Announcement of Hanguang 800 AI Accelerator
| Artificial Intelligence | Cloud Computing | Semiconductor Technology |
Updated By: History Editorial Network (HEN)
Published: | Updated:
4 min read
The Hanguang 800 AI Accelerator represents a significant advancement in artificial intelligence hardware, developed by Alibaba's T-Head and DAMO Academy. This accelerator is built with 17 billion transistors using a 12 nm manufacturing process, which allows for enhanced performance and efficiency. The design focuses on high-speed image processing, boasting a capability of 78,563 images per second (IPS) for inference tasks. Additionally, it achieves an impressive efficiency of 500 IPS per watt when running the ResNet-50 model, a popular deep learning architecture used for image classification. This level of performance positions the Hanguang 800 as a competitive option in the AI accelerator market, catering to various applications that require rapid image processing and analysis.
The Hanguang 800 is not only a technological achievement but also a strategic move by Alibaba to expand its cloud services. By making the accelerator available for rent on Alibaba Cloud, the company aims to provide businesses and developers access to powerful AI capabilities without the need for significant upfront investment in hardware. This approach aligns with the growing trend of cloud-based AI solutions, allowing organizations to leverage advanced computing resources for their AI projects. The introduction of the Hanguang 800 is expected to enhance the capabilities of AI applications across different sectors, including e-commerce, finance, and healthcare, by enabling faster and more efficient data processing.
Primary Reference: Announcing Hanguang 800: Alibaba's First AI-Inference Chip

Explore the Life Moments of Alibaba Group | 