Highlights:

  • The Hailo-8L delivers up to 13 trillion operations per second in PCIe card configurations between 52 and 208 TOPS, according to the Tel Aviv-based firm that has raised nearly USD 224 million in funding.
  • Hailo-8L PCIe cards can operate ResNet50 at up to 10,000 frames per second.

Hailo Technologies Ltd., a startup that manufactures processors optimized for artificial intelligence, announced that it has expanded its Hailo-8 AI accelerator product line to include a Peripheral Component Interconnect Express card for high-end deployments and a Hailo-8L product for entry-level applications.

The Hailo-8L delivers up to 13 trillion operations per second in PCIe card configurations between 52 and 208 TOPS, according to the Tel Aviv-based firm that has raised nearly USD 224 million in funding. It is designed for processing-intensive applications such as video stream management.

On the ResNet50 benchmark model, a convolutional neural network, the Hailo-8 Century PCIe cards reportedly attain 500 frames per second per watt. Hailo-8L PCIe cards can operate ResNet50 at up to 10,000 frames per second.

The 52 TOPS card begins at USD 249 and, according to Hailo, can reduce edge AI deployment costs by up to 70%. They are compatible with industrial temperature ranges and transformer models like the Segment Anything Model, Contrastive Language-Image Pre-Training, and Vision Transformer at the network edge.

According to the company, Hailo-8 Century and Hailo-8L are designed for low-latency, high-efficiency processing of complex pipelines with multiple real-time streams and concurrent processing of numerous models and AI tasks. The scalable processors are compatible with Hailo-8 software.

Hailo Technologies has advanced rapidly. It introduced a new line of processors that execute computer vision models at speeds of up to 20 TOPS and eliminated the need to send data to the cloud for processing five months ago.

The company focuses on imaging applications in markets such as automotive, video surveillance, smart cities, and retail, where a significant quantity of image processing must be performed on local devices.

Orr Danon, Hailo CEO, said, “All of the companies we work with are training in the cloud and deploying at the edge. You create a signature in the cloud and then distribute it to the edge,” which reduces the need for expensive cameras and other intelligent devices.