AMD Shares Fresh Mantle Numbers, Promises Driver to Support it Soon
Posted on January 30, 2014 12:00 AM by Rob Williams
We’ve been hearing about AMD’s Mantle API for a number of months, and while the wait for it to arrive has been a bit rough, the first driver to support is due soon. In the interim, AMD is sharing some performance data to help whet our appetites, and for those who run modest PCs and have compatible GPUs, this is one piece of tech that’s worth taking note of.
As a quick recap, Mantle is AMD’s low-level API solution that aims to negate some of the performance overhead of higher-level APIs (DirectX and OpenGL). AMD has touted it as “bare metal” programming, where developers will have unparalleled response times from the GPU to render graphics faster. Overall, the goals include reducing the overhead of processing commands, adding control of the command buffer, reducing the shader compilation overhead, and perhaps best of all, properly scale to multiple CPU cores.
AMD stresses that Mantle works best when a game is CPU-bound, not GPU-bound – if a game is GPU-bound, it tends to stress the card to such a degree that Mantle-like optimizations will not help much at all. That said, those who will see the best benefit of Mantle will have modest PCs – something we’ll see evidenced in the results below.
First up, we have Battlefield 4, the primary game that AMD has mentioned in relation to Mantle since the tech’s unveiling.
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)
As seen, when a modest CPU is used along with a huge CPU, a 41% gain at 1080p can be seen. This contrasts with the 2.7% gain seen when using the highest-end CPU on the market along with a midrange GPU.
Oxide Games’ StarSwarm is AMD’s other proof-of-concept. Oxide consists of industry veterans from Firaxis and Microsoft Studios, so they seem like good candidates for executing Mantle at its full potential.
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)
Well, that’s quite a different picture. A 319% increase might seem a bit outlandish, but the reason for it has to do with some of the tasks Mantle takes off of the CPU. In this particular case, Mantle allows more units to be introduced into the game, an issue that a high-end CPU negates thanks to its extra performance potential. Even so, with Mantle enabled, a small gain can still be seen on high-end rigs. It’s worth noting that a demo of StarSwarm is due out Thursday afternoon on Steam.
AMD’s Radeon R7 260X is a perfect candidate for improvements from Mantle
Alongside these results, AMD had intentions to include Mantle support in its just-released Catalyst 14.1 beta driver (edit: it appears that AMD might be a day late on this one), but things didn’t quite go as planned. However, AMD did get a couple of neat features added that are worth talking about.
First and foremost, 14.1 is the first driver to enable HSA support on Kaveri APUs. As we mentioned at Kaveri’s launch earlier this month, HSA will enable the processor’s CPU and GPU cores to work together in certain scenarios where the developer compiled HSA support in. What that means is, an app might use the GPU for certain things, and use the CPU for the rest. At the moment, LibreOffice 188.8.131.52+ and Corel AfterShot Pro 184.108.40.206 both support HSA.
The other addition is “phase 2” frame pacing support, which affects games run at above 1600p on cards that don’t support XDMA (HD 7000, R9 270X, and R9 260X), along with dual-GPU configurations (even including Kaveri paired with a Radeon R7 250).
Overall, those are some nice additions to Catalyst 14.1 (and there may be more), but I’m sure I’m not alone in wanting Mantle support to get here sooner than later. That said, even when the initial support does get here, AMD stresses that it’ll be a true beta, and that performance will no doubt be enhanced over time, as well as launch bugs (such as the inability to take advantage of Mantle in a multi-GPU setup) prevalent. Even so, bring it on.