The GTC 2017 organized by NVIDIA was held in San Jose from May 8th to 11th. The theme of this conference is depth learning and artificial intelligence, involving biomedical, accelerated analysis, autopilot and computer vision and many other directions. As one of the main sponsors and invited participants of GTC, Sugon has demonstrated a complete set of XSystem hardware and software integration systems with years of precipitation in high performance computing and cloud platform technology.
The highest intensity, the strongest computing power of the depth of learning high - performance computer SuperBOX
SuperBOX is a new generation of high-intensity depth learning server, 4U space has 20 built-in PCI-E interfaces, can carry GPU, FPGA, MIC and other processing units. PCI-E Fabric architecture and 960Gbp/s communication rate to ensure the efficiency of parallel computing. When equipped with 20 latest NVIDIA P100 GPU, stand-alone computing capacity can reach up to 184 TFLOPS, which is the world's strongest server with single computing power. While it is perfect support for NVIDIA's next-generation Volta architecture GPU processor, which is the high-end depth of learning user's best choice. If adopt with a 36-port InfiniBand switch, it can even expand into a powerful computing cluster with 36 SuperBOX nodes and total of 648 GPUs, providing incredible computing support for deep learning model training.
Heterogeneous High Intensity Computing Node W780 - G20
The W780-G20 is a GPU server product specifically designed for deep learning and training platform. It supports maximum 8 GPUs in 4U space, and integrate NVIDIA DIGITS interactive depth learning system. It provides a perfect deep learning experience for small users through software and hardware.
The best choice for enterprise users W580-G20
Sugon W580-G20 depth learning machine integrate NVIDIA DIGITS interactive depth learning system, excellent system architecture to ensure that users in a single system can support four industry mainstream GPU acceleration card, while strict control of the system power dissipation, excellent scalability and high cost-effective to make it become the best choice for enterprise users.
Tuning, algorithm, management - Trinity software platform
In addition to the depth of learning hardware products, Sugon also deployed the three-dimensional software platform concentrate by a set of tunings, algorithms, management of, constitute by the depth of learning framework Xsharp and visual high-performance computing cluster management platform Gridview together.
XSharp depth learning framework provides a large-scale algorithm tool set, and deployed in distributed parallel system optimization, distributed parallel machine learning execution mode optimization, embedded module optimization, the performance is far more than Caffe (cuDNN).
Gridview is the Sugon self-developed visualization of high-performance computing cluster management platform. Can help users monitoring of the running status of the GPU in real-time and job scheduling, to provide performance monitoring, mobile control, cluster management tools and other modules.
Based on the Sugon-ICT-NVIDIA joint laboratory platform and the depth of the R & D team supported by the post-doctoral mobile station, the Sugon of the R & D team as a key R & D unit of the E-level prototype will integrate high-performance computing and future development in the future Cloud platform and other technical advantages, to create a broader user groups for the outstanding product system, pioneering the depth of learning applications, help in-depth learning technology blossom in all walks of life.