Please consider a donation to the Higher Intellect project. See https://preterhuman.net/donate.php or the Donate to Higher Intellect page for more info.

Difference between revisions of "SGI Indigo2"

From Higher Intellect Wiki
Jump to navigation Jump to search
 
Line 294: Line 294:
 
File:28301649897 53cb5305d4 o.jpg
 
File:28301649897 53cb5305d4 o.jpg
 
File:42950261121 46a6f5de8b o.jpg
 
File:42950261121 46a6f5de8b o.jpg
 +
File:Indigo2-front-1901.png
 
</gallery>
 
</gallery>
  

Latest revision as of 17:41, 11 October 2019

SGI Indigo2
30066309397 ed38f1de2f o.jpg
44283158334 2fb6c74eb0 o.jpg
44757864071 c1d2878d18 o.png
43602537314 4f2846cf44 o.jpg

The Indigo2 was introduced by SGI in 1992 and discontinued in 1997.

The SGI Indigo² and the SGI Challenge M were Unix computers marketed by SGI from 1993 to 1997, with production of IMPACT models ending in 1998. The Indigo² was a desktop workstation while the Challenge M was a server which differed from the Indigo² only by a slightly differently colored and badged case, and the absence of graphics and sound hardware. Both systems were based on the MIPS processors.

Both systems supported EISA and GIO64 expansion buses via a riser card. Indigo² systems also supported SGI IMPACT graphics hardware. The IMPACT graphics boards draw more power than the GIO-64 bus can deliver, so the IMPACT-ready systems have additional power connectors on the expansion riser card, with a separate connection to the power supply. An IMPACT-ready Indigo² must have an IMPACT-ready riser card, an IMPACT-ready power supply, and a sufficiently recent PROM revision.

CPU Options

The Indigo² has three distinct variants, each with a specific motherboard and "IP" number or designation:

  • IP22 supports an R4000, R4400, or R4600 CPU clocked at 100-250MHz
  • IP26 supports the R8000 CPU clocked at 75MHz
  • IP28 supports the R10000 CPU clocked at either 175 or 195MHz

IP26 systems were typical referred to as POWER Indigo2, while the IP28 systems usually had a grill badge that read "10000."

SGI Part Number Description
030-8097-002 IP22 R4000 100MHz 1MB SC
030-8116-004 IP22 R4400 150MHz 1MB SC
030-0814-001 IP22 R4400 200MHz 2MB SC
030-8255-002 IP22 R4400 200MHz 1MB SC
030-0808-001 IP22 R4400 250MHz 2MB SC
030-0658-002 IP22 R4600 133MHz 512KB SC
030-0751-001 IP26 R8000 75MHz 2MB SC
030-1143-002 IP28 R10000 175MHz 1MB SC
030-0966-004 IP28 R10000 195MHz 1MB SC

Memory Subsystem

All three variants had 12 SIMM slots on the motherboard, organized into three banks of four slots each, and took industry standard fast page mode (FPM) 72 pin SIMMs with parity. Speeds should be 60 or 70 nanoseconds, and the internal organization must be 36 bits wide - 8MB x 72 bit parts will not work.

IP22 systems will support up to 384MB with 32MB SIMMs. While the IP26 and IP28 systems both support 64MB SIMMs, published limits for these systems reflected concerns about the amount of heat generated by then-current DRAM chips. According to SGI the R8000-based IP26 systems would only support 640MB (2 banks of 256MB, one of 128MB), while the R10000-based IP28 would support 768MB (3 x 256MB).

Eventually 64MB SIMMs became available that generated less heat, and denser 128MB SIMMs became available. Both the IP26 and IP28 can use these 128MB SIMMs, but with limitations. IP26 systems require some banks to use lower-profile SIMMs to clear the CPU carrier, and SGI described limits on the mix of different density SIMMs in these systems. However owners have reported working configurations of up to 896MB (1 x 128MB, 1 x 256MB, 1 x 512MB bank).

With the IP28 it is possible to achieve a total of 1GB of RAM (2 banks x 512MB, or 2 x 256MB + 1 x 512MB), but unfortunately this appears to be a hard limit based on address logic.

(Much of this information was taken from Ian Mapleson's Indigo2 GigaRAM page. Thanks also to Bert Heise, Roger Garcia and David Cummings.)

