SGI Crimson General Maintenance

Status panel and I/O door layouts of the IRIS Crimson (SGI installation guide, Figure 4-2)

Routine care of the SGI IRIS Crimson is mostly a matter of its power supply, its cooling and the battery that holds its PROM settings. The Crimson is now over 30 years old. Its Power-One or Cherokee power supply and its battery-backed settings memory are the parts owners most often have to deal with. SGI's own documents cover installation and board replacement but not the inside of the power supply.

Site and environment

Site requirements (SGI installation guide, chapter 3)

Item	Requirement
Mains outlet	NEMA 5-20 (20 A) in the US and Canada; NEMA 6-15R elsewhere
Room temperature to maintain	15 to 35 °C (59 to 95 °F)
Relative humidity	20 to 80%, non-condensing
Ventilation	Keep objects at least 12 in (30 cm) from every side of the chassis
Service access	About 3 ft (91 cm) around the system
Heat load	4777 BTU/hour from the chassis; about 0.44 ton of air conditioning

Table 3-1 in the same guide gives the operating range as 10 to 35 °C at sea level. The system weighs 82 kg, so plan where it will stand before moving it.

Safety before opening

Warning: the Crimson's power supply rectifies the mains to about 300 V DC on an internal bus, and large parts of the Power-One supply are not isolated from the mains.

1. Shut down IRIX and switch off at the front power switch.
2. Switch off the main circuit breaker at the back and unplug the power cord. SGI's guide says this is the only way to remove power from both the system and the supply.
3. Expect the power supply fan to run for about 20 seconds after the breaker is switched off.
4. Wait before touching the supply. jan-jaap found about 70 V still on the Power-One's main filter capacitors more than ten minutes after the fan had stopped, and destroyed the bias supply's switching transistor by shorting it with a probe at that point.
5. Wear an earthed wrist strap and put boards on an antistatic mat. SGI also says not to use an ohmmeter on the boards.

Opening the chassis

1. Release the retainer latches on the front door and swing it open to 90 degrees. The door can be lifted off its hinges for easier access.
2. Release the I/O door retainer clip and remove its three screws. Disconnecting the external cables first makes the I/O door easier to open.
3. Before pulling the IP17, unplug the 64-pin serial ribbon cable from it. Before pulling an MG1/graphics board, unplug its cables from the I/O panel.
4. Use the board ejector tabs to remove and seat boards. A board is home when both ejector lips are vertical and it has snapped into the backplane connectors.
5. To reach the fan controller board, remove the plastic back panel and then the sheet-metal cover behind it; the controller sits just above the backplane.

When fitting a board, hold any cables in the slot to its left flat against their board so they are not crushed.

Cleaning

Dust builds up in the card cage and on the fan tray. CB_HK stripped his Crimson to the bare chassis and cleaned every part, checking for unseated cables, loose memory and broken wiring as he went. Another forum member recommends a residue-free contact cleaner on connectors and isopropyl alcohol on memory contacts. jan-jaap's caution in the same thread applies: if the status panel shows over-voltage lights or the system powers itself on, the fault is in the power supply, and cleaning contacts will not cure it.

The front power switch on the status panel can stop working through dirt. CB_HK's system would not start with a good supply until he worked contact cleaner into the switch; dexter1 replaced his.

Memory and jumpers

All memory is on the IP17 board, in 32 sockets arranged as four blocks of eight sockets (four banks of two in each block).

• Fill bank 1 of each block first, and fit SIMMs in groups of eight.
• High-density SIMMs are 8 MB (64 MB per group of eight); low-density SIMMs are 2 MB (16 MB per group).
• All eight SIMMs in one location must be the same density. Mixing them in one location causes memory errors, and the diagnostics check only one of the eight SIMMs in a location.
• High- and low-density groups can share the board only when the high-density groups are in the lower-numbered locations. SGI lists fourteen allowed layouts, from 16 MB to 256 MB.

Two jumpers matter when the IP17 and IO3B are out of the machine. Jumper B009 at the upper left of the IP17 must not be fitted. Jumper C0F0 at the lower left of the IO3B must be on the pins nearest the backplane connector; in the middle position it causes memory errors. The same C0F0 setting is what adapts an IO3B for the IP17, since the board is also used with Power Series processor boards. SGI's guide warns that only the IO3B supplied with the IP17 kit works with the IP17; older IO3 boards will not.

VME slots

Fit VME boards from slot 1 upwards with no empty slots between them. The backplane passes the VME grant signals from board to board, and an empty slot breaks the chain unless jumper blocks are placed on the pins at the rear of the chassis, to the left of that slot's P1 connector. sgistuff.net adds that the BG/IACK jumper must be removed for each slot that holds a board, and that a Genlock board always goes in slot 4 with the genlock jumpers beside its lower connector in place.

