Strong Video Processing Capabilities
Supports video decoding up to 320 channels of 1080p@30fps and encoding up to 160 channels of 1080p@30fps, and is capable of efficiently process HD and UHD video data, fully meeting the stringent requirements for high-performance video processing in video transcoding, remote gaming, and other applications.
High AI Processing Power
Uses five self-developed RISC-V general-purpose intelligent computing SoCs to provide up to 200TOPS of processing power at INT8 precision for various AI applications.
Large Memory Capacity and High Bandwidth
Boasts up to 320GB memory capacity and provides sufficient storage for the efficient operation of large AI models, significantly improving the reasoning efficiency of complex AI tasks and large data sets.
High-Precision LLMs
The built-in dedicated processing units for DNN inference can efficiently handle the inference tasks of high-precision LLMs, meeting the needs of complex model architecture and large-scale parameter computing.
Flexible Interconnection Capabilities
Supports interconnection of up to eight GPUs within a single server and server clustering, and can flexibly build large-scale distributed computing systems to meet the extension needs of AI inference servers at different scales.
Parallel Video Processing Capabilities
Uses five high-performance intelligent computing SoCs and has large memory capacity to process massive amounts of video streams at the same time and handle heavy-duty tasks such as parallel video transcoding.
High-Performance RISC-V Architecture
Adopts an advanced RISC-V architecture, features high performance, low power consumption, and high flexibility and scalability, and can swiftly respond to the scheduling needs of different computing tasks.
Support for Hybrid AI Computing Acceleration
Provides strong hybrid AI computing acceleration capabilities, widely applicable in diverse applications such as AI inference servers, video cloud platforms, and smart park management systems.
京公网安备 11030102011303号