[CB_HK]

I’m looking for assistance from anyone who might be familiar with troubleshooting Max IMPACT issues.

I’m attempting to boot an Indigo2 R4400 IMPACT system with a Max board set installed. The issue presents as the computer starting, playing the boot tune, and then hanging (led green to solid orange).

I can tell the machine is attempting to get the graphics running: the screen pops up with signal detected and then goes blank about three seconds later. This continues to happen over and over until power is removed.

I have so far accomplished the following troubleshooting:

1) Pull and reseat everything in the machine
2) Disassemble the board set (wearing ESD gloves)
2b) Clean all contacts, inspect every board for missing SMD components, reassemble and ensure all boards are mated (Note: No TRAMs are present or installed)
3) Swap in a Solid IMPACT board: machine boots with no issue
4) Swap in an Extreme board set: machine boots with no issue
5) Machine passes IDE with Extreme board set installed

With all that said I’m a bit stuck. Is it a common failure mode for the board set to just stop working?

I’m hoping someone might have experience with resurrecting this set as I’d like to migrate it to my R10K IMPACT system.

Thanks!

[Trippynet]

Do you have any other cards fitted to the system? Also, are you sure your PSU is beefy enough to handle the MaxIMPACT card?

[CB_HK]

Do you have any other cards fitted to the system? Also, are you sure your PSU is beefy enough to handle the MaxIMPACT card?

I do not have any other cards installed. It's just the graphics set at the moment.

I've been referencing information from Obsolete Hardware:



However, your question about whether or not the power supply was capable of handling the Max IMPACT set had me question whether or not the PSU would in fact run it due to age.

The R4400 I'm using has a 060-8001-001 PSU installed. It should be able to run the set without issue. Even so, I decided to try it in my R10K system.

The R10K has a 060-0027-002 PSU installed. It's known working and has been solid for me since acquiring it. Interestingly, rather than the display flipping between signal and no signal, like on the R4400, it now stays steady black with a signal detected. Same issues, otherwise: solid amber LED and nothing after initial boot tune.

I'll also add that the system is not getting far enough to post information to the terminal. It's almost as though it's hanging during diagnostics.

Lastly, the board set is definitely getting power to all three planes as they are warm to the touch (ESD gloves on) and I have confirmed with an IR thermometer for tough to reach sections in the middle.

Best guess I have at the moment is that one or more boards is stuck and not working 100%.

[Irinikus]

Have you tried the card in both positions that it can be slotted into the GIO64 bus?

[CB_HK]

Have you tried the card in both positions that it can be slotted into the GIO64 bus?

I have. I used the bottom slot as well as the one above it.

The User's Manual for the IMPACT Indigo2 indicates that a solid amber LED with no display is due to a graphics board failure. Which, if the board is truly broken, will be a real shame.

[weblacky]

Boy that sucks and I don't think this the first Max Impact we've heard about dying. It has been suggested that MaxImpact boards get too hot and undergo thermal expansion over time and end up with cracked solder joints. If you don't have the extra TRAMs (as you said) there are still built-in TRAMs that may or may not need reflowing.

But only the very experienced members (like Jan-Jaap) could tell you if the graphics will boot without working built-in TRAM and just show you artifacts or whatever. Assuming you have nothing else to lose, with the board you might try diode and Ohm mode measurements with a multimeter on the power pins (where the white connectors on the backplane are) on the boardset and see if anything looks amiss (local short).

I do fully understand it's a board sandwich and you're not going to get detail views of what's going on while it's being powered. I might suggest you try to find the header pins (that connect all the boards together in a stack) that transmit the raw power from the backplane to the internal headers and gently separate the boards, perhaps a 1V-3V limited current power injection on each individual board (outside the system of course) with a bench power supply might reveal a power-hungry board that MIGHT indicate a local short of some kind.

You could so just perform a very detailed visual inspection of each board and look for poor solder joints and attempt a gentle touch up of each one you find. I'd not image you'd want to initially fiddle with the large chips, I'm more thinking about the header connections and SMD passives (caps, diodes, resistors).

I'd cross my finger that between a low-voltage (current limited) power injection test (test for immediate draw) and an inspection for touch ups...perhaps you might find the issue.

Obviously if it's a real damaged IC then, the ONLY way we'll find that is using another (working board sert) MaxImpact and performing measurements on EACH IC leg and comparing to the same position on the dead graphics board.

[jan-jaap]

A couple of things:

1. The way forward are the 'ide' diagnostics. These have to be run from a serial console connection. First of all because the graphics tests would reset and claim the console, but also because because in this case the system doesn't boot to a graphics console. So get your console cable out, disconnect keyboard, mouse and see what shows up on serial.

2. The problem can be something trivial, like a dead Dallas chip causing scrambled NVRAM settings. With the settings for 'console' and 'bootfile' scrambled you will have an unbootable system with no display output as well. Again, serial will show.

3. It's quite possible that it will crash with a graphics error, or the IDE diagnostics fails with a graphics error. In that case you will have to get more familiar with the MaxIMPACT graphics. It's a sandwich of three PCBs, interconnected with flex cables. If you're lucky (?) it may have the TRAM option. All MaxIMPACT cards have one MB of TRAM, and with the TRAM option you have 3 more MB for a total of 4MB. The MaxIMPACT card set itself seems to be fairly reliable. I've never seen the flex cables come undone, but it doesn't hurt to double check. I've never seen the 1MB 'base' TRAM fail, but I have seen the heat sink come off once. Oh, on MaxIMPACT is basically HighIMPACT times two, like an NVIDIA SLI. So the base 1MB and the TRAM option are double as well.