Compatible 128MB SIMMs

A number of mainstream servers used 128MB SIMMs that are compatible with the Indigo2. Here is a list of models:

  • HP/Compaq NetServer LX Pro 6/166
  • HP/Compaq NetServer LX Pro 6/166 SMP
  • HP/Compaq NetServer LX Pro 6/200
  • HP/Compaq NetServer LX Pro 6/200 SMP
  • HP/Compaq NetServer LXe Pro 6/166
  • HP/Compaq NetServer LXe Pro 6/200
  • HP/Compaq NetServer LXr Pro 6/200
  • IBM PC Server 704 (8650-4MO)
  • IBM PC Server 704 (8650-5MO)
  • IBM PC Server 704 (8650-6MM)

Nekochan member smj found that HP part number D4893A was a reliable search term for finding compatible parts on eBay circa 2010.

Graphics

Indigo2.guide.p4.1.gif

The graphics boards available for the Indigo² were the pre-IMPACT Newport and Express boards (which included the SGI XL24, SGI XZ, Elan and SGI Extreme) and the MGRAS IMPACT boards (the SGI Solid IMPACT, the SGI High IMPACT, the SGI High IMPACT AA, and the SGI Maximum IMPACT). IMPACT graphics is not supported by the Power Indigo² (R8000 CPU). The Indigo2's replacement, the SGI Octane, offered an upgraded bus but featured the same graphics options, albeit in repackaged form.

Identifying Indigo2 Graphics Options

Most Indigo2 graphics options can be identified easily by looking at the rear panel of the machine. First generation graphics options (XL/Newport and Express) have a round 3-pin mini-DIN connector for stereo glasses. XL/Newport can be differentiated from XZ or Extreme as the Express graphics options (XZ or Extreme) have a small mini-BNC genlock connector in addition to the stereo port and the 13W3 monitor connector.

XZ (2x GE7) can not be differentiated from XZ-Elan (4x GE7) without the machine running as the boards are exactly the same, with the XZ-Elan having 4 GE7 chips packaged in the GE7 MCM, whereas the original XZ had 2 empty wells.

Extreme graphics has 3 boards which you should be able to discern through the ventilation grating next to the slots.

IMPACT-class graphics have the 13W3 and a DE-9 connector for the stereo glasses. The class of graphics is easier to determine in IMPACT because the DE-9 is on the base board, whereas the 13W3 connector is on the top board.

SGI Solid IMPACT graphics (1 board) have both the 13W3 and the DE-9 on the same board.

SGI High IMPACT is a two board set, with the DE9 underneath the 13W3. It is impossible to tell from the back whether it is a High-AA IMPACT or not (High-AA IMPACT boards have two GE11 Geometry Engines whereas original SGI High IMPACT boards come with a single GE11. The relationship is the same as the relationship of original XZ to XZ-Elan). If the cover can be removed, the High-AA IMPACT will have two large PGA chips on the lower board, whereas the original High IMPACT will have one large PGA chip and one empty pad.

SGI Maximum IMPACT is a three-board set, so there will be a blank rear plate between the DE-9 and the 13W3.

There is no way of determining whether or not the SGI High IMPACT and SGI Maximum IMPACT boards have the 4MB Texture RAM expansion option installed or not from the outside. If you can open the Indigo2, you can look at the top board to see if the expansion module is present or not.

As always, if the system is operational, hinv and /usr/gfx/gfxinfo will give you complete information on the system hardware and graphics subsystem.

I/O Subsystem

  • PS/2 Mouse and Keyboard ports
  • One 10Mbps 10baseT Ethernet interface, with AUI and RJ45 connections
  • Two RS422 serial ports (mini-DIN, to 38.4kbps)
  • One bi-directional parallel port
  • IRIS Audio (up to 48KHz sampling on inputs)
  • Two 10MBps SCSI controllers (WDC 33C93B)
  • EISA expansion bus (4 slots for Extreme gfx, 3 for IMPACT)
  • GIO64 expansion bus (3 slots for Extreme, 4 for IMPACT)
  • One 5.25" SCSI drive bay (externally accessible)
  • Two 3.5" SCSI drive bays (one externally accessible)

Note that the EISA and GIO slots are interleaved on the same riser, so all may not be occupied at the same time.

ATM, FDDI, Fibrechannel, HPPI, and Token Ring interfaces were available from various vendors for either the EISA or GIO bus. At least one GIO64-VMEbus adapter was offered (Bit3).

