![Story image](https://itbrief.co.nz/uploads/story/2024/05/01/img-Pu8BKjl0AkAEZqbXgNhFaafX.webp)
IoT Analytics unveils six key trends in edge AI at Embedded World 2024
IoT Analytics has released research highlighting six key trends in edge Artificial Intelligence (AI) following the Embedded World 2024 Event. Key insights with a focus on edge AI were shared, having been derived from a 67-page report by a team of on-the-ground analysts.
The Embedded World 2024 Event Report compiled by the IoT Analytics team presented significant trends in IoT chipsets and edge computing. The researchers described edge computing as an array of intelligent computational resources situated near the data consumption or generation source. In turn, edge AI involves deploying AI models on a device or equipment at the 'edge', thus making AI inference and decision-making independent of constant cloud connectivity.
Satyajit Sinha, Principal Analyst at IoT Analytics, commented on the ever-growing necessity for CPU vendors to develop high-performance multi-core CPUs and integrate specialized NPUs into their System-on-Chip designs to help cope with the edge AI shift. Such a change was highlighted during Embedded World 2024 and was particularly noticeable in some key trends observed throughout the event.
The major trends noted in edge AI included NVIDIA setting the pace in AI technology adoption with their renowned GPUs, and companies such as Aetina and MediaTek showcasing their partnership with NVIDIA. Additionally, a rise in platforms that accommodate AI model deployments on-chip, like the EdgeAI SDK platform by Advantech, was noted, allowing developers to test AI model deployment without purchasing physical hardware.
NPU integration into edge devices, boosting AI inference abilities and efficiency, was also featured. Solutions such as NXP's new MCX N Series Micro Controller Units (MCUs), which deliver 42 times faster Machine Learning (ML) inference than CPU cores by themselves, were displayed.
A shift towards AI model training on thick-edge locations, AI-enabled cellular IoT devices allowing for localised decision-making, and the growing presence of small-sized AI/ML in everyday objects and tools enabling autonomous decision-making without reliance on the cloud rounded off the key trends noted.
Examples of the progression towards edge AI included: Quad-NVIDIA-GPU-integrated MAINGEAR PRO AI workstations by MAINGEAR and Phison, Aetina's AIP-FR68 Edge AI Training platform accommodating four NVIDIA GPUs, Aetina's AIB-MX13/23 edge AI solution for industrial fault detection using NVIDIA Jetson AGX Orin GPU, numerous demonstrations of NXP and ARM's edge AI chipsets, and Fibocom's Qualcomm-powered intelligent mowing robot, among others.
Analysts predict that the rise of edge AI could usher in considerable industry changes, particularly in healthcare, automotive, and robotics. The move away from cloud-centred models towards the edge will further reduce reliance on hyperscalers and promote broader AI usage outside centralised infrastructure.