SCSI cabling and disks

As shipped, port 0 of the IO3B feeds both the upper and lower drive bays, daisy-chained by a jumper cable, and port 1 is unused. The drives can be split so that port 0 serves the lower bays and port 1 the upper bays, which SGI calls striping. Port 1 can instead be taken to the I/O door for external devices using the flat ribbon cable SGI supplies for that purpose; the other internal cable will not work there.

Devices per SCSI channel in the two internal configurations (sgistuff.net)

Configuration	Channel 0	Channel 1
Jumpered (bottom connectors bridged, all bays on channel 0)	4 internal, 0 external	0 internal, 7 external
Terminated (each bay pair on its own channel)	2 internal, 5 external	2 internal, 5 external

jan-jaap runs his Crimson on a 50-pin narrow SCSI disk. Owners report trouble with later SCA drives on adapters: Geoman's 18 GB Quantum and a newer IBM DDRS drive gave errors where an older IBM DCHS worked, and jan-jaap could not get a Seagate Cheetah 15K to work in his Crimson.

Fans and cooling

The fan tray below the card cage has four blowers in early Crimsons and six in later ones. The six-blower tray (SGI part 013-0564-001) places three blowers directly under the graphics slots and is required for RealityEngine. A thermostatic fan controller board above the backplane sets the blower speed, and the default speed is low. For RealityEngine, SGI's procedure removes the thermal sense connector from the two-pin header on the controller and fits a jumper across the header so the fans run at full speed continuously. SGI warns that without the jumper the fans always run at low speed, even under adverse conditions.

Check that every blower turns at power-on. The thermal circuit breaker on the status panel trips when the exhaust air reaches 60 °C.

Power supply

Crimsons have either a Power-One SPM5 modular supply or a Cherokee supply, both rated for 1050 W total output. Remove the plastic back panel and the metal sheet under it to see which is fitted; the supply sits above the mains inlet, silver for Power-One and black for Cherokee.

Facts that matter when checking a Power-One unit, from jan-jaap's repairs:

• The SGI version is SPM5-AD-AE-LB-S180. The AD module supplies 5 V at 170 A; the AE module supplies +12 V, −12 V and −5 V. SGI's configuration is rated 5 V at 170 A, −5 V at 5 A, +12 V at 16 A and −12 V at 7 A.
• It does not switch between 110 V and 220 V by itself. A wire link on the mains terminal block sets the input range, and 220 V applied to a unit linked for 110 V destroys it.
• With the breaker on, the input module puts 300 V DC on the internal bus and the fan runs, but the outputs stay off. The front power switch releases the inhibit signal, the outputs come on and the fan speeds up. A Cherokee supply does not run its fan until the system is switched on.
• The SGI version has an extra board between the 12-pin control connector and the supply's control circuit. As jan-jaap traced it, it appears to switch the outputs off if the supply overheats. Keep it when building a supply from a donor.
• A DIP switch bank on the supply's backplane configures it. Copy the original settings to any replacement.

Test a suspect supply on the bench with a dummy load before refitting it; see SGI Crimson Troubleshooting Guide for the procedure and SGI Crimson Capacitor Replacement Guide for the capacitors. Neither SGI guide gives a procedure for the supply's internals.

Settings battery

The PROM environment (console, boot mode, boot device and so on) is held in a battery-backed SRAM that sits in a Dallas DS1216 socket on the IO3B. The DS1216 contains potted lithium cells, so when they run down the whole module has to be replaced or modified. Owners have fitted an external coin cell:

1. Remove the SRAM from the DS1216, and the DS1216 from its socket. Its pins are fragile.
2. Disconnect both potted cells from the circuit. CB_HK drilled small holes near pins 8 and 14 to cut the links to the cells.
3. Check that there is no continuity between ground and pins 4 and 14, which shows that neither old cell is still connected.
4. Solder the positive lead of a coin cell holder to pin 4 (BAT1) and the negative lead to ground. CB_HK used pin 8; GanjaTron took ground from the unused J2 connector at the front of the IO3B.
5. Tie pin 14 (BAT2) to ground. GanjaTron found his module still lost its settings until he did this.
6. Mount the holder on the IO3B; there is little room between the IO3B and IP17 boards, so keep it flat.

After the work, the PROM reports an error once and the variables have to be set again with setenv; this PROM has no resetenv command. jan-jaap gives console=d to use the serial console and bootmode=c to boot automatically after the power-on tests. For lithium cell hazards see Battery Explosion, Capacitor or Corrosion Damage.

Date and time

The Crimson is not year-2000 ready: after a reboot it may think it is in the 1970s. Set the date after booting, or let IRIX take it from a time server on the network.

Related pages

• SGI Crimson
• SGI Crimson Troubleshooting Guide
• SGI Crimson Capacitor Replacement Guide
• SGI Crimson Documentation
• Recommended Tools