SGI: Hardware

Hi

I have recently aquired a Crimson VGX and a PowerSeries 4D/310 GTX.

Before I go and break something by just flipping the switch. Is there anything I should look out for?

I seem to recall that PSU:s tend to fail on these machines, perhaps a preventive recap is in order?

Edit: ooh, I get to update my sig now

Pontus wrote: I seem to recall that PSU:s tend to fail on these machines, perhaps a preventive recap is in order?

There are two types of PSUs for these systems: Power One ("silver") and Cherokee ("black").

The Power One is a modular PSU sold to many customers, in many different configurations. If they fail they tend to just not start up. The Cherokee is less fancy and tends to just self destruct (charred PCBs etc). For this reason, and the fact that they come apart nicely (easier to repair!) I prefer the Power One. Usually you find the Power One in older Power Series and the Cherokee in Crimsons. I have always suspected that the Cherokee PSU was a matter of production cost saving. The PowerOne PSUs were marketed in nuclear and medical applications so they probably did not come cheap.

Otherwise my biggest concern would be the cooling (fan tray), which is not monitored. For that reason alone I would probably pull the cards and do a test run without any cards in the card cage so you can inspect the fan tray. There were two different generations of fan trays for these systems: the original and the "RE-ready" one. You'll probably find the latter one in the Crimson, the fans are positioned slightly different to better cool a Reality Engine.

Next up do a test run with the cards installed and monitor the LEDs on the cards. If they are "solid" (there is supposed to be LED activity during operation): good. If they all blink "nervously" in sync: power down immediately because you've got a PSU problem. Do not run the system for too long with the card cage open.

Oh, and if you dig deep enough into the USENET archives (comp.sys.sgi.*) you'll find discussion about a 1050W PSU versus a 1500W PSU, with the latter assumed to be RE ready. This is all nonsens. Both the Power One and the Cherokee have exactly the same output specs. The Power One, being modular, has a chassis capable of (at most) 1500W power intake from the mains, but the modules installed in this PSU configuration have a combined power output of 1050W. A different module configuration might get closer to that 1500W number, and of course there's a loss between input and output. You have to remove the PSU from it's chassis before the label which spells it all out becomes visible. Bigger PSUs existed but were used in 4D "Predator" and "Skywriter" racks.

Random final remark: beware of the capstan rollers of QIC150 cartridge tape drives. They "rot" and turn into chewing gum.

Have fun!

congrats Pontus. seems you got yourself a pair of real classics there

Thank you jan-jaap. Excellent information as always. Is the color of the PSU visible from the back? In that case I think both machines have the silver Power One PSU, but I'll double check.

Are the board or board connectors themselves fragile. I've broken boards in other computers simply by loosening them (both retainer clips and components). I'll be careful of course, but it never hurts to ask.

@foetz: I've been looking for a crimson on and off for about eight years, then these two come from different directions within a month Sometimes you are lucky. By coincidence they come from the same university, though not the same department.

Pontus wrote: Is the color of the PSU visible from the back? In that case I think both machines have the silver Power One PSU, but I'll double check.

Maybe with a flashlight. It's on the left hand side, upper half. There are ventilation openings, but there's a metal chassis cover behind the skin, still. This is what a Power One looks like installed in the system, with the skin and the metal cover removed: viewtopic.php?f=3&t=16729650#p7379686

My 4D/440VGX is one of my favorite systems. Can't really explain why, it must be a combination of the SMP config, the "big" graphics and the deep humming of the system fans. I'm quite happy I brought it back to life

This PowerSeries is a 310, which makes it a single processor IP7 system (according to sgistuff.net). Your 440 is a quad processor IP15. The Crimson is a one processor IP15. Can the Crimson backplane handle one or more IP7 or IP15, to make it multiprocessor?

Oh, here are pictures:

Pontus wrote: This PowerSeries is a 310, which makes it a single processor IP7 system (according to sgistuff.net). Your 440 is a quad processor IP15. The Crimson is a one processor IP15. Can the Crimson backplane handle one or more IP7 or IP15, to make it multiprocessor?

The Crimson is IP17 . The 4D/440 is an IP15 hardware wise, IP7 software wise. If you pull the CPU board of a 310 you'll see that it is essentially a half populated IP7.

Crimson uses the same MP backplane as the PowerSeries, but you cannot install multiple IP17s. The Crimson has RAM on the CPU card so it doesn't need to access it over the MP bus on a separate MC2. Crimson was a way to get the 64bit R4000 out there, SMP had to wait for Challenge/Onyx. The R4000 was already limited by the 4D backplane.

