[GanjaTron]

Hi everyone,

while replacing the dead batteries in my Crimson's DS1216B NVRAM with an external CR2032, I started getting progressively weirder issues with the Reality Engine such as (see attachments):

• The PROM console sporadically complained about an invalid EEPROM or failure to open the graphics console gfx(0)
  
• An MP bus error right after booting (never had that before!)
  
• Corrupt graphics when the console did occasionally initialise.
  

After swapping the DG2 with a spare, I got a sane picture again, so I'm wondering what happened. Has anybody seen something like this, and where exactly is the mentioned EEPROM located, and what does it contain?

The DG2 has only one socketed chip, and it's an 8-pin Xilinx 1765DPC PROM which stores the FPGA configuration. The VPP and VCC supplies are tied together, so the chip is hardwired for read-only operation, and must be externally programmed, which doesn't make it an EEPROM. Is anyone familiar with this chip and what happens if the contents are corrupt? Or should I be looking elsewhere?

Thanks for any infos, really scratching my head over this one (again).
Damn, these things are high maintenance! 

--Roland

[weblacky]

I don't own a system like this, but I've heard this before...

I think you're dealing with this: https://wiki.preterhuman.net/InfiniteRea...M_Commands

I was under the impression that this is a real EPROM so I think it has to do with the installing OS routine needing to set some graphics board parameters, if you added the graphics board later (after OS install) this likely didn't happen?

This is as far as I can lead you, good luck.

[GanjaTron]

There's a similar tool to ireeprom called dg2prom that lives in /usr/gfx/ucode.

P.S. Strangely, "dp2prom -i" reports the ucode state in the EEPROM as "good"...

[weblacky]

I didn’t write anything about a dg2prom. But regardless an eprom that’s okay may mean it’s physically okay, not the data values inside. The page I showed above claims a clear command is often used when a corrupt or black screen in graphics mode appears.

I assumed you need the valves reset.

[GanjaTron]

I didn’t write anything about a dg2prom.

Sorry, apparently my initial reply got trashed during the edit...

On RE there are tools called dg2prom and flashprom in /usr/gfx/ucode which do something similar to ireeprom. I believe the latter specifically pertains to the 28F020 flash ROM on the GE8, while the former presumably deals with the 1765DPC on the DG2. However, dg2prom's terse help doesn't suggest it can clear/reset the (ee)prom.

[GanjaTron]

So a quick update: last night I checked the suspect DG2 again and noticed that the messed up image reacts whenever I wiggle the paddleboard connecting the DG2 directly to the RM4(s). I got all sorts of crazy stuff, including a completely absent output and an illegal RM4 termination message I forgot to mention in my original post. That said, the flaky connection could also lie in the backplane. In any case, it's not really confidence inspiring. This crate has been around a bit...

--Roland

[jan-jaap]

Ah, the wonderful world of the Reality Engine ... I've decided to never remove any of these boards from my Crimson unless I have absolutely no other choice. There's too much flex in the PCBs, and at the same time they're filled with large QFP chips that will lift their legs from the PCB if that happens, especially now that the soldering joints become brittle with age. Same for that front interconnect.

The older stuff (VGX) was all through-hole tech on thicker PCBs and is pretty much bullet proof. The later stuff (Onyx IR) has metal reinforcements of the PCBs and a much more sturdy interconnect. More reliable, especially the Onyx2 IR boards. But the RE and RE2 are the weak link, I'm afraid ...

[weblacky]

Ah, the wonderful world of the Reality Engine ... I've decided to never remove any of these boards from my Crimson unless I have absolutely no other choice. There's too much flex in the PCBs, and at the same time they're filled with large QFP chips that will lift their legs from the PCB if that happens, especially now that the soldering joints become brittle with age. Same for that front interconnect.

The older stuff (VGX) was all through-hole tech on thicker PCBs and is pretty much bullet proof. The later stuff (Onyx IR) has metal reinforcements of the PCBs and a much more sturdy interconnect. More reliable, especially the Onyx2 IR boards. But the RE and RE2 are the weak link, I'm afraid ...

Jan-Jaap - Out of question (since I know you're an actual workshop with real equipment), has doing manual reflowing (with large area board preheating) on every component (slowing working your way around) on these huge boards ever yielded any fix?  I've seen you give the "wobbly board with broken joints" warning several times throughout the years, have any of them ever been brought back from the dead or dying?  It seems like large-graphics inventory is getting lower and lower (find another) that fix would become more necessary.

Also is there any hope for a space on the edge of these large boards for some kind of reinforcement (metal frame or the like) to be added to working boards to rigidify them?

[GanjaTron]

Hi JJ, lone time no see/hear! Good to see you're still around here!

RE is indeed extremely sensitive and fiddly, as SGI transitioned from TTL to SMD. I didn't even touch the RE but apparently vibrashunns from mucking about with the IO3B is all it took. We had similar experience with the RE2 in our Onyx back at Fraunhofer.

I might take a stab at reworking those solder joints for that front interconnect, but I'd need to find some fine gauge good old Pb-based solder first after running out of the stuff. I assume this is way before RoHS conformance.

---

I've seen you give the "wobbly board with broken joints" warning several times throughout the years, have any of them ever been brought back from the dead or dying?  It seems like large-graphics inventory is getting lower and lower (find another) that fix would become more necessary.

The connections on those boards are definitely flaky, and there's considerable wobble. Reflowing is worth a try, tedious as it is, but I agree this needs to be done with proper equipment without damaging the board. I've reflowed brittle solder joints inside my old Trinitrons (with success), and those probably run way cooler than an RE. Of course the density of those joints was way lower.

AFAIK these boxes originally shipped with horizontal plastic (?) supports mounted across the top and bottom of the cardcage, which I believe were meant to apply rearward pressure to the boards. I've never seen one in person, and they're missing in both of my chassis.

Thanks for the feedback! Regards,

--Roland

[weblacky]

One thing I've learned from doing old electronics repairs, unless you have to reflow, don't literally reflow. By reflow I mean, suck, clean, reapply.

Many of PSU joints I've dealt with and old Indy mainboards I'd replace caps on (so small sample) have had environmental corrosion start on the all the joints. While it's just surface, if you literally just reflow/reheat/remelt the same solder it will compromise the alloy and be crap (long term) as that corrosion mixes in with the molten solder.

I also always use kester 63/47 leaded solder...always. I've found it better to desolder (but not really remove any component), suck the free solder away, clean and re-apply fresh solder/flux when repairing connections. For chips that have internal ground regions and/or BGA/QFN packages...you don't have much choice, so just reflow. But any component with through-hole pins or long enough legs (like QFP), I'd try to suck/wick solder and replace what I could, one side at the time (to prevent shifting/movement of the IC). You won't get it all, but you could do a desolder wick hit on one side, then add more solder to refresh...won't get it all but maybe get some and hopefully get the surface corrosion off onto the wick before it penetrates the actual molten joint.

I know a lot of the SGI connectors are really through-hole. So after preheating the board to a comfortable level, you could suck several joints, reapply, suck the opposite end of the pin row, reapply, go back to first side, do more, etc. I've been told to help prevent connector melt, if you have lots of connections (through-hole) to do, jump around like the star pattern you use for apply lug nuts to a car tire, don't cluster them because it focuses heat, if you skip around and come back it's supposed to help prevent local heat build up that can damage the plastic of a connector because you went linearly instead of jumped around it spread out the heat duration/intensity.

But it would be a labor of love...no doubt.