's Onyx 350 guide sets them at 30, 35 and 40 °C for that system.[7][8] The env and log commands are described in the L1 and L2 Controller Software User's Guide.[9]

Wrong position, wrong pipe order

[edit | edit source]

Each brick's L1 shows its location as a string such as 001c12 (rack 001, unit 12 from the bottom). If it is wrong, the boot brick and pipe order can be wrong too, because both depend on the rack positions; reprogramme it with the L1 brick rack and brick slot commands described on SGI Onyx 4 General Maintenance.[10]

To find which brick holds pipe 0, run /usr/gfx/gfxtopology: the location after the @ on the pipe 0 line matches the string on that brick's L1 display. Pipe 0, channel 1 is the lower left DVI-I connector on that brick. With more than one graphics brick, pipe 0 is normally the one lowest in the rack.[11]

Graphics bricks missing from hinv

[edit | edit source]
  • Check the yellow and green link LEDs and reseat the XIO or NUMAlink cable.[5][4]
  • Check that the brick's L1 shows the right rack and slot.[10]
  • Some IRIX versions list both board versions as SG2, so the hinv line does not tell an SG2 from an SG2.1.[12]
  • kaigan's hinv -v -m output lists each board of his G2N-brick as a PCI Graphics Processor with vendor ID 1002 behind a CGBRICK Bridge.[13]

No picture or wrong picture

[edit | edit source]
  • If a digital monitor stays dark on the second output, check the board version. On an SG2 board, the version without a sticker reading 030-2003-xxx on its rear panel, only channel 1 (the left DVI-I connector) can drive a digital monitor; channel 2 is analogue only. The SG2.1 can drive either type from both outputs, and the two versions can be mixed in one system.[12]
  • Each pipe's Device section in /etc/X11/XF86Config-4 can carry Option "MonitorLayout" "channel1type, channel2type" with AUTO, TMDS (digital), CRT (analogue) or NONE. Without the line, both channels are set to AUTO.[14]
  • For monitors or cables that do not follow the VESA DDC standard, add Option "NoDDC" "on" to the Device section.[15]
  • The modes SGI configured by default are 1024 × 768, 1280 × 1024 and 1600 × 1200 at 60 Hz, 1920 × 1080 at 72 Hz, 1920 × 1200 at 60 Hz, and 1280 × 1024 at 96 Hz for stereo; others are added in the Monitor sections of XF86Config-4.[16]
  • After a graphics brick is added, at boot IRIX writes a new configuration as XF86Config-4.N when it finds new graphics hardware, but keeps using the old file unless the xf86config-autoreplace flag is on. Replace the old file with the .N file and reboot, or regenerate it by hand as described on SGI Onyx 4 General Maintenance.[17]
  • The boards are ATI cards driven through XFree86 and support OpenGL only. kaigan found that programs written for IRIS GL do not run at all, and that even Maya 6.5 had glitches to work around.[13]

IRIX level

[edit | edit source]

SGI recommends IRIX 6.5.24, or 6.5.22 or 6.5.23 with patch 5448, and says earlier systems shipped with older releases should be upgraded. uname -R shows the release, and versions patchSG0005448 lists the patch if it is installed.[18] Stereo sync works only on IRIX 6.5.21 with patch 5208, or 6.5.22 and later.[19]

Replacing a graphics board

[edit | edit source]

kaigan identifies the boards as ATI FireGL X1 cards on an AGP riser, and reports that they carry SGI-specific firmware.[13] Raion adds that the FireGL cards used here are not the same as those sold for x86 machines, and that earlier attempts to use other cards did not work.[13] Woof reports that the firmware and its flashing tool are on the Onyx4 installation media.[13] The AGP riser is SGI part 030-1872-007, according to vishnu.[20]

The board swap itself is on SGI Onyx 4 General Maintenance. A 7-pin synchronisation cable runs from the riser to the underside of the board; SGI warns not to pull the board far out before unplugging it, and not to trap it when refitting.[21]

G2N-brick node and interface boards

[edit | edit source]

kaigan's hinv -vvvm lists the same 2U interface board (2U_INT_53, part 030-1809-003) in his G2N-brick as in his base compute module, and a four-processor node board listed as IP59_4CPU (part 030-1989-003) in both.[13] The interface board's L1, Dallas chip and voltage regulator faults reported on the Origin 350 are on SGI Origin 350 Troubleshooting Guide.