Internal Drives

The internal drive bays of the Indigo2 take model-specific carriers. These carriers mate to a backplane using a non-standard connector, and contain devices using a standard 50 pin IDC ribbon cable connector. There may not be enough room in the carrier to use an adapter board to use 68 or 80 pin devices unless using a down-sized device, e.g. a 2.5" hard drive in a 3.5" drive carrier.

Faster Ethernet Options

In addition to the 10Mb Ethernet controller on the motherboard, there were several 100Mb LAN options available from 3rd parties, either via EISA or GIO64 expansion cards. The two best known and most widely used Indigo2 network cards are 3Com 3C597-TX 100Mbit EISA Card, and the Phobos G160 GIO64. The second one offers better overall performance due to using the superior GIO64 bus, which also has the effect of reducing the CPU utilization through use of DMA transfers.

Motherboard I/O Connector Pinouts

Indigo2 io backplane pinout.jpg

Add-On Options

Digital Media

Pre-Impact

Impact


Operating Systems

Versions of IRIX

The SGI Indigo2 was initially delivered with IRIX 4D1-4.x. Support for the Indigo2 finally ended with IRIX 6.5.22.

Support for the Indigo2 IMPACT was available with point releases of IRIX 5.3 and IRIX 6.2. Patches to IRIX 6.2 were also shipped with the addition of the R10000 CPU option.

Important Patches

IRIX patch 7009 is best known for fixing an upgrade/installation bug with rqsall(1m) under 6.5.22 (and possible 6.5.21) on systems featuring R8000-based IP21 or IP26 boards. These are usually found in the POWER Indigo2 and POWER Onyx workstations. The key to upgrading these systems to 6.5.22 is to make sure patch 7009 is applied before you run rqsall(1m) at the end of your installation. This way the corrected rqsall(1m) will be used to setup the system for reboot.

Linux

Linux was ported to the Indigo2 along with other SGI models of the Indy-Indigo2-Octane period. More information can be found on the Linux/MIPS.org wiki, specifically on their IP22 page, a catch-all for SGI hardware from this period.

XL graphics were commonly supported by distributions in the 2.4 kernel timeframe. A driver for IMPACT graphics and patches for XFree86 and X.org servers are available courtesy Peter Fuerst.

Troubleshooting and Repair

Power Supply

The machine may refuse to power up at all, though the fans move briefly when the power switch is thrown. Or the machine to switch off at random, regardless of whether the environment is warm or cold. Briefly disconnecting the machine from the electrical outlet may clear the condition for a time. According to this article the problem may be bad capacitors in the high-voltage section of the power supply.

Note that the article describes a procedure working with high-voltage electronics, and could be dangerous.

You can see more information about Indigo2 power supplies at Indigo2 Power Supply Differences.

Impact Graphics TRAM: Troubleshooting, Installation and Removal

> MaxImpact has two TRAMs. IIRC, the 2nd one is on the middle PCB.

Yes, it's on the middle PCB.

Careful how you extract the 2nd TRAM. Remove the end plate first (the one that goes over the monitor and stereo socket), then remove the top 4 locking screws. When you bend the board upwads, don't force it all the way to 90 degrees. Just keep it at around a 50 to 60 degree angle. Use a pair of tweezers or needle-nose pliers to extract the plastic locking pins from each end of the TRAM. If you use anything like a flat-bladed screwdriver to lift up the TRAM away from the PCB, be careful to do so at points on the PCB where there are no surface mount components that could be harmed.

When you're testing a Max set, do so with either no TRAMs installed, or both. Don't try it with just one TRAM. This is why I tried very hard to find a HighIMPACT set so that I could test TRAMs one at a time, then move pairs onto a Max set once I had 2 TRAMs that are ok.

Also, do not test IMPACT boards with TRAM installed when the top-lid for the Indigo2 is not fitted. Alternatively, use a separate fan to help cool the TRAM on the top board if the top-lid is not in place. I use a fan connected to the molex cable of a disk sled in the upper disk bay.

In all the above, observe antistatic precautions and make sure the bottom PCB does not scrape against any surface or object which might result in other surface mounted components failling off.


