Final Thoughts

As it is turning out, the capability for Kaveri-based Dual Graphics frame pacing is only portion of the change that is coming with the 13.35 driver release this month.  Tahiti-GPU Eyefinity CrossFire scaling is in play as are TrueAudio and Mantle possibly; it looks like January and February are looking to be more exciting for PC gamers than initially expected.  

Dual Graphics, previously known as Hybrid CrossFire, is a feature that has always looked great on paper.  Taking a processor with integrated graphics on it, in particular one with the performance of the AMD A10 APU line, and improving on that with a low cost discrete graphics card for upgradeable mainstream gaming sounds amazing.  For years we all thought it was fine too – frame rates as reported by in-game benchmarks or tools like FRAPS showed increases in performance.  The problem was that it was all just a facade, something that tools like FCAT and our Frame Rating graphics performance methodology have proven.  Higher frame rates don't really mean much if half of those frames are thrown away, not visible, and not affecting user experience.

The ability to frame pace multi-GPU graphics has finally arrived to Dual Graphics with the Catalyst 13.35 beta and the results are pretty good.  Of the five games we tested, four of them saw significant improvements in playability and experience with the new driver that did not exist before.  The only game that didn't see improvement was Skyrim, a game that is still using a DX9 rendering engine.  AMD has clearly told us that only DX10/11 games are enabled for Frame Pacing right now, and in all honestly I don't think we'll see DX9 implemented at all.  The game development world has moved on, and unfortunately I don't think AMD sees the value of spending time on that right now.  

Scaling with Dual Graphics, the jump in average performance going from the Radeon R7 250 discrete single GPU to the coupling of the R7 250 and A8-7600 APU, ranges from 15% to 35%.  GRID 2 sees the least performance scaling likely due to the already CPU-centric performance limitations the game often displays.  Modest amounts of frame rate spiking also occurs in the this game which somewhat limits the improvements in real-world performance.

Battlefield 3 shows the most impressive change with very little variance and a solid 35% increase in average frame rate.  Yes, I realize that isn't the 60-90% we are used to seeing in the discrete graphics world, but considering the architectural differences between the APU and the discrete graphics card, this is a solid result.  Battlefield 4 actually looks worse for this configuration.  Even though it scales up by 25% there is quite a bit of frame variance and spikes in the frame times that we would like to see improved.

The fix isn't perfect, but the driver came as such a surprise to me that it's hard to not be impressed by the results.  As I said at the start of this story, it has been a long and painful trek for AMD over the past year, having to rebuild its Catalyst software as well as its credibility with the enthusiast community.  They are getting there, but it's going to take a lot of consistent software support to maintain it.

« PreviousNext »