In its first attempt at the MLPerf Benchmark performance in the Closed Edge Power category, the US-based machine learning start-up SiMa.ai, which focuses on solutions for the embedded edge, achieved the top score.
The start-up was founded by Krishna Rangasayee (Founder and CEO), and its staff specialises in software, semiconductor design, and machine learning. Its concentration is on the embedded edge market. Initially, it concentrated on finding solutions to problems in the fields of intelligent vision, robotics, Industry 4.0, drones, and autonomous vehicles, among others.
The company’s Machine Learning System-on-Chip (MLSoC) platform achieved the highest inference results on the ResNet-50 benchmark, outperforming the market leader in terms of power consumption and performance (frames per second). These MLPerf results show that the MLSoC Platform lives up to its promise of providing a simple, push-button, Any, 10x solution for ML deployment.
The ResNet-50 benchmark, an image classification inference benchmark (test), is frequently used as a benchmark for evaluating the efficiency of machine learning accelerators.
“While it is exciting that we beat the incumbent leader at MLPerf in performance and power, what is super rewarding is that we are delighting customers globally with a ‘Any. 10x. Pushbutton’ experience that in real life applications far exceeds any other alternative,” said Rangasayee. “Our focus remains doing ML software right for our customers.”
Established by industry leaders in 2018, the foundation for MLCommons aims to accelerate machine learning innovation. The MLPerf inference benchmarks are released bi-annually and define a fully standardised way of measuring performance and power for a variety of ML applications, enabling end-users to easily sort out company claims, and provide industry-standard metrics.
SiMa.ai’s MLSoC hardware combined with its Palette software delivers a purpose-built platform with push button results, enabling effortless ML deployment and scaling at the embedded edge while achieving a claimed 10x better performance at the lowest power.
With this methodology, SiMa.ai says it is able to achieve dramatic results without needing to employ a massive team while delivering results in minutes versus competitor technology that requires months.
For the autonomous driving technology, the start-up claims it can support the automotive industry for ADAS and electrification roadmap with its solutions that can take care of all the computation needs for less than 10W for L2+/L3 applications and less than 100W for complex L4/L5 automotive systems.