The problem child are the TRAM option cards. They mechanically/thermally challenged. Let me explain: the TRAM chips themselves are QFPA packages. Like this: http://2.bp.blogspot.com/-BGVLK6EHxrI/VS.../comco.png  Three of them sit on the TRAM option PCB, and a heat sink is mounted to them with kapton tape. Kapton tape is heat resistant and doesn't stretch. The way this is done tends to bend the PCB which is really not what you want with these large QFPA packages. It puts pressure on the TRAM chips, which has to flow to the PCB via the chip legs. On top of this, they run very hot. The chips handle this fine, but the combination of thermal stress from many heat cycles as the system is powered up/down and mechanical stress from the design leads to brittle soldering joints and lifted chip legs.

This only affects the TRAM option. And this is how the MaxIMPACT fails in my experience: the card and the 1MB 'base' TRAM works just fine, the option cards fail, and not because the chips burn out but because of failed soldering joints.

Oh, the very act of removing / installing the TRAM is problematic to the TRAM itself. The TRAM has to be pressed into the main PCB, in other words: you press on the heat sink and the pressure 'flows' though the chip legs when it's inserted. The other way around is equally problematic: when you pull the TRAM option you tend to flex it, so if it wasn't broken already, it may now be. Or next week, after a couple of more heat cycles. In short: don't mess with these things unless you have to.

SGI probably realized this as well. The Octane TRAMs are much more rigid and use the heat sink as a structural element. They rarely break, though they require re-seating occasionally.

Oh, the TRAM option cards are supposed to be secured with a split pin. If someone added the TRAM option to your card later , they likely aren't there. It's near-impossible to undo them without breaking them. In this case, especially after shipping, your TRAM option cards may simply need re-seating. All the precautions (see above) apply.

If the 'ide' test reports graphics errors related to texture memory, memory test pattern mismatches or maybe DMA errors, in short anything that points at TRAM failure, you may decide to remove the TRAM options cards. You don't do this on a whim, but you may have no other choice. Chances are that the card will work fine without them. In that case: repair of TRAMs with lifted legs is possible, but not trivial. I've successfully repaired them, but I have access to a lab with good equipment and a microscope. It's a very time consuming exercise, it will take a couple of iterations of inspect / repair / run diagnostics. Ideally you should have a known-good HighIMPACT card set handy which enables you to test 'repaired' TRAMs individually rather than in pairs.

---

Update: I see now you don't have the TRAM option cards. Sorry about that.

Still, diagnostics over the serial line are the first step.

[CB_HK]

My apologies, I thought I had mentioned it in the beginning, I had a serial console hooked up when trying to get the machine to boot. I’ll re-evaluate and see if I need to try a different cable or if something else wasn’t set correctly because I was unable to get anything from the console with the Impact board installed. If the machine could boot all the way I could get to the console. Not sure what the issue there is.

[saq]

My apologies, I thought I had mentioned it in the beginning, I had a serial console hooked up when trying to get the machine to boot. I’ll re-evaluate and see if I need to try a different cable or if something else wasn’t set correctly because I was unable to get anything from the console with the Impact board installed. If the machine could boot all the way I could get to the console. Not sure what the issue there is.

Check your handshaking on the serial. I need to get my machines up again to check for SGI, but there are some computers that only work right with 3-wire for the initial console b/c the handshaking isn't properly done. Also curious if you have tried it with dual graphics boards and the gfx console plugged into the second Solid/XL.

Jan-Jaap, your comment is getting me thinking about pulling out the one from mine to look at "slitting" the heatsink similar to what they did on the IP30 revision to reduce stress. IT would be an operation but might be doable. IIRC the PCB is pretty thin on those texture modules though so insertion/removal might still be a problem.

CB - how did the serial line look with the different graphics installed? If it isn't coming up at all until IRIX is loaded it might be the 3-wire vs hardware handshake issue. If it does come up it might be something with the HQ on the card that is hanging during diags and not returning either pass or fail, though you'd think it would timeout at some point.

[CB_HK]

My apologies, I thought I had mentioned it in the beginning, I had a serial console hooked up when trying to get the machine to boot. I’ll re-evaluate and see if I need to try a different cable or if something else wasn’t set correctly because I was unable to get anything from the console with the Impact board installed. If the machine could boot all the way I could get to the console. Not sure what the issue there is.

Check your handshaking on the serial. I need to get my machines up again to check for SGI, but there are some computers that only work right with 3-wire for the initial console b/c the handshaking isn't properly done. Also curious if you have tried it with dual graphics boards and the gfx console plugged into the second Solid/XL.

Jan-Jaap, your comment is getting me thinking about pulling out the one from mine to look at "slitting" the heatsink similar to what they did on the IP30 revision to reduce stress. IT would be an operation but might be doable. IIRC the PCB is pretty thin on those texture modules though so insertion/removal might still be a problem.

CB - how did the serial line look with the different graphics installed? If it isn't coming up at all until IRIX is loaded it might be the 3-wire vs hardware handshake issue. If it does come up it might be something with the HQ on the card that is hanging during diags and not returning either pass or fail, though you'd think it would timeout at some point.

saq,

Thanks for the suggestion about the 3-wire serial cable. I ordered one up that matches the Indigo2 service manual. Hopefully I can get it to work that way.

To answer your question about two cards, I did attempt to boot with a Solid IMPACT card in the bottom-most slot with Max just above, video coming from the Solid card. All I got was a fuzzy gray display and the same amber light that never changed. I would guess that as soon as I can get the system talking on the serial port with the correct gear then I should be able to hopefully see what the issue is.

I'll update soon when I have a chance to crack into it!