Over the weekend, AMD let loose on on the Catalyst 14.1 1.6 Beta drivers which contain fixes for a number of performance issues with recent AMD Radeon GPUs and the APIs for Mantle and TrueAudio. Everything AMD has worked towards for the past two years has led up to this point and this specific driver relase. How significant is it for games that don’t pack in Mantle? Is frame pacing finally fixed? Hit the jump to see AMD kept their promise.
AMD’s latest driver, for now, is in a very, very early beta stage. In fact, it’s so early that there are several bugs that still have to be addressed, but AMD felt that it delivered enough of a change that it was ready for testing by consumers. The driver includes a lot of changes to how Radeon GPUs handle frame pacing and AMD promises much better performance and smoother gameplay in Crossfire with Eyefinity setups, 120/144Hz gaming and with triple-Crossfire. Its a big boost from what Radeon owners have been used to for all these years and it deserves a closer look.
But let’s get the big kahuna out of the way – Mantle.
Mantle works and is a screamer in some cases
That line pretty much sums up any of the results you’ll be seeing of Mantle tests from here on. The people who benefit most from Mantle fall into a handful of categories – gamers who are CPU-limited (or using an AMD processor), gamers who seek more efficiency from Crossfire setups, gamers who play at resolutions higher than 1080p with slightly weaker systems and multiplayer games that are bottlenecked by CPU performance. Much of the tests you’ll see on the net will concentrate on these scenarios and for this reason the performance improvements that Mantle brings may vary wildly.
PC Perspective’s testing of Battlefield 4‘s single-player revealed the kind of scenario where Mantle won’t help people much. If you’re running a high-end CPU and a single high-end GPU together, Mantle is only slightly more efficient than DirectX rendering. More consistent CPU usage was observed, but because Battlefield games typically have GPU-limited single-player campaigns, these results won’t mean much.
Now take a system that’s similar, but shove in a AMD A10-7850K, a APU that is considered a lot weaker than the Core i7-3960X. In the above results, the numbers in the end are close when you look at frame times and frame rate averages. But there is a major different in terms of CPU usage – it’s a lot more constrained and Battlefield 4 takes up less CPU time across the benchmark than DirectX. This is one of the first clear indicators of the benefits of Mantle – the CPU is less bogged down with code relating to the GPU driver.
Mantle’s main aim, in case my discussion on it hasn’t pointed this out clearly, is to help remove CPU bottlenecks in a system and allow graphics cards to stretch out their legs a lot more. Clearly, this is working well already, with a slight boost to minimum framerates as well. It scales up from there, too.
In a multiplayer test with the same rig, the benefits are finally seen in full force. Battlefield 4 in multiplayer is a known torturer of processors and here the load is lightened with Mantle’s help. Minimum framerates are exactly the same (its still a game that’s riddled with bugs as I write this) but average FPS jumps up by 27%, close to AMD’s official claims. The CPU time taken up by the game and the graphics driver is much more constrained and the GPU as a result can perform in a much more consistent manner at a much higher framerate.
Those benefits scale up as well. Moving up to 2560 x 1600 at Ultra settings, the benefits are even more pronounced. Minimum and average framerates see a huge jump, while the frame times see a massive drop and become a lot more consistent, resulting in a superior and smoother experience. This is the kind of performance jump commonly seen in hardware generational jumps and only very rarely in software.
In terms of overall efficiency, CPU and GPU time sees a big boost and almost matches up perfectly to the frame time results. This represents a jump in performance of 32%. Due to limited time spent testing the new beta driver, PC Perspective wasn’t able to compare the same multiplayer test with the Intel system, but it would be interesting to see how they match up.
All of this mostly tallies up with AMD’s own expectations to how CPU performance increases with the Mantle rendering path. As expected, because Intel processors are already very good at running code that is largely single-threaded in nature, there’s not a lot of change at the high-end of the scale. Even AMD’s FX-8350 won’t see very many performance benefits, although it’s still free performance for both companies no matter which way you slice it.
The benefits are more pronounced the lower you go, with almost 20% performance gain for the aging Phenom II 1090T six-core Thuban processor (and almost certainly the same increases for anyone still running a Core i7-920). It appears that this also benefits processors without L3 cache more than processors with L3, with the A8-6600K and A10-7700K seeing around 40% jumps in overall performance. The catch is that this is with a Radeon R9 290X, a pairing that will be unfathomable for a lot of people.
Still, Mantle is here and it works, even on a game that has been buggy from day one. There are bugs to iron out, a lot of work still needs to be done to get this into the general public’s hands and more awareness needs to be built up around the use of Mantle. But, judging by the fact that this will be in all Frostbite 3.0 games from now on, expect these performance jumps to be a common thing for games coming from EA’s stable.
Discuss this in the forums: Linky