Ambarella Revolutionizes AI Camera Processors with Double the Performance for Transformer Networks

Ambarella has unveiled two new 5nm AI camera processors designed for in-vehicle fleet telematics.

The CV75AX processor offers double the AI performance compared to Ambarella’s previous generation SoC. It supports advanced transformer neural networks for enhanced accuracy and fewer false positives without requiring extensive training for every object. This processor also powers complex driver-scoring models, making it ideal for next-generation vehicle gateways.

The CV72AX processor can handle up to 10 cameras and delivers six times the AI performance of its predecessor. It supports both vision transformers and multi-modal vision language models. This allows for immediate event alerts, efficient multi-channel video categorization, and natural-language searches—again, without the need for extensive training.

Both processors come with support for LPDDR5 memory and USB 3.2 interfaces, suitable for 5G modems. They’re equipped with an AI accelerator, a CPU, an image signal processor, and an H.264/5 video encoder, enabling multi-channel recording for evidence collection.

A notable feature of the CV75AX is its low-power mode, consuming less than 5mW, which monitors vehicle sensors when the car is parked.

These processors are manufactured using a finfet process at a Samsung foundry and can operate in extreme temperatures ranging from -40 to 105°C. They are currently available for sampling.

For development, Ambarella provides support through its Cooper platform. This includes Cooper Metal, which offers hardware tools, and Cooper Foundry, which supplies a multi-layer software stack.