Xilinx introduced Versal, a processor designed for artificial intelligence reasoning, using TSMC’s 7-nanometer FinFET process technology. Xilinx claims that its Versal AI Core series has eight times the artificial intelligence reasoning performance of Nvidia’s GPU Tesla V100.
The processor uses the Adaptive Compute Acceleration Platform (ACAP), which combines a scalar processing engine, an adaptable hardware engine, and a smart engine with advanced memory and interface technology blessing. The chip provides powerful heterogeneous acceleration for any application. Xilinx claims that the chip has far more than the artificial intelligence processing capabilities of the FPGA chip.
The Versal chip combines Xilinx’s own FPGA technology, two high-performance Arm processor cores, two low-power Arm processors, and a dedicated artificial intelligence computing engine. Xilinx claims that its artificial intelligence reasoning performance is eight times that of the Tesla V100 and 43 to 72 times that of the Intel Xeon CPU.
Xilinx currently only works with several early partners. The Versal Prime series and Versal AI Core series will be officially launched in the second half of 2019. In addition, the software development environment for its Versal products, including drivers, mediation software, libraries, and software frameworks, will also be available in 2019.