Ambarella’s CV7 SoC leverages the CVflow computer vision architecture to bring 8K image processing and advanced AI inference to the edge. It supports simultaneous processing of multiple video streams at up to 8K at 60 Hz, making it well suited for a wide range of consumer and industrial AI perception applications, as well as multi-stream automotive systems—particularly those running CNNs and transformer-based networks.

Built on a 4-nm process, the CV7 delivers low power consumption, reducing thermal management requirements and extending battery life. Compared to its predecessor, it consumes 20% less power while integrating a quad-core Arm Cortex-A73 CPU that doubles general-purpose processing performance. A 64-bit DRAM interface further improves memory bandwidth.

The highly integrated CV7 SoC includes a third-generation CVflow AI accelerator, delivering more than 2.5× the AI performance of the previous CV5 SoC. It also integrates an image signal processor and hardware-accelerated video encoding for H.264, H.265, and MJPEG formats.

CV7 SoC samples are now available, along with a CNN toolkit for porting neural networks developed using Caffe, TensorFlow, PyTorch, and ONNX frameworks.

CV7 product page 

Ambarella

The post Vision SoC powers 8K multi-stream AI appeared first on EDN.