Back in January I posted a review of Gigabyte’s latest GTX 560 Ti card. My opinions of the card have not changed during the past 4 months, and even now have not changed.
Sadly, however, starting sometime this week my card began to exhibit very strange problems — but only on 2D (or presumed to be 2D) surfaces. It’s not something easily noticeable in a 3D-based game, but playing something like Dwarfs!? made it apparent within seconds. Then I tried Two Worlds (which uses a 2D texture for its sky) and noticed occasional problems there too.
Later, I ran Misha’s Video Memory Tester application which confirmed that the likely source of the problem was bad GDDR5 RAM on the card, or a bad bus/path between the RAM and the GPU itself.
Here are some screen shots of Two Worlds, Dwarfs!?, and Video Memory Tester. In the latter case, you can even see some discoloured pixels in GDI, confirming the issue is with RAM on the video card and not system RAM:
- Dwarfs!? #1
- Dwarfs!? #2
- Dwarfs!? #3
- Dwarfs!? #4
- Dwarfs!? #5
- Dwarfs!? #6
- Two Worlds #1
- Two Worlds #2
- Video Memory Test results #1
- Video Memory Test results #2
I tried doing things like decreasing both the core and memory clock on the card to no avail. Temperatures appear to have no bearing on the failure; the card can be running extremely cool or extremely hot and the problem occurs with the same regularity. Using tools like Furmark worked fine, ATITool’s artefact tester is useless since it gives false positives (search the Web for validation of that statement), and EVGA’s artefact tester only works on EVGA-branded cards.
I’ve put in an order for an MSI N560GTX-TI Twin Frozr II to hold me over while the GV-N560OC-1GI is being RMA’d with Gigabyte.
It probably doesn’t help that these cards are slightly overclocked from their reference specs either; too bad nobody’s selling Gigabyte’s GV-N560UD-1GI, which is a GTX 560 Ti with reference specifications (no overclocking).
Misha’s Video Memory Tester is ultimately what convinced me the problem was with the GDDR5 RAM. All the errors shown indicate primarily a series of single-bit errors (e.g. I wrote $FF00 (%1111111100000000) but what I got back was $EF00 (%1110111100000000)), and some multi-bit errors. Since the error locations are mostly within the same region (I had to look at the error log to confirm this), this would appear (visually) as anomalies in a game (texture problems or random bits not being set in texture/RGB data) — or, in the case of GDI, certain pixels being mangled on the desktop.