For those of you who read my overview on AMD’s Trinity that launched two days ago, you might be interested in how OpenCL benefits you even if you’re not on the new platform with the “Devastator” GPU. Gaming aside, AMD’s biggest flaw is single-thread performance and an overall lack of computing muscle compared to Intel’s last-generation part, Sandy Bridge.

OpenCL looks to give customers heterogeneous computing abilities no matter the platform, allowing applications to take advantage of GPU acceleration together with your CPU to make short work of things like video encoding, photo editing and animation, to name a few workloads and AMD’s Bulldozer cores currently suck at. With Trinity, AMD has pushed forward their agenda to open the OpenCL standard to all applications that could use it and Anandtech showed the world how it could help Intel users, too.

OpenCL doesn’t care which platform its on. Its an alternative to the proprietary QuickSync technology used by Intel which is only supported in paid-for applications like Cyberlink’s MediaEspresso converter. Intel’s Quicksync is really, really good at accelerating the transcoding process while preserving framerate, image quality and picture-perfect colour. by comparison many didn’t favour Nvidia’s approach using CUDA cores, as the final video was more leaning towards performance and a faster, smoother frame rate than image quality. AMD’s Stream achieves the kind of quality seen with Quicksync, but doesn’t have nearly enough chops to do it at the same frantic pace.

 In the test Anandtech ran with MediaEspresso, OpenCL shows that Trinity’s transcoding time nearly halved and now draws up squarely to Intel’s best chips with Quicksync disabled. Not shown here is AMD’s Llano under the same OpenCL benchmark, but I assure you it reaches a similar level of performance, nearly halving the time taken to complete the video. Trinity will never rival the 12-second record set by QuickSync but it doesn’t have to – its almost $300 cheaper than Intel’s flagship.

For those of you who really weren’t paying attention in class and staring at your classmate with the pretty eyes/nice smile/big boobs, OpenCL is actually royalty-free software, free to implement wherever you choose. Who designed it? Apple Inc. No, I’m not joking. All those rumours that Apple was looking to put AMD chips in its laptops in future? Looks a lot more likely now than it ever was, doesn’t it?

Moving onto other applications, a really popular option for those who don’t want to pay the ridiculous fees for MediaEspresso can grab Handbrake, the open-source GPL-distributed video transcoding software. Handbrake is multi-threaded and has been a popular option for many media enthusiasts, but suffers from a lack of GPU acceleration. Nvidia’s CUDA, Intel’s QuickSync and AMD’s ATi Stream are all licensed technology and a free program like Handbrake wouldn’t be able to recoup the kinds of losses it would otherwise incur. It’d just be another media converter that costs roughly $25 to do the same thing as everything else. Handbrake’s dev team is working on an OpenCL-capable build and Anandtech had an early version in their hands for testing. Both Intel and AMD receive a boost from the standard, but wait ’till you see the scores!

Two things should be noted here. First, Intel’s Sandy Bridge platform doesn’t support OpenCL on their HD3000 GPU. Thats surprising considering that it still does better, finishing second behind Ivy Bridge. Both get around a 5fps boost, suggesting that there was a slight amount of power left. It should also be noted that this is with QuickSync disabled, since there’s no discernable benefit. Trinity’s scores double, bringing it right up to the Intel Core i7-3720QM. For a low-end solution, that’s pretty impressive. Even more surprising is Llano’s improvement, more than doubling its original score.

Its a win for AMD either way, and consumers who focus on content creation but need something on a low budget, they’d be happy enough with a capable Llano or Trinity notebook/Ultrabook. Intel users have nothing to fear for now, but this may change in future.

Source: Anandtech

Discuss this in the forums: Linky

More stuff like this: