[Nichosen]

Awesome work

[weblacky]

That's very good write-up and great results! I'm just a little lost so I'd like to back up, if I may. What gave you the idea that the SRAMs were related to the IC being called out in the diag? You talked about corrosion, did the corrosion kind of lead you there or was there other info/input that gave you the idea to look at the SRAM chips?

Did you find there were connected in a conductivity testing between the IC legs and the legs of the SRAM chips?

I'm just trying to understand your process...I don't think I'd have made that leap so quickly. I'm trying to learn what brought you there.

[Irinikus]

Nice work!

[matze584]

Hi Weblacky!

That's very good write-up and great results!

Thank you much!

What gave you the idea that the SRAMs were related to the IC being called out in the diag?

I've read the Octane Technical Report (the one with Impact graphics) to get some clue of the architecture of the Impact graphics series. I remembered VC3 from this paper, as i read the error message. I was hoping to find this Chip and analyze the issue further down from there.
I am into hard- and software development for custom industrial automation and control for a living (and for life!!).  I started with my electronics hobby at age 10 and got a leg into retro computing by designing/building a Z80 based CP/M System from scratch using enamelled copper wire and Xilinx CPLDs for the custom VGA and IDE interfaces at age 15 (took me around 3 years onwards to gather all skills for this though!).  So i may have developed an eye (and some intuition) for what and how i am searching and how i track down problems systematically. Including what to look for, if it says "SRAM", how it looks physically and what it does technical wise. And gladly, this SRAM error was quite well to find due to the fact, that one (and every time the same) half of a 16 bit word was failing.
BUT!!  It was a great portion of luck involved. Much of this was possible by lucky circumstances. If any of the BGA chips had failed, it would have been hands down impossible for me. All the other SMD parts are day to day business, but a BGA would have ended my hopes to get it up and running instantly. Not to say, impossible for me to find the majority of other failures deeper inside the pipeline because we have no schematics anyways. As such i was happy to experience this particular failure rather than any other else My old SolidImpact was such kind of thing. I can't do anything about it. No informations, no readable error messages, nothing. no clue where to start: dead...

Did you find there were connected in a conductivity testing between the IC legs and the legs of the SRAM chips?

After finding these two chips be SRAM in general, i did some conductivity checks between VC3 and the control signals of the SRAMs.  They were indeed connected to each other.
Unfortunately this didn't tell if the SRAMs were the problem or if the VC3 is faulty. My first theory was this: If this dust had conductivity to it, it may have pulled from VCC to /WE on the SRAM. This pins are located next to each other on Pin 28 (VCC) and /WE (WriteEnable) on Pin 27. This would not have damaged the SRAM itself. It is used to get driven from outside onto WE, as this is an input signal. But it most likely didn't got written at all. I can tell this, as the errornous half reads differently by a certain pattern in the message. If it would have been written, it would be more random than just 0x00 and 0xFF. Most/Some SRAMs have a property, to have a distinctive pattern of 00 and FF after power up. This is what we can see in the error-log on the console. All upper halves are fine, and all lower halves show a regular consecutive pattern of 0xFF and 0x00 on a per address basis. This also shows, that it can (and gets) be read. If the Databus of the SRAM was faulty or the ReadEnable was the problem, the lower half output would have been all zeros or all ones, depending what way the designers of VC3 pulled the bus in case of tri-state on the data lines.
So i suspected the VC3 itself first. It's output drivers would'nt be very comfortable with a (strong?) pull to VCC, that might have been there due to the conductive dust.
But i then found both SRAM chips control signals beeing connected to each other. This leads to the fact, that both SRAMs are read and written at once. Knowing one of them to deliver back the data correctly, the "VC3-Pindriver-is-dead-theory" can most likely be set back for later. After i got this far, the second theory: it is also possible, that the SRAM itself got a problem and simply does not react to write enable anymore.
With this beeing the defect most likely now, and the easier task to do, I decided to swap both SRAMs over from the Solid and took the system for a test. It came up on the first try!
I am now going to recheck the old SRAMs and try to find out what happened to this SRAM to fail and see what it does in the component tester.

But for now, there is a whole new thing: Learning IRIX
Installing the Demos was an experience for itself. Fighting network settings and depecies. Got some of them now, but i need to find out how to setup my network properly. This is my anti-boredom-activity for today and i am scrolling through the web and this forum at the moment :-)
Currently it won't install the Indigo2Impact Demos because of a missing Performer package, doesn't mind installing the Octane demos, etc. I may be a hardware guy, but i definitely have no clue on IRIX yet If just everything would be as good described as the Reanimator IRIX install! This was a great step by step experience :-) I am going to start a new thread, if i dont get it with the existing infos, if i may :-)

Cheers!

[weblacky]

I appreciate the expanded explanation.

I totally get the pattern hole theory, but your reading the architecture overview for Octane was an interesting tidbit. I’ll see about doing that myself.

In terms of troubleshooting your SolidImpact, I think only comparative testing (what I do) would work. Find an exact duplicate card, that works, and measure similar points on both cards with various tools to “find the difference”. I bought a used Huntron 3200S for this purpose on my SGI collection specifically! But diode and resistance measurements with a high quality multimeter might be enough to found differences to investigate.

Yeah, the real reason I can even work on much of this stuff is it only uses BGA for a few things. Otherwise SGIs chip/bus speeds didn’t require BGA due to high frequency noise issues, they could use packages that had legs and therefore a human being can solder them (even the SDRAM main memory on these predates the BGA require due to low clock speeds)!

Also, later SGI’s have a large number of test points around nearly every chip so you don’t even have to sneak a probe into the legs that often. In these graphics cards you don’t have good test points but in many of the newer systems, where the chip density is lower, they called out a lot of the legs with test points that they used for quality control/assurance.

We’ve had comments from years ago about how the Indigo2 graphic card series had a lot of heating and expansion issues with its TRams or similar add-ons. No one‘s talked about the boards themselves directly but the fact that there’s a lot of legs and so little solder does lead you to wonder how much of it is simply cracked solder joints, or the like, instead of actual component failure.

[chulofiasco]

- I also recapped the powersupply completely even before the first contact to the grid. Four of the HV-Board electrolytics were dry. 

Do you have a bom of what you used? 

Could you share notes?  What you did to the broken MaxImpact is great.  🔥🔥

[matze584]

- I also recapped the powersupply completely even before the first contact to the grid. Four of the HV-Board electrolytics were dry. 

Do you have a bom of what you used?
Could you share notes?  What you did to the broken MaxImpact is great.  🔥🔥

Thank you! :-)

Yes, i made a list of all capacitors, that i purchased for replacing the electrolytic capacitors in my PSU.  It is a Zytec unit.  I also took a picture of both of the PCBs and added the values as overlays for reference.   The part numbers in this list are for the Mouser webshop. I replaced most of the caps by Panasonic FR type, because those are reasonably good allround Low-ESR types with 105°C rating and have a long lifetime specified.  The Wurth capacitors i used were in stock at my workplace, so i checked datasheets against the original ones and found them to be a good replacement.  Please note, there is -dangerous- high voltage inside this PSU even if it was powered off for a longer period of time. Danger of fire if something is wrong and all the like. I cannot guarantee for correctness of my pictures and parts list for any other than my own PSU for which this Informations were written. It's on everyones own risk to use this infos or doing service on the PSU in general!  (Just to be on the safe side!) :-)

 

Code:

2x 647-LGU2D182MELC   Nichicon        200VDC  1800uF  35x47mm
7x 667-EEU-FR0J682SB  Panasonic FR    6.3VDC  6800uF  16x20mm
3x 667-EEU-FR0J331    Panasonic FR    6.3VDC   330uF  6.3x11mm -> They are much smaller physically then the original type!
1x 667-EEU-FR1A332B   Panasonic FR     10VDC  3300uF  12x20mm 
1x 667-EEU-FR1C332SB  Panasonic FR     16VDC  3300uF  16x20mm
1x 710-860160672002   Wurth WCAP-ATLL  50VDC     1uF  5x11mm
1x 667-EEU-FR1V331U   Panasonic FR     35VDC   330uF  10x12,5mm
1x 710-860020772010   Wurth WCAP-ATG5  63VDC    10uF  5x11mm
4x 710-860020572006   Wurth WCAP-ATG5  35VDC    47uF  5x11mm
1x 667-EEU-FR1V101    Panasonic FR     35VDC   100uF  8x11mm
4x 667-EEU-FR1V471    Panasonic FR     35VDC   470uF  10x16mm 

But should'nt we put this infos into a seperate thread or place? If someone else is looking for this informations, he may possibly not find it deep inside a thread about graphic card problems. What would be good practice? :-)

Cheers!

[weblacky]

Just a note, unless you measured/saw something very specific about this PSU, most SGI PSUs have bleed resistors across all the filter caps (not literally, but in circuit). You shouldn't have any voltage after a few minutes. If you use a thermal imager you can often see them light up and stay lit several seconds after PSU shutdown to identify them.

You can of course check this to make sure by measuring the voltage across the AC filter capacitors and you should see it dropping fairly quickly to near nothing. This isn't unusual for PSUs, things like CRTs never do this because the process of building up charge is huge, so they like to try to stay charged...switching power supplies normally don't.