Pontus wrote: Oh, here are pictures:

Nice, looks complete and in good shape! Is that a VAX next the Crimson or an Abekas Framestore?

Doh, I meant to write IP17. Hypothetically, could the IP17 be pulled and replaced with an IP15 ? In other words, is the backplane the same?

To the left of the Crimson is a MicroVAX II and to the right is a MicroPDP 11/73. They are unrelated to the Crimson. (I have all the plastics for them, just not mounted)

Both the Crimson and 4D is complete, some scratches and stains but nothing major.

I've been looking for documentation, the techpubs goes back to Challenge and Onyx but nothing for the PowerSeries or Crimson unfortunately. There is some information in the hardware developer handbook which is nice. But aside from looking for actual dead tree media is there any other source, like the IRIX CDs ?

Pontus wrote: Doh, I meant to write IP17. Hypothetically, could the IP17 be pulled and replaced with an IP15 ? In other words, is the backplane the same?

'Yes' on both counts, the backplane is the same. If you want to install one (or two!) IP7 or IP15 boards you have to install an MC2 as well. You can't mix IP7 and IP15 in one system. Technically a Crimson is a 4D/510 PowerSeries, or put differently: a Crimson is a PowerSeries with an IP17 CPU board and an IO3B I/O board. Most PowerSeries came with an IO2 or IO3.

My first Crimson was probably an upgraded PowerSeries. It didn't have the red skins:

Pontus wrote: I've been looking for documentation, the techpubs goes back to Challenge and Onyx but nothing for the PowerSeries or Crimson unfortunately. There is some information in the hardware developer handbook which is nice. But aside from looking for actual dead tree media is there any other source, like the IRIX CDs ?

If you're looking for an online (InSight) owner's manual: nope. These systems used to come with ~ 2m of brown binders, mostly hardcopy of IRIX manpages, GL and Fortran manuals:

I think one is called the "Server Owner's Manual" or similar, but it's generic and not nearly as in depth as what you get for later systems. I have a scanned 3rd party Crimson Field Service Guide which has some info, but by far the most informative is probably This old SGI , and there's good info in Simon Pigot's Crimson/Power Series/Onyx RealityEngine/RealityEngine2 Info .

IRIX does include standalone diagnostics software for IP7, IP15 and IP17 so you can exercise the CPUs, I/O and memory. There's a separate field diagnostics disc for IRIX 5.3 which can exercise RealityEngine and VGX, probably GTX as well.

Pontus wrote: Are the board or board connectors themselves fragile. I've broken boards in other computers simply by loosening them (both retainer clips and components). I'll be careful of course, but it never hurts to ask.

Not especially. Pull the levers simultaneously, or "wiggle" them to release the board from the backplane. I've seen one or two boards with a missing or broken lever, but it never happened to me.

Be careful to to put stress on the boards when they're out, e.g. don't stack them.

bla bla ESD precautions bla bla. You know the drill.

I tackled the 4D tonight. I pulled out the boards and left them halfway out in the card cage. The machine is very clean inside, I only found and removed a little dust in the front air intake.

I plugged it in and I hear a fan go on, I suppose it's a 220V fan for the power supply? Flipped the switch on the front: nothing!

*sigh* ... but I didn't give up and figured it might need a bit of load to start up. I inserted the IO2, IP7 and Memory board and tried again. Nothing! blast.. thinking of giving up and start with the Crimson.

I noticed that when I pulled the plug there was a blink on the two LEDs above the power switch. So gut feeling told me to hold the reset button and after a few seconds the machine lit up!! Indicator leds on IP7 and IO2 blinked in a sensible pattern and all fans in the card cage spun up and coughed up a bit of old dust in my face. The second time I powered it up I had to do the same (hold the reset button) but now it seems to start every time I power cycle it.

Then I tried for an hour or so to hook up a serial connection. But neither my laptop with USB to Serial dongle or my trusty VT220 got anything, not even garbage :/ Not sure what to do next, perhaps the check the SGI serial port pinout or give up on serial and put the graphics back in.

Any suggestions on how to get to the serial console? Is it the port labled 1A or one of the other that is the console?

jan-jaap wrote: bla bla ESD precautions bla bla. You know the drill.

I even wore a wrist strap!

Pontus wrote: I tackled the 4D tonight. [...] now it seems to start every time I power cycle it.

Awesome!

