Hailo-8 deep learning processor capable of 26-TOPS
Hailo said it is sampling the chip to companies across multiple industry sectors but with a focus on automotive.
The Hailo-8 processor includes memory and runs at up to 26 tera operations per second (TOPS), and when running the ResNet-50 benchmark consumes 1.67 watts and achieve 2.8TOPS/W. Hailo offers a software development kit (SDK) that was co-developed with the hardware.
Areas such as advanced driver-assistance systems (ADAS) and smart cities and smart homes often require the use of high-performance cameras to perform tasks such as semantic segmentation and object detection in real time – tasks which Hailo’s chip performs at full resolution while consuming a few watts.
According to preliminary results comparing Hailo-8TM to Nvidia’s Xavier AGX, which runs NN benchmarks such as ResNet-50,
Hailo-8 consumes almost 20 times less power than Nvidia’s Xavier AGX, while performing the same tasks under ResNet-50.
Hailo was founded in 2017 by former members of the technology unit of the Israeli Defense Forces.
Related links and articles:
News articles:
China backs Israeli deep learning processor startup
Israel startup funded to develop deep learning processor
Analog and digital circuits for machine learning
Vayyar increases resolution of radio sensor
