VDSP-PiCubeT20 

       We have designed and manufactured the first 3D stereo parallel processor PICUBE-T20 based on the stereo computing PiCube-ISA 108 VDSP open source project, and developed a full range of AI deployment ecosystems with supporting software compilers and basic software libraries. PICUBE-T20 innovatively proposes a distributed high-density computing solution, providing a general-purpose, high-efficiency, low-power, low-cost computing platform for high-density computing in AI or other fields, empowering enterprises, especially small and medium-sized enterprises (SMEs), AI industry practitioners, and AI application creators, and promoting the popularization and personalization of AI and the popularization of AI applications.

Universal AI Compute Card-AGILE256G

 Hardware parameter

       1)Integration of 32 pcs 3D stereoscopic parallel processors PICUBE-T20

    2)Provides 65536 computing cores

    3)Computing precision FP16

    4)On-card storage capacity up to 256GByte

    5)Interface is pcie gen 4 x 4

    6)Single card power consumption 200W

 Software Ecology
       1)Support for common neural network frameworks

    2)Supports inference, fine-tuning, knowledge distillation

  Scalability
     The number of integrated processors PICUBE-T20 can be customized or tailored according to the amount of computational tasks.

Universal AI Server - SOBER

 Hardware

     Integrated with 8 AGILE 256G compute cards

     Air-cooled,energy saving, ECO friendly

  Software

     Supports common neural network frameworks 

  Software Ecology

     Supports common neural network frameworks

     Arithmetic assistance independent of CPU and GPU platforms

     Supports full training

 Scalability

    The number of AGILE256G cards can be cut or increased according to the amount of computing tasks.

General Purpose AI Arithmetic Cluster-SAGER

Hardware

    Distributed architecture based on SOBER servers

    Air-cooled

Software ecology

    Realize aggregation of computing power, efficient scheduling and load balancing of cluster servers.