FREMONT, Calif., May 15, 2017 /PRNewswire/ -- Inspur, a leader in intelligent computing and is ranked by Garter as the top 5 server manufacturer worldwide, introduced Caffe-MPI on the first day of GPU Technology Conference 2017. According to testing results based on the Internationally-accepted ImageNet dataset for deep-learning models, The Caffe-MPI shows excellent parallel scalability and performs nearly two times faster than Google's TensorFlow.
Caffe and TensorFlow are currently the two most popular deep learning frameworks. Caffe is a deep learning framework made with expression, speed, and modularity in mind. Caffe is developed by the Berkeley Vision and Learning Center (BVLC), as well as community contributors and is popular for computer vision. TensorFlow is a software library for numerical computation using data flow graphs, developed by Google's Machine Intelligence research organization.
Caffe-MPI, developed by Inspur, is the world's first high performance MPI cluster version of the Caffe deep learning computing framework. Based on the Berkeley Caffe framework, using MPI technology for parallel optimization of Caffe's data, it integrates parallel computing of multiple devices and cards to enable deployment of large scale training platforms for complete large scale data sampling training. Under the performance of 16 GPU cards in 4 nodes, the Caffe-MPI is 13 times more efficient than a single card, and processes almost twice the number of images per second than the TensorFlow 1.0 with the same cluster structure.
Caffe-MPI's superb performance comes from its groundbreaking parallel computing design. Caffe-MPI has two layers of communication modes: The communication between GPU cards in a nodule and the RDMA's global communication between nodules which tremendously reduce the bottleneck of network communication and overcome the effect of uneven bandwidth between network and PCIE in traditional communication. The Caffe-MPI unleashes NVIDIA's latest CUDNN capabilities fully to archive high performance expansion.
During GTC 2017, Inspur showcased a complete product solution for artificial intelligence users, including
- AGX-2, the world's first 2U 8 P100 GPUs with NVLink or PCI-E interface, AI supercomputer.
- SR-AI Rack, jointly launched by Inspur and Baidu, is a GPU Box with the most advanced fully integrated hyperscale solution for AI.
- AIStation is an artificial intelligence Deep Learning management system developed by Inspur to support many types of deep learning frameworks, with the ability to deploy deep learning training environments rapidly, to manage deep learning training tasks comprehensively, and to provide an efficient and user-friendly platform for AI users.
Artificial intelligence deep learning is one of Inspur's key three computational intelligence services. In April this year, Inspur has established a new AI Department to introduce cutting-edge computing platforms geared towards developing innovative AI applications and continue to inspire intelligent computing. Inspur has already become the major contributor of AI computing and GPU processors for the world's best web services companies such as Baibu, Alibaba and Tencent. We are also working very closely with iFlytek, Qihoo 360, Sogou, Toutiao, Face ++ and other leading companies in the field of artificial intelligence, helping customers improve the magnitude of performance in voice, image, video, search, network and many more.
SOURCE Inspur Electronic Information Industry Co., Ltd