One of the highlights of AMD’s Hawaii event last month was the introduction of AMD Mantle, a new, close-to-metal API that would supposedly unlock huge additional performance and address many of the issues that plague DirectX. The implications of this move, given that AMD technology is powering every upcoming console and the current Wii U, was that the new API might also be supported cross-platform, across multiple consoles. AMD itself made no such promises, but a number of people ran with the idea anyway. Mantle, it was rumored, was the Xbox One API ported over for mainstream PCs. This even made some sense — Mantle was pitched as the secret sauce that would allow the Xbox One’s weaker graphics processor to go toe-to-toe with the PS4. Efficiency beats brute strength! Or something like that.
Yesterday, Microsoft released a blog post that kills the notion of Mantle as a direct port from Xbox One. According to the company, Xbox One’s DirectX 11.x support “provides a superset of Direct3D 11.2 functionality. Other APIs such as OpenGL and AMD’s Mantle are not available on Xbox One.” The post goes on to discuss the extensive use of tiled resources in Direct3D 11, improved performance for sharing data across CPU and GPU, reduced presentation overhead, and improved rendering latency.
In short, these are many of the advantages that AMD has also claimed for Mantle. But similarity of function doesn’t imply similarity of code, and Microsoft’s unequivocal statement implies that Mantle and Xbox One’s API are two different beasts. Given the similarities between the Xbox One and PS4 APU, it’s probably fairly safe to assume that the PS4 also doesn’t support Mantle.
Brilliant strategy, or bridge too far?
So far, the only company that’s publicly declared for AMD’s Mantle is Dice, which will deliver a patch for Battlefield 4 that enables Mantle support sometime in mid-December. The game launches on October 29, implying that it’ll take some additional work — though not too much — to bring the API up to snuff. Dice’s presentations on Frostbite 3 make it clear that the company is adding Mantle support to more than just BF4; the Frostbite 3 engine will be able to use the new API in all upcoming titles as well. Thus far, the new Mirror’s Edge prequel, Need for Speed: Rivals, Dragon Age: Inquisition, Plants vs. Zombies: Garden Warfare, the next-generation Mass Effect, and the Star Wars: Battlefront title (scheduled for 2015) are all confirmed as being based on Frostbite 3. That’s a solid beginning for a new API, but it’s just a beginning.
Dice’s own explanation for why they’ve adopted Mantle bears touching on. In its presentation at the AMD Hawaii event, the company showed the following slides:
The situation with Mantle, therefore, seems similar to the situation with HSA. The Xbox One and PS4 don’t claim to be fully HSA-compatible, but still offer most, if not all, of the actual features of that platform. Mantle is designed to give the PC space many of the benefits of console APIs, while preserving the broad compatibility that allows the PC space to be more flexible. It’s also possible that Mantle offers a particular benefit to APUs, where Intel has been steadily closing the gap between itself and Sunnyvale for several product generations. If Mantle can offer a 10-15% real-world performance benefit over DirectX 11 running on an APU, that gives AMD a substantial upwards kick. Combine that with Kaveri’s GCN-based hardware, and AMD’s next-generation APU could finally bridge the “good enough” gap in gaming, once and for all.
But the catch, of course, is adoption. It’s not surprising that Microsoft would avoid Mantle. There’s an intrinsic tension (likely exacerbated by Microsoft’s deep-rooted organizational power struggles) between improving gaming on the Xbox One and improving gaming on Windows. Historically, meanwhile, GPU features that are only supported by one company tend to die slow deaths. AMD incorporated a tessellation engine into every Radeon HD 2000, 3000, and 4000 card, but until DirectX 11 and Fermi, the capability went unused. Nvidia’s hardware-based PhysX enjoyed modest success for a few years, but never became the must-have feature that Nvidia was hoping for when it bought Ageia. The company’s briefings and presentations barely mention hardware PhysX at all these days, and the release of new titles has slowed to a bare trickle. The last successful third-party API was 3dfx’s Glide, which enjoyed massive success for several years — but Glide was released at a time when 3D gaming was in its infancy, OpenGL support was badly fragmented (and slow), and Direct3D barely existed
- 1 of 2