Recently, YUAN High Tech has become a video capturing I/O devices manufacture partner of NVDIA GPUDirect for Video. NVIDIA GPUDirect for Video is a technic to effectively reduce latency from external video source to internal GPU, further to accelerate the development of AI video recognition for different markets.
Please Visit : NVIDIA GPUDirect for Video
Traditional image recognition requires the image data to be first received from the image capture card to the memory, then a copy of it is transferred to the GPU for time-consuming data analysis, where the image loading time highly affects the performance of the application. As the resolution of images is getting higher, the amount of data is also becoming larger. For instance, the work of transmission and copying has caused a heavy burden on the CPU. With the support of NVIDAI GPUDirect for Video, the data movement and access of NVIDIA data center GPUs are enhanced. Developers can synchronize transmission to allow GPU and image capture card to work together without manual waiting state to ensure completion. The delay can also be reduced by transmitting smaller data chunks without having to wait for the entire field or frame to complete before transmitting.
Huengpei Lin, the general manager of YUAN said: “We are entering a new era - from disclosure to enclosure. AI applications on videos have changed the way how people used to communicate, interact and learn with audio-vision. With the popularity of streaming video, the changing speed will continue to accelerate in the future ”
YUAN provides HDMI and SDI video solutions up to 4K / 8K in the B2B professional video and audio market. It provides PCIe / miniPCIe / M.2 video capture cards, USB2.0 / 3.0 video capture boxes, video conversion boxes, streaming encoders and other products to meet the needs of various markets. All kinds of audio-visual products of all interfaces use the same set of audio-visual software development kits. In 2011, YUAN began to enter the field of AI visual analysis. With four major modules including streaming service and identification analysis, even embedded systems can enjoy the advantages of low latency of ultra-high-definition images!
YUAN also participated in the NVIDIA GTC, Global Artificial Intelligence Technology Conference, on November 9, 2021. Introducing how YUAN capture cards and SDK develop real-time image AI edge computing for medical systems. The underlying architecture is integrated with TensorRT, and the float point 16 quantization is applied to improve the calculation performance of the inference. In the past, it took Pytorch about 20ms, while TensorRT only takes 9ms now.
Accelerating Med Tech Deployment in Emerging Markets [ A31515 ]
Yuan High-Tech ( GTSM: 5474 ), established in 1990 with headquarters in Taipei, Taiwan, is the leading force in the digital home entertainment, broadcasting and surveillance market. We deliver an extensive range of high-quality video/audio products for system integrators and ODM customers. As one of the major providers to these markets, YUAN’s solutions can be found within the product lines of worldwide leading ODM customers and channel partners.
We provid PCIe/miniPCIe/M.2 video capture card series, USB2.0/3.0 capture box/dongle series, video converter boxes series and streaming encoder boxes series to meet different markets’ needs. All video capture products with different form factors are supported by one driver and the same SDK which follows Windows Directshow or Linux v4l2 standards.
Our R&D teams design and develop hardware, drivers, firmware, FPGA and SDK packages to provide the best technical support for our customers. For more information, please visit : www.yuan.com.tw