3

I recently acquired an MSI Radeon R9 280X video card that experiences visual glitches/flickering patches/artifacts in games and in 2D mode. Once I tried to open a video in fullscreen mode from Youtube and the screen just went entirely flickering black until I restarted.

I tried the common fixes of R9 280X/7970 artifacting - underclocking memory, undervolting the card, underclocking the core, increasing the voltage, updating the BIOS, increasing the idle clocks - nothing helped.

Basically the card is bad. Under the assumption that I cannot RMA the card, what are the possible uses of a video card that produces graphical artifacts? Can it be used for Bitcoin mining? OpenCL computations? Folding@home? Sold for scrap parts? Basically I need to salvage this card in any way possible.

I ran a GPU memory error checker (best answer from this question: How can I test my GPU memory/RAM?) and it reported no errors but I'm still wary.

I still haven't put the card in the oven or freezer.

Improve this question
11
  • 2
    No; It cannot be used for anything. Why can't you RMA it? The R9 280x hasn't been released long enough for it to be out of warranty.
    – Ramhound
    Jul 25, 2014 at 13:47
  • 1
    Did you tried the card at other pc?
    – Watsche
    Jul 25, 2014 at 13:48
  • 1
    GPU chips make pretty nice keychains :)
    – gronostaj
    Jul 25, 2014 at 13:50
  • @Ramhound It's a long story involving a dispute with the Amazon marketplace seller who sold me the card. Luckily Amazon covered the money but I still have the card.
    – Sevag
    Jul 25, 2014 at 13:51
  • 1
    You normally don't go to the seller of the card to handle warranty claims.
    – Ramhound
    Jul 25, 2014 at 13:55

2 Answers 2

Reset to default
0

Can it be used for Bitcoin mining? OpenCL computations? Folding@home?

Yes ^ it should be possible. BOINC as well. Only way to know for sure is to test it and see if the computes fail or not. With BOINC you would be able to assign all GPU workloads to the artifacting card and zero to your primary card.

Have you tried beta drivers? Have you monitored the temperature when in use?

Improve this answer
6
  • Temperatures are OK. ~60C during benchmarks, 80C in Furmark IIRC. I haven't tried new drivers yet. I ran the Furmark donut test and that produced no artifacts. I also ran the OCCT test with memory error checking with no problems or artifacts - it feels like they're random because when I try to force them out with stress tests/error checking, I don't see them. I'll try messing with the drivers.
    – Sevag
    Jul 25, 2014 at 14:05
  • 80C is approaching the limits or right at them. Usually good to target around 70C +/- 5C with a fan profiler. Unfortunately I am not familair enough with AMD GPU's to know the options but you may want to look for something like EVGA Precision for NVIDIA cards that allows you to set an adaptive fan profile based on the temperature. Doing that alone I can keep my card about 7-10C cooler than it would otherwise be at during intense games. 60C is certinaly fine though so maybe nothing to worry about here. Is it all games that seem to artifact?
    – Enigma
    Jul 25, 2014 at 14:09
  • I've got an NVIDIA card that will regularly crash with some games but not others at stock clock speeds. If I underclock memory by about 230mhz and shaders by about 80mhz it will never crash. It could be that you just need to underclock some more or find the safe values.
    – Enigma
    Jul 25, 2014 at 14:13
  • If the card is producing artifacts, it is all but certain that it will produce calculation errors. If it is not reliable when doing the calculations for the graphics, it's pretty much guaranteed not to be reliable when doing the calculations for bitcoin mining, as this uses the exact same hardware.
    – ChrisInEdmonton
    Jul 25, 2014 at 14:15
  • I can overclock an old GPU to the point of producing artifacts when in games and still get successful computes in BOINC. It's a different process and not 1:1. For one thing, it's not displaying or rendering anything on screen.
    – Enigma
    Jul 25, 2014 at 14:20
0

AMD uses ECC on newer cards and if any error occurs, it resends data or recomputes until it reaches target.

When there is so much instability, it stucks in an infinite re-send loop or bsods. So it is very hard to find a point which it is %0 error compute frequency. An error could be present even at 1 MHz(not even GHz) working frequencies because of quantum-mechanics' tunneling probability. Anything can happen in nano world at anytime.

Only you can know if error happens.

Lets say r9-280x starts artifacting %99.99 at 2GHz(before it explodes) .

@1.5 GHz, it is %90 per cycle (just trivial numbers to have an idea of tunneling)

@1.3 GHz, it is %10 per cycle (trivial again), ECC could stand for a second before getting in bsod or smt like that

@500 MHz, it is %0.01 per cycle but ECC takes care of it and performance impact is minimal.

@1 MHz, it is %0.000000000000001 but it is there and will not vanish. ECC makes it nearly impossible to lock your game/compute.

If gaming cards were %100 stable, there wouldnt be a need for server-grade components to run a web site 7/24.

You just need to watch out for decreased gaming/computing experience when crossing factory frequencies.

If it is stable for 24 hours, it may stay stable for another 24 hours(or not).

So, if it is not stable for 24 hours for any frequency range, then you should not use it for anything.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .