Thinking of slapping together a couple of Radeons to power a 4K display or a multi-monitor setup? Yeah, you might just want to hold off on that.
The folks at PC Perspective have conducted some new testing using an updated version of Nvidia's frame capture tools. Their results show that serious performance issues arise when CrossFire multi-GPU configs are paired with multiple displays—or 4K monitors, which the graphics hardware treats as dual monitors.
In addition to the runt frame problem we and other sites detected earlier this year, using a CrossFire setup to drive a 4K monitor or Eyefinity config introduces another issue: something PC Perspective calls "interleaved frames," whereby multiple slices of a single frame are shown on top of another.
That issue is apparently more visually distracting than regular screen tearing or runt frames. Not only that, but when frame capture tools are adjusted to compensate for interleaved frames, the resulting performance numbers—which are more representative of perceived performance—show very little scaling from one to two GPUs in games like Battlefield 3 and GRID 2. That means adding a second Radeon barely improves performance on a 4K or Eyefinity setup.
This may come as no great surprise to some. After all, when AMD addressed its multi-GPU microstuttering problems with a driver update last month, the company clearly stated that the fix, known as frame pacing, only worked with single-monitor setups at resolutions of 2560x1600 or less. PC Perspective asked AMD specifically about the interleaving issue this week, and it was told that the company has "committed to addressing these cases in a future update." AMD added, "Details of the rollout for this solution will become very evident, and made public in the very near future."
Translation: a fix should be forthcoming.
Still, this problem is bad news for CrossFire users hoping to upgrade to a higher-res monitor—or to grab a couple of extra ones. 4K displays are already selling at Newegg, and it seems AMD isn't quite ready for them.