PERFORMANCE OF THE ELBRUS-8SV MICROPROCESSOR FOR TECHNICAL VISION TASKS UNDER POWER CONSTRAINTS

Abstract

Research in the field of creating specialized computing systems for robots is conducted in many world scientific centers, including our country. The development of capabilities of sensor systems, global navigation systems, growth of computing power and improvement of algorithms allow creating onboard computing systems with broad intellectual capabilities. An important, butunsolved problem remains the equipping of such computing systems with domestically produced microprocessors. The emergence of domestic computing systems and software of the new generation, such as microprocessor "Elbrus-8SV" and OS “Elbrus” opens up new opportunities for developers of robotic complexes. The peak performance of "Elbrus-8SV" microprocessor is more than 0.25 TFLOPS of double precision, which allows solving computationally complex tasks, such as technical vision tasks, on the microprocessor. Another important requirement of onboard computing technology, in addition to computational power, is low power consumption. As a rule, on general-purpose microprocessors, high performance is impossible with low power consumption, and specialized processors, such as vector or neuroprocessors, are used to solve computationally complex technical vision tasks. To reduce the power consumption of general-purpose microprocessors, there are special methods, among which the authors considered: switching off the physical cores, reducing the clock frequency, switching off the pipeline, switching off synchronous pulses in the idle state. The authors reviewed typical technical vision tasks solved by onboard computing systems. An experiment was conducted to estimate power consumption and execution time of vision algorithms when the clock frequency is reduced and the microprocessor cores are switched off. The experiments showed the possibility to decrease the power consumption of the Elbrus-8SV microprocessor cores by 36-46% with an increase in the program execution time. Based on the results of the experiment, conclusions were made about the applicability of the Elbrus-8SV microprocessor for creating advanced onboard computing systems with the ability to operate both in high performance mode and with reduced power consumption. The results obtained by the authors indicate the prospects of import substitution in the field of robotics.