The Vivante CC8000 IP family delivers 576GFLOPS to 2TFLOPS compute performance for embedded devices.
VeriSilicon Holdings Co. Ltd has released the Vivante CC8000, an IP series based on parallel, scalable memory efficient shader (MESH) processor architecture targeting embedded applications such as natural language processing, advanced signal processing, machine vision, scientific computing and cryptography. The CC8000 claims to deliver compute performance using less energy and much smaller die area.
Included in the Vivante CC8000 family is the CC8400 compute engine that is cantered around a proprietary MESH architecture. A single CC8400 is comprised of 256 MESH cores; each core has a 32-bit single-precision compute unit that delivers 576GFLOPS of compute performance at clock rate of 1GHz. The CC8400-Quad has four CC8400 compute engines operating as a single IP unit delivering more than 2TFLOPS of compute performance. The CC8400 compute engine offers intrinsic support for complex math instructions and double precision operation.
The CC8000 series is OpenCL 1.2 FP and OpenVX compliant and implements all aspects of OpenCL including transcendental functions natively in hardware. It also provides OpenCV for Vision applications. The Vivante CC8000 software solution is operating system agnostic and can be ported to virtually any required software platform.
The CC8000 series is currently in production and available as an IP license.