ARM emphasizes that with the Mali-T658 SoC designers are able to make use of a carefully crafted system-level approach to multicore design. That approach that includes ARM Cortex processor cores, the little-big power efficiency technology and cache-coherent interconnect.
As a result designers are expected to target high-end smartphones on 28-nm silicon with quad-core Mali-T658 coming to market in 2013 and eight-cored Mali-T658 graphics units to be 20-nm silicon in 2015. The core is also expected to find application in tablet computers, smart-TVs and automotive infotainment systems.
Mali-T658 will be able to take on computation tasks in applications such as image processing or augmented reality. The core has been made compatible with the recently announced A7-A15 little-big coupling so that as computation is moved on to the T658 it may accompany the movement of the core program down from the A15 to the A7, said Jem Davies, vice president of technology for the media processing division at ARM. The autonomous nature of the Mali Job Manager, and its ability to carry on graphics processing with a reduced load on the CPU, means it is suited to working alongside a big-little CPU system. By using the right processor for the right task the Mali-T658 is able to handle GPU compute tasks in parallel with the CPU handling the always-on always-connected tasks. ARM CoreLink system IP enables system-level cache coherency across clusters of multicore processors, including the Cortex-A15 and Mali-T658.
In addition the Mali-T658 is compatible with the ARMv8 full 64-bit instruction-set architecture, as is the Mali-T604.
ARM lead partners on the development of the Mali-T658 are listed as Fujitsu Semiconductor, LG Electronics, Nufront and Samsung.
As would be expected the Mali-T658 GPU supports all popular graphics and compute APIs, including Microsoft DirectX 11, Khronos OpenGL ES, Open VG, Khronos OpenCL, Google Renderscript and Microsoft DirectCompute.
Related links and articles:
ARM mulls 'big-little' scheme for graphics
ARM adds cache coherency to AMBA 4
Imagination raises outlook for graphic cores
Analysis: 'Rogue' design win hurts ARM
ARM Plays Catch≠up in Graphics IP