Baikal Electronics showcased a working prototype of its GPGPU core at the Microelectronics 2026 exhibition. The system demonstrated real-time human recognition capabilities. Currently, the core operates on an FPGA, which allows the company to test architecture and software stacks before moving to silicon production.
The prototype includes a second-level cache, a RISC-V control processor, and a proprietary software stack ranging from compilers to runtime environments. It supports existing computer vision models. Developers claim compatibility with Nvidia CUDA, which aims to facilitate the migration of neural network applications to the new hardware.
The upcoming Baikal-AI-E1 processor is positioned as an analog to the Nvidia Jetson Orin platform. Future applications for the chip include robotics, autonomous systems, industrial computer vision, and smart cameras. Local processing is intended to reduce latency and enable operation without an internet connection.
The next development phase involves transitioning the architecture from FPGA to a physical silicon chip. Only after this transition will the company be able to evaluate final performance metrics, power consumption, core area, and mass production feasibility.
