Think Silicon NEOX GPU family with models optimized for graphics (NEOX|G) or artificial intelligence (NEOX|A) is based on the RISC-V RV64C ISA instruction set with adaptive NoC, and offers up to 64 cores delivering up to 409.6 GFLOPS at 800MHz with support for FP16, FP32 and optionally FP64 and SIMD instructions.
The NEOX GPUs can be integrated into microcontrollers, crossover processors, and even more powerful application processors, and target AI, IoT/Edge, and media processing in consumer and industrial devices.
Each shader of the GPU is a programmable 64-bit RISC-V (RV64GC) core running a real-time operating system (RTOS) and the GPU is supported by lightweight graphics and machine learning frameworks. The multi-threaded GPU system can be customized for graphics, machine learning, vision/video processing, and general-purpose compute (GPGPU) workloads.
The solution is meant to be integrated into 32-bit SoCs designed for smartwatches, augmented reality (AR) eyewear, video surveillance, and smart display terminals running Linux, Wear OS, or real-time operating systems. Think Silicon says NEOX IP pre-evaluation systems based on Xilinx SoC FPGA are available for customer testing together with System Verilog RTL, configuration tools, the NEOX SDK, LLVM and GCC toolchains, as well as libraries with OpenCL 1.2 EP and OpenGL ES 2.x APIs support, among other deliverables.
This “news” may not be that new since it looks very much like the NEOX|V RISC-V GPU announcement from December 2019, except there are now two models optimized for either 3D graphics or AI. I’m unable to find any SoC based on the NEOX GPU so far, but progress may have been made and the company will be demonstrating both 3D/2D graphics and machine learning applications at Embedded World 2022. Further details may be found on the product page and recent announcement.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.