[edit | edit source]

References

[edit | edit source]
  1. ↑ 1.0 1.1 SGI, 007-4634-002, appendix A, p. 81. General procedures.
  2. ↑ SGI, 007-4634-002, chapter 2, p. 28. Providing power to G2-bricks and G2N-bricks.
  3. ↑ SGI, 007-4634-002, chapter 2, p. 33. Verifying system connections.
  4. ↑ 4.0 4.1 4.2 SGI, 007-4634-002, appendix A, p. 82, Table A-1. Troubleshooting chart.
  5. ↑ 5.0 5.1 5.2 5.3 Silicon Graphics, Silicon Graphics Onyx4 UltimateVision User's Guide, 007-4634-002, June 2004, chapter 1, pp. 3–6 (File:SGI Onyx4 UltimateVision User's Guide (007-4634-002).pdf). Front panel LEDs, rear panel, XIO and NUMAlink LEDs.
  6. ↑ 6.0 6.1 SGI, 007-4634-002, appendix A, pp. 90–91, Table A-3. Power supply LED.
  7. ↑ 7.0 7.1 7.2 SGI, 007-4634-002, appendix A, pp. 83–84, Table A-2. L1 controller messages.
  8. ↑ Silicon Graphics, SGI Onyx 350 Visualization System User's Guide, 007-4632-001, June 2003, chapter 8, pp. 144–145 (File:SGI Onyx 350 Visualization System User's Guide (007-4632-001).pdf).
  9. ↑ Silicon Graphics, SGI L1 and L2 Controller Software User's Guide, 007-3938-006, chapter 3, pp. 66–78 (File:SGI L1 and L2 Controller Software User's Guide (007-3938-006).pdf).
  10. ↑ 10.0 10.1 SGI, 007-4634-002, chapter 2, pp. 11 and 19. Rack locations, programming L1 rack and slot numbers.
  11. ↑ SGI, 007-4634-002, chapter 3, p. 68. Locating pipe 0.
  12. ↑ 12.0 12.1 SGI, 007-4634-002, chapter 3, p. 63. Using both DVI channels on a card; SG2 and SG2.1.
  13. ↑ 13.0 13.1 13.2 13.3 13.4 13.5 kaigan, Raion and Woof, "It Lives! SGI Onyx350 Prototype/Onyx4 - "Artemis"", IRIX Network forum, July 2020 to May 2024, forums.irixnet.org.
  14. ↑ SGI, 007-4634-002, chapter 3, p. 57, Table 3-2. Monitor types.
  15. ↑ SGI, 007-4634-002, chapter 3, p. 49. Dual-channel configuration and the NoDDC option.
  16. ↑ SGI, 007-4634-002, appendix B, p. 119, Table B-1. Video modes supported by default.
  17. ↑ SGI, 007-4634-002, chapter 3, pp. 38–40. Regenerating the XF86Config-4 file.
  18. ↑ SGI, 007-4634-002, chapter 3, pp. 36–37. IRIX version requirements.
  19. ↑ SGI, 007-4634-002, chapter 3, p. 44. Stereo requirements.
  20. ↑ vishnu, reply of 6 September 2025 in "ONYX4 AGP RISER", IRIX Network forum, forums.irixnet.org.
  21. ↑ SGI, 007-4634-002, appendix A, pp. 112–118. Replacing a graphics card.


Retrieved from "https://retrotechwiki.com/index.php?title=SGI_Onyx_4_Troubleshooting_Guide&oldid=29651"