AMD’s Mantle has been the talk of the industry for quite some time now and they’ve admittedly hyped it up quite a bit, but they’ve allowed developers like John Andersson of DICE to really drive the discussion. Although I believe that there was a time when AMD let the discussion get away from them, they appear to finally be ready with the Mantle API after a delay from December to January that we reported on. While it seems as though the delay itself wasn’t necessarily their fault, they clearly take some responsibility. And based on the haste with which AMD presented this announcement to us, it seems like it was a bit rushed. I believe it may have been fed by rumors that AMD was delaying (again) into February by Sam Reynolds at VR-Zone.
They clearly wanted to squash any such rumor and are launched today at midnight to prove their January date, although at the time of writing we still haven’t been provided the Catalyst 14.1 driver that we were promised to test this new Mantle capability. They have since stated that some issues arose with the driver and that they would be providing us with the driver tomorrow for testing since they had accidentally sent out an earlier driver without any Mantle profiles which probably resulted in a lot of very confused reviewers.
If you don’t quite understand what Mantle does, we actually have a few videos that we took at APU13 last year that, one explain Mantle, and two answer any questions you might have about development with Mantle. Both videos are pretty lengthy but do a pretty good job of explaining Mantle.
Getting that out of the way, Mantle is designed to make performance improvements in CPU-limited scenarios and reduces API overhead by reducing draw calls and maximizing the GPU’s overall capability. Based on this, the numbers that AMD provided us with seem fairly in line what one would expect, even though we will need to independently test them ourselves. The driver itself should actually be publicly available at the publication of this article right here (we don’t know when, exactly that’ll be), so you can actually test it out for yourself if you have an AMD GPU with GCN architecture.
Since there are two primary titles that AMD had been working with, namely Battlefield 4 (EA/DICE) and StarSwarm (Oxide Games), there are only performance figures available for the two of them.
o Battlefield 4 (EA-DICE)
- CPU-limited scenario: 40.9% (1080p) and 40.1% (1600p) performance improvement under Ultra settings and 4xAA on the AMD A10-7700K with an AMD Radeon? R9 290X.
- GPU-limited scenario: 2.7% (1080p) and 1.4% (1600p) performance improvement under Ultra settings and FXAA on the Core i7-4960X with an AMD Radeon? R7 260X
- Average uplift for 1080p: 13.28% (Average of 290X and 260X data on the i7-4960X, A10-7700K, FX 8350 and i5-4670K)
- Average uplift for 1600p: 11.35% (Average of 290X and 260X data on the i7-4960X, A10-7700K, FX 8350 and i5-4670K)
o StarSwarm (Oxide Games)
- CPU-limited scenario: 319% (1080p) and 281% (1600p) performance improvement in the ?RTS? test on Extreme settings with the AMD A10-7700K and an AMD Radeon? R9 290X.
- GPU-limited scenario: 5.1% (1080p) and 16.7% (1600p) performance improvement in the ?RTS? test on Extreme settings with the Core i7-4960X and an AMD Radeon? R7 260X
- Average uplift for 1080p: 115.65% (Average of 290X and 260X data on the i7-4960X, A10-7700K, FX 8350 and i5-4670K)
- Average uplift for 1600p: 75.19% (Average of 290X and 260X data on the i7-4960X, A10-7700K, FX 8350 and i5-4670K)
Looking at these preliminary figures, we can see that AMD has clearly achieved their goal of improving low-end performance, especially when CPU-bound. They’ve also really improved performance in games that tend to be very ‘CPU-heavy’ with lots of calculations and units. Good examples of games that will shine under Mantle are Oxide’s StarSwarm, Supreme Commander and even possibly games like Company of Heroes or maybe even Starcraft 2. What Mantle could enable is the ability to play games like Starcraft 2 or Company of Heroes on ‘lesser’ systems with weaker CPUs and GPUs as the CPU tends to be the bottleneck in many of those games.
In addition to the improvements that Mantle promises, the Catalyst 14.1 driver also promises to ring in the second phase of AMD’s frame-pacing initiative. This update adds fram pacing to DX10 and DX11 applications on resolutions higher than 1600p that don’t already have XDMA technology, which AMD introduced with the R9 290X to resolve the issues experienced in mutli-GPU setups.
Catalyst 14.1 is also AMD’s first HSA-enabled driver for Kaveri that enables the GPU and CPU silicon to operate together, already supporting applications like Libre Office, CorelAfterShot Pro and many more. We are currently in the process of testing the Kaveri APU, but it looks like we’re going to have a new driver to test with in order to make sure we provide you with the most up to date information.
I’m not entirely sure why AMD launched this driver announcement or Mantle in such a way, and its frankly perplexing to me the way that they handled it. What will be the most interesting thing for me, though, will be to see how a game performs that is specifically designed with Mantle in mind, rather than an afterthought. DICE clearly had some issues with Battlefield 4 that they needed to fix before they focused on Mantle, which I believe was one of the core reasons for the delays, but I once again feel like AMD has sort of lost control of the story and how Mantle is perceived by the press and their readership. We’ll hopefully get testing in the new Mantle Catalyst driver soon and will let you know how much it really does improve performance. Because to me, if the performance difference is at least double digits, Mantle is a success. If it isn’t, then it is probably a waste of time and resources.
Based on the numbers we’ve been given, it looks like a pretty solid proposition to value gamers that don’t want to spend a crazy amount of money but still want playable games. I also believe that Mantle will give AMD an advantage in terms of their weaker CPU cores and it will make gaming on AMD CPUs less problematic. Its no secret that AMD’s CPUs are weaker in terms of IPC than Intels, but with Mantle it becomes much less relevant. It will also be interesting to see whether or not Mantle will affect the ability
to do compute programming and whether or not OpenCL programmers can get more performance out of AMD’s GPUs using Mantle as well. Because right now, Mantle seems to be an API very heavily focused on gaming and graphics, but I think there may be some benefits to compute as well.