Drivers are very interesting things; they usually work out of sight but are just as important as the GPU itself. It is with these thoughts in mind that I would like to post something very interesting today: a benchmark comparison (courtesy of DG Lee over at IYD.KR) of all the driver releases between and including Catalyst 13.9 WHQL and 14.7 RC. I would also like to point out a very interesting anomaly that I noticed in the benchmarks.
Curious Watch Dogs Anomaly – Catalyst 13.9 WHQL to 14.7 RC Shows 4% Improvement Overall
Let me begin by saying that the benchmarks were conducted by DG Lee and all measures were taken to ensure that the system remained constant and only the drivers were changed. Catalyst versions were completely and thoroughly removed before the next version was benchmarked, since drivers are notorious for sticking around after they have out stayed their welcome. Lets begin with the testing rig used:
- Intel Core i7 4770K
- AVEXIR DDR3 PC3-12800 CL9 16GB (4×4)
- AMD R9 280X 3GB
- ASUS Maximus VII Hero
- Team L3 SSD 240GB
- Corsair AX 750 GOLD
- Windows 7 Ultimate
That is a pretty high end build with all bottlenecks accounted for completely. The drivers that were tested are: Catalyst 14.7 Release Candidate, Catalyst 14.6 Release Candidate, Catalyst 14.4 WHQL, Catalyst 14.3 WHQL, Catalyst 14.2 Beta, Catalyst 14.1 Beta, Catalyst 13.11 WHQL, Catalyst 13.10 WHQL and Catalyst 13.9 WHQL. A very wide array of games was tested, including Watch Dogs, Crysis 3, Battlefield 4, Metro 2033 and many more. Without any further ado here are some of the benchmarks:
As you can see there is a fixed pattern to the benchmarks. Nearly all of the data lies within very small parameters and variation is very slight, if at all. And while in some cases you see a performance increase going up the Catalyst versions, in others you actually see downgraded performance. However, this is to be expected and quite standard. Now this is where the Watch Dogs anomaly comes in. In the driver benchmark of Catalyst 14.7 RC, the frames per second take a massive leap. See for yourself:The frames per second jumps from 46 to 58. That is an absolutely insane amount of performance increase just from a driver update. I looked up at AMD Catalyst 14.7 RC notes and there was no mention of a performance increase this significant so the next thought that came to me was human error. Contacting IYD netted me the following two possibilities: A weird one of a kind software glitch (not human error) or the bench being completely legit. You would recall that there was some pretty hot talk between AMD and Nvidia over Game Works and Watch Dogs. AMD stated that the reason for the poor performance was that the Game Works library was closed source, however Nvidia stated that they had nothing to do with it.
If the benchmark is legit then this raises a very interesting question. What exactly happened? I am pretty sure Nvidia didn’t hand over the source files to Game Works so how did Red manage this significant an increase? Well, not to take sides, but I think there is a slight chance that the issue was with Catalyst drivers from the start and AMD realized that and patched it up. Since no other game had anywhere near this kind of performance increase, we are not looking at miracle optimization either. This is certainly good news for AMD users who were getting the short end of the stick with Watch Dogs.
— Usman Pirzada (@usmanpirzada) August 22, 2014