The most common fault with TRAMs is the Alpha channel. Use the tex_cube demo to check initial Alpha functionality. The image should appear consistent whether the window is small or large. If it works ok, check more complex Alpha features with the Macau demo - watch how trees and lamp posts appear, look for artefacts around the billboard edges (press T to show the billboard shapes), and also watch how building textures in the distance appear - they should not show any dithering or 'rolling' pixel effects if the TRAM is working ok.

If the Alpha is working ok, I then test more basic texturing using the Mirrors demo, Matterhorn, Huge Engine Model (including transparency mode, and test shading/wireframe with textures off), A10, F18, and of course try an IrisGL task aswell (I use Beethoven model from buttonfly).

Other symptoms:

- If running a 3D app causes the display to lock up and then resets the X server back to the login screen, this is normally a fault with the GE board. Often this is the case when the display looks ok after initial power on and after the system has fully booted and logged in, ie. the DG and raster elements are ok, but a bad GE isn't exercised until a 3D app is executed. This fault results in gfx timeout errors in SYSLOG.

- If the display is already bad after initial power-on, ie. when the Press ESC for Maintenance message appears (typically visible as lots of jagged horiztonal line effects), then this is normally a bad DG board.

- If the system makes a normal power on tune and sounds like it's booting ok, but the screen is black, this can be the result of one of the TRAMs not being seated properly. Check by removing both TRAMs, ensure the base Max set is ok, then refit the TRAMs, ensuring they are correctly seated.

If you decide to swap boards between Max sets, be very careful. Don't use the tiny levers to disconnect the ribbon cables, they are too fragile; I use a small flat-bladed screwdriver to just lift up the connector either side of the socket. Take the opportunity to remove any dust with a soft brush and/or inert air duster.

Don't bother using the PROM ide for testing, it's not accurate, often reporting false positives.

If you can, use the lower three slots to test Max sets. This gives better cooling for the TRAM at the top.

R4400/250 and lockups, audio buzzing

There is some sort of CPU bug on this particular system which will cause buzzing sound in audio playback or system freeze/lockup if the CPU becomes too idle. It seems to happen after a certain period of idle time and is fixed by hitting a key on the keyboard or using the mouse, as this generates a hardware interrupt or something along those lines.

A timing issue of some description, possibly internally within the CPU cache can cause momentary lockup of the system, and if anything is being played (ie mp3s) a buzz is heard from the speaker. This problem appears to manifest itself more when using the sound hardware, however anecdotal evidence suggests it can happen anytime. Furthermore

  • the network
  • the real time clock

all seem to have some effect - or are effected by this problem. For example stopping NTPD reduced the effect.

The "effect" can be stopped by engaging with the Indigo2 somehow. This can be via movement of the mouse or pressing any key. It can also stop if sufficient network activity is generated. Reading the time from the timeserver for example or being prompted for a login to the FTP server or simply pinging wasn't enough.

Developing from a previous simple amendment to someone elses UDP client/server scripts, I have amended the same persons Multicast UDP scripts. Taking mcastServer.c and mcastClient.c from pont.net I have created a new multipurpose script which can act as client or server.

I'm not a C coder, hence the childishly ripped source, however to I have made the following modifications.

  • client: endlessly loop repeating the sending a \0 char
  • server: do nothing

You can rig up your rc startup scripts to start the server process on your Indigo2 and also to make a call on the client machine (in my case my always on Debian box) using rsh (or more secure alternative) to execute the client.

Archived at https://web.archive.org/web/20040721072035/http://thecave.reality.dsvr.net/irix/indigo2_lockup_workaround/ but r4ki2-buz.c source is lost.

Dual head Newport/XZ - GL only on one screen

This configuration was known as NEWPRESS or Newport/Express dual head. If you leave both graphics cards in and reinstall eoe.sw.gfx you should get the correct library. If this doesn't work you might want to try installing and specifying the "-m GFXBOARD=NEWPRESS" option to inst or swmgr.

Indigo2 Gallery

External Links

SGI Documentation:

More information can be found at:

Manuals

  • 007-3244-002 - Indigo2 IMPACT 250 MHz CPU Upgrade
  • 007-9096-050 - Indigo2 Workstation Owner’s Guide
  • 007-9206-003 - Indigo2 and Indigo2 IMPACT Dual Head Installation Guide and Notes for Developers
  • 007-9207-002 - Indigo2 CPU Upgrade Installation Guide
  • 007-9226-001 - Cosmo Compress for Indigo2 Installation Guide