The Furian architecture has been designed to cater for higher volumes of graphics rendering and GPU-compute that are likely to come on future consumer equipment, Imagination said, but it provided scant technical information saying more would be provided on its website next week.
When compared with a PowerVR Series7XT Plus GPU in the same process technology and at the same clock frequency and for cores with a similar silicon area Furian improves GFLOPs and fill-rate, Imagination claims. The GFLOPS density improvement is 35 percent and the fill-rate density improvement is 80 percent.
However, Imagination is already offering PowerVR Series8 cores so it would be interesting to see if Furian cores perform better than PowerVR Series8 (see PowerVR GPU IP targets mid-range markets).
Furian has also been designed with virtual and augmented applications in mind and emerging applications such as autonomous driving assistance systems (ADAS) and machine learning.
Power VR Furian will join Power VR Rogue and is intended to provide enhanced power efficiency and scalability so that immersive graphics can be supplied for sustained time periods on battery operated mobile devices.
Furian features a 32-unit wide ALU cluster and a novel instruction set architecture (ISA) in the primary and secondary ALU pipelines. It also allows function-specific pipelines such as those for ray-tracing, to be integrated with the GPU.
To address GPU-compute requirements it provides APIs including OpenCL 2.0, Vulkan 1.0 and OpenVX 1.1 as well as a bi-directional GPU/CPU coherent interface for data sharing.
Next: Still using TBDR