€5.5m to optimise embedded AI power consumption
Swedish startup Embedl has raised €5.5 million to optimise generative AI for power and latency in embedded applications.
The company is based on research by Professor Devdatt Dubashi of Chalmers University of Technology who has developed Model Optimization technology to help AI systems run efficiently on existing hardware, avoiding costly upgrades.
The funding will be used to expand the Embedl Hub platform to help OEMs deploy generative AI models across multiple hardware platforms. Benchmarking can compare models for specific applications and compare hardware. The hub currently covers the Nvidia Orin processor, Qualcomm Snapdragon 8 and the Raspberry Pi 3 and 4.
The technology can reduce the energy consumption up to 83%, and manufacturers can halve the cost of their hardware by optimising their models. It is already used by Bosch, Zenseact, Tobii and Kodiak Robotics.
A software development kit (SDK) provides tools to prune, quantise, and compress deep learning models, reducing size and speeding up inference for applications such as voice recognition, text generation and image classification. A modular design lets developers tailor components for specific needs and apply their domain knowledge and built-in visualisation tools make it easy to track model changes during optimisation.
“The world needs to make AI more energy efficient, fast. While the applications and usage of AI continue to skyrocket, we can’t increase energy consumption at the same level. Our solution will also help bring robotics and autonomous vehicles to the market faster, as we can help optimise the hardware’s energy efficiency while assuring the highest quality data being transferred instantly,” says Hans Salomonsson, CEO and co-founder of Embedl (above).
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
