Xilinx, known for its decisions in the field of programmable logic and adaptive computing systems, introduced a new series of Versal AI EDGE devices, oriented to peripheral and signals and processing signals for machine vision, radars, lidarov, SDR, etc.
This series will replenish the ACAP (Adaptive Compute Acceleration Platform) – chips that have become a company response to the market needs in decisions combining FPGA flexibility with ready-made processor cores, memory, interface controllers and other hardware blocks. By connecting them in one crystal, Xilinx received a platform to accelerate computations with flexible adaptation capabilities under a wide range of possible applications called Versal.
In AI EDGE, the company uses the VERSAL 7-nm architecture, which optimized for II calculations and reduce power consumption, up to 6 W in the minimum configuration. The engine will receive an increased up to 38 MB of memory. An additional 4 MB of RAM can be built in addition. Maximum performance reaches 479 Tops on int4-operations. External interfaces are represented by blocks DDR4-3200 / LPDDR4-4266, 32 Gbit / s transceivers, 40GBE, PCIE 4.0 with CCIX, as well as MIPI with the sensor subsection by resolution from 8 MPIX. The latter are critical for the autopilot (adas) systems of the second level and above.
All this, according to Xilinx, makes it a platform leader in the field of peripheral computing – thanks to the highest II-performance in the world in terms of WT, it leaves far behind the decision based on specialized ASSP chips (Application Specific Standard Product) and GPU, including And Nvidia Jetson Xavier.
However, the delivery of the first samples of new products is scheduled for the first half of 2022, and the sets for testing and developing are on the second half. Therefore, the comparison with the decisions of the last two years looks strange – by the time the Versal AI Edge is released, competitors will also update product portfolio.
Be that as it may, the announced Xilinx platform will be a fairly interesting solution. Thanks to the presence of FPGA products based on it will be able to obtain a longer life cycle, since over time they can be updated by adding new features or optimizing available.
Versal AI EDGE is expected to use in autonomous driving systems with increased security functions, prediction systems, robotics, medicine, as well as defense and space products. Easy to developers, the development of a new platform will allow branded design tools Vivado, Vitis and Vitis AI, as well as specialized libraries, frameworks and OS for specific applications.