Muxi's first AI inference GPU is unveiled: INT8 has a computing power of 160TOPS!

According to news on June 9, domestic GPU manufacturer Muxi Integrated Circuit (hereinafter referred to as "Muxi") announced through its official WeChat that at the 16th China International Security Expo held in Beijing on June 7, Muxi (MetaX) It exhibited its first AI inference GPU accelerator card - XiSi N100 and the application of the solution in the security field.

According to reports, Xisi N100 is an AI inference GPU accelerator card for cloud data center applications. It has a built-in heterogeneous GPGPU general-purpose processor core "MXN100" and integrates HBM2E high-bandwidth memory and INT8 integer on a single card. The computing power reaches 160TOPS, and the FP16 floating point computing power reaches 80TFLOPS, with high bandwidth and low latency characteristics. It supports high-definition video processing capabilities of 128 channels of encoding and 96 channels of decoding, and is compatible with multiple video formats such as HEVC (H.265), H.264, AV1, AVS2, etc., and supports up to 8K resolution.

At the same time, Xisi N100 is also equipped with the MXMACA software stack independently developed by Muxi, which is closely coupled with the hardware architecture, supports network models of multiple mainstream frameworks, and supports mainstream computer vision processing and multimedia processing frameworks. Customers can quickly migrate various application scenarios and business models, and can use ModelZoo immediately for continuous improvement.

At the Security Expo, Mu Xi demonstrated the practical applications of XiSi N100 in security fields such as face recognition, vehicle detection, and license plate recognition. Mu Xi has also joined the Beijing Security Industry Association.

According to reports, Xisi N100 has achieved large-scale mass production and can be widely used in smart cities, smart security, smart transportation, cloud computing, intelligent video processing and other scenarios.

In addition to the MXN series of GPUs (Xisi) for AI inference, Muxi also has the MXC series of general-purpose GPU (GPGPU) chips for AI training, AI inference and general computing, and the MXG series for graphics rendering acceleration. GPU, this also means that Muxi has full-stack GPU chip products.

Mu Xi said that its products all use completely independently developed GPU IP, have instruction sets and architectures with completely independent intellectual property rights, and are equipped with a complete software stack (MXMACA) that is compatible with the mainstream GPU ecosystem, and are highly energy efficient and versatile. With its natural advantages, it can build comprehensive ecological solutions integrating software and hardware for customers, which is the cornerstone of computing power in promoting data center construction and industrial digitalization and intelligent transformation and upgrading in the context of "dual carbon".

Editor: Xinzhixun-Rurounijian

The above is the detailed content of Muxi's first AI inference GPU is unveiled: INT8 has a computing power of 160TOPS!. For more information, please follow other related articles on the PHP Chinese website!