AMD constantly stresses that their Llano and now Trinity processors are in fact APU. It means that their architecture has been optimized for different types of tasks, which can be executed by joint effort from the traditional x86 cores and the streaming processors of the integrated graphics core. Of course, special software is necessary for these principally different computing resources to operate in harmony. And if a year ago it was a death sentence for the APU concept, now things have started to change dramatically. The developers of multiple popular applications started to take advantage of the hybrid solutions. As of today, the current or upcoming version of several applications will be able to benefit from the computing potential of the integrated graphics cores. These applications are: Adobe Flash 11.2, Adobe Photoshop CS6, GIMP, ArcSoft MediaConverter 7.5, CyberLink MediaEspresso 6.5, Handbrake ? WinZip 16.5.
Today we cannot test the new Trinity processor in any of these applications, but we can definitely try and estimate the practical performance of the Devastator core under GPGPU load created via OpenCL and Microsoft DirectCompute interfaces. This is where we decided to use SiSoftware Sandra 2012.10.18.74 suite.
The computing performance of the Devastator graphics core looks really good. VLIW4 architecture provides superb efficiency during general-purpose calculations, allowing Radeon HD 7660D to get way ahead of the previous version of the graphics core in Llano processors and Intel HD Graphics 4000 core, but also of the discrete Radeon HD 6570 graphics card. As a result, we can expect Trinity to do really well in applications supporting OpenCL.
The situation is quite similar in the encryption test. Namely, by using VLIW4 architecture in their new hybrid processors AMD was pursuing a specific goal: to demonstrate the advantages of combining x86 computing cores with streaming graphics core. Since software developers continue checking out hybrid processors potential, it is a very timely move. At this point AMD should not only demonstrate that this approach is possible, but also prove it efficient and advantageous.