1

I have got a faulty Radeon HD 6850. I was only told that it stops working after windows boots and that the former owner is going to throw it away if I don't take it. So well, now I am trying to resolve the problem:

What I have observed is quite interesting:

  1. I tried the card with Ubuntu without any extra drivers. It works under this setting without problems. Haven't tried gaming, nor instaling any third party driver.

  2. I installed windows xp and it worked as long as I did not install the drivers. After installing the drivers windows 'tries' to boot, but as soon as the welcome screen should appear the screen turn like half of the aspect ratio and stays black.

  3. I have observed that BIOS does show me wierd characters all over the place. And I investigated this problem further:

    • it displays an y instead of an i
    • it displays an u instead of an e
    • ... if you follow this pattern, and if you check the binary ascii codes behind it you will see that there is a pattern behind it:

-> in all cases the fourth bit is always high, but it should be low instead.

It could be a faulty memory of course, but my idea is that it is as simple as a faulty communication line. It seems that i have tested three different operating modes. Like a basic 'failsafe without driver' mode which seems to work flawless.

A text only mode which seems to work in some points:

  • it displays the characters right in term of the ascii code (if I would like to write an 'a' it would display an a - not chopped or anything)
  • it seems that the memory which saves the distribution of characters on the screen also works without problems. There is no flickering or some stuff like that
  • but it does not write the right character into the memory. so if the bios tolds the gpu to write an 'e' to the 'display memory', then the card writes an 'u'

And there is this third mode, which is the 'drivers installed mode'. And my idea is now that the OS is trying to put the GPU in some kind of 'operating mode'. It is trying to do an initialization sequence or something like that. And this fails because of this communication error which I mentioned earlier. But I don't think that there is anything written to the 'display memory' in this phase.

So well, I think that the problem is pinned down to a 'small' area. It has to be the part where the signal comes from the bios to the 'processing unit'.

The problem is that I have no idea of the technical realization of this communication. I would be able to resolder some stuff, but I actually don't know where to look for some faulty stuff.

Therefore i would be really happy if someone could give me some hints or some resources to learn this stuff. I am also happy if a vivid discussion develops about anything mentioned. Every input can be valuable.

Thanks! ;)

dmi
  • 11
  • 1
  • 3
    I wish fixing computers was like that but I can tell you nobody will have a clue! computer technicians just swap stuff out when it's not working. And calling electricians won't help either, 'cos even if some could fix it in theory, they wouldn't want to and they'll claim to not have a clue. And even electricians nowadays just swap boards! – barlop Jan 24 '15 at 00:53
  • Your best bet is going to some electronic specific forums, and asking the hobbyists there. You deserve a (small) metal for figuring out the ascii pattern though! – barlop Jan 24 '15 at 00:54
  • I'm in agreement with Barlop (although s/metal/medal/). Your issue might not be a communications channel; it could be bad memory. Perhaps bad memory on the card, which causes the data corruption before communication occurs. Your issue might be inside a chip, or a communications pathway. It might not be visible with a naked human eye. Your best bet may be to use the drivers that are actually functional. That may not be a way to get optimal performance from the card, but the card may still be good enough to be a dumb terminal useful for web browsing. – TOOGAM Jan 24 '15 at 08:10
  • http://superuser.com/q/333617/10165 is a similar issue, but it was with an nvidia card. In that specific model's case, reflowing the chip would have fixed it. – Journeyman Geek Jan 24 '15 at 12:56
  • Unfortunately today's computer hardware is proprietary and manufacturers don't want to disclose how their boards work (unless you're a big company, agree to sign an NDA and its profitable for them to give you that information), as opposed to back in the old days where you can get a schematic and know exactly how that hardware works. –  Jan 24 '15 at 14:09
  • What about trying to buy the exact same card new, and pretending to have got a defective one by sending this one instead ? That's kinda mean but it may work. –  Jan 24 '15 at 17:36

1 Answers1

1

First step would be to examine the card thoroughly for any visible physical damage.

This could be along the lines of:

  • bent or broken 'pins' on the PCIe connector
  • scratches or damages to the PCB (circuit board) that could be causing a short circuit
  • visible signs of heat damage (burnt parts and/or melted plastic)
  • loose components (not likely in this case)

Then I would try the GPU in PCIe x1 or x4 mode to see if that helps. Your mainboard might not support setting this manually in BIOS, however.

Try alternative drivers if you can find any, to see if it could be a software issue. With your description it is a long shot, but at least easy to do.

Good luck with the patient!

Kristian
  • 3,102
  • 13
  • 21