There are two branches: High performance Series7XT for smart phones and tablets, which scales from 100Gflop to 1.5Tflop, and Series7XE which is optimised for chip area and aimed at TVs, set-top boxes, wearables, and lesser mobile devices.
The basic unit of Series7 GPUs is the ‘unified shader cluster’ (USC) which consists of 16 processing pipelines. Each pipeline has either two 32bit floating point ALUs, or four 16bit ALUs – so there are 32 32bit ALUs in one cluster. Imagination’s definition of an ‘array’ is 0.5-16 clusters, or 8-512 32bit ALUs. Two USCs share one texture unit.
7XT GPUs have two or more clusters, while XT GPUs have one and below.
So far defined are: GE7400 (0.5 cluster = 16 cores), GE7800 (1 cluster) GT7200 (2), GT7400 (4), GT7600 (6), GT7800 (8), GT7900 (16 clusters).
For scientific applications, there are optional 64bit floating point ALUs, one per USC.
Compared with the firm’s Series6 (which maxed-out at eight clusters), architectural improvements have increased cluster throughput by up to 60%. These improvements include: co-issue capability added to the instruction set, a new hierarchical layout structure that enables scalable polygon throughput and pixel fill rate improvements in addition to increased clock frequencies, and GPU compute setup and cache throughput improvements.
For security against inter-app hacking, hardware virtualisation now completely separates applications from one-another.
Hardware tessellation and OpenGL ES 3.1 support are included for the Android Extension Pack (AEP). “The new pack ensures maximum content compatibility with the Android 5.0 ‘Lollipop’ release,” said Imagination.
For Microsoft operating systems there is a DirectX 11.2 feature set in Series7XT cores, and the 64bit flating point scientific extension supports OpenCL.
Electronics Weekly Electronics Design & Components Tech News