Pontus wrote: Then I tried for an hour or so to hook up a serial connection. But neither my laptop with USB to Serial dongle or my trusty VT220 got anything, not even garbage :/ Not sure what to do next, perhaps the check the SGI serial port pinout or give up on serial and put the graphics back in.

Any suggestions on how to get to the serial console? Is it the port labled 1A or one of the other that is the console?

Yes, console should be on the 1A, the top one under the keyboard port.
Serial pinout is not the same as a PC! Refer to techpubs for the pinout: http://techpubs.sgi.com/library/tpl/cgi ... 7/serial.z

Also, the ports on the system are female, where PCs are male. You'll have to make your own console cable. Null modem (TX, RX crossed), 9600/n/8/1.

Also you need a decent serial port. A friend brought by a Challenge L to our local retro-computing group and we tore apart his serial cable before trying another serial port. Turns out my USB serial port made enough power but his laptop wasn't able to deliver enough volts for the Challenge to register input. We kept seeing the menu but it wouldn't take input from the terminal.

I haven't search my drawers for serial cables and adapters just yet.

But I read on sgistuff and noticed the backplane layout. Slot 1-5 is VME, 6-7 is CPU and 8 is MEM. My system has memory in slot 7. Which is correct?

I sincerely hope that this hasn't destroyed anything :/

Pontus wrote: I haven't search my drawers for serial cables and adapters just yet.

But I read on sgistuff and noticed the backplane layout. Slot 1-5 is VME, 6-7 is CPU and 8 is MEM. My system has memory in slot 7. Which is correct?

I sincerely hope that this hasn't destroyed anything :/

For a single tower (Diehard) chassis:
1...4 are VME
5...8 are the PowerPath bus
9...14 are the graphics bus

VME slots are normally filled left-to-right, the exception is the IKON (genlock) board which must in in VME #4

The IO2/IO3/IO3B must be in slot 5 (and physically won't fit anywhere else)
Normally CPU boards are in slot 6 and 7, and memory in 8.

I happen to have a spare Diehard backplane here and had a quick look; I have the impression slots 6 .. 8 are functionally identical. I'd still move the MC2 to slot8, though.

I still have a couple of MC2 boards so don't despair

I had just connected another adapter between my vt220 terminal and the 4D. It might have been connected for a minute at the most.

The area close to the fuse is squeky clean. No blown caps, charred IC or burnt PCB. No smell or smoke when i happened.

I'm at the limits of my electronics knowledge so I'm not going to take any chances or connect anything before getting some more help. I'll see if I can get som help locally.

It looks like there's a ton of Silastic on the transformer wires, that's going to make it hard to see if components are burned.
One thing about RS-232-C serial circuits is that they use higher voltages than just about anything else in a computer. The -12V DC rail is chiefly to supply power to the serial drivers and receivers, along with +12V. So perhaps there is some hidden path between the DC fans and the serial ports? Which module is -12V DC output from on the supply?

I looked at the pinout of the combination of serial adapters that I used. The chain was IRIS 4D <-> D-sub 9 gender changer <-> D sub 9 to MMJ adapter <-> MMJ DECconnect cable <-> MMJ adapter to D sub 25 <-> D-sub 25 gender changer <-> VT220.

The pin connections are as follows: (left is the computer, right is the terminal)

• TD 2 <-> 3 RD
• RD 3 <-> 2 TD
• CTS 5 <-> 6 DSR, 8 CARRIER DETECT
• NC/GND 6 <-> 20 Data Terminal Ready
• GND 7 <-> 7 GND

I looked at the PCB where the serial port is connected and saw that pin 6 which is not used in the SGI pinout is connect to ground while the VT220 gives about 5 Volts out (I just measured).

I'm thinking/hoping/wishing that this could have caused the fuse to blow somehow, but I suspect that only shows my limited knowledge :/

Last weekend I spend a few hours with my friend Mattis and with his help (well... I mostly watched and posed tons of questions) we could conclude that at least two components are not operating properly. The power MOSFET "BUK456-1000B" and the switching controller "UC3844N".

I've ordered replacement parts and I hope to get them before Christmas. OH CRUD!! I forgot to add the fuse to the order... where is the facepalm smiley when you need it :/

Anyway debugging wasn't all that difficult, mostly thanks to Mattis experience and Jan-Jaaps invaluable help but also thanks the the modular nature of the PSU. It was just a few screws to get the backplane out and we could apply a small voltage to drive the circuit on the bench and watch the switching wave form on an oscilloscope or rather the lack of a waveform since the circuit didn't work.

Here is Mattis in action: