HUAWEI CLOUD ModelArts Wins First Prize in Stanford DAWNBench Deep Learning Benchmark Competition
Mar 20, 2019 GMT+08:00
On March 20th, Stanford University published the latest DAWNBench results. HUAWEI CLOUD's one-stop ModelArts platform for AI development received first place in image classification training and inference performance. These awards further ModelArts' global leading position as a deep learning platform.
The ModelArts substantially reduced model training time andachieved excellent inference performance. In only 4 minutes and 8 seconds,ModelArts trained ResNet-50 on ImageNet with 128 V100 GPUs. This is more thandouble last December’s speed of 9 minutes and 22 seconds, and four times the speed of fast.ai running on AWS. In image classification performance, it was 70% faster than the second-place vendor. Moreover, it was 3 times faster than AWS's solution and 8 times faster than Google's.
ModelArts' Leading Deep Learning Platform Technology
Deep learning is a fundamental technology within the field of artificial intelligence. Recently, deep learning has been applied to myriad fields, such as autonomous driving, Internet, security, and medical care. The required data has been increasing alongside AI model training capability. Requirements for AI computing resources and time are increasing, making deep learning training and inference vital.
HUAWEI CLOUD ModelArts supports massive data preprocessing,
large-scale distributed training, automated model creation, and on-demand"terminal-edge-cloud" deployment. It helps users quickly create and deploy models, and manage every aspect of AI workloads. ModelArts' superior distributed model training and rapid inference technology secured its spot in Stanford's
DAWNBench competition, the benchmark suite measuring a variety of deep learning training and inference tasks.
Inference performance rankings
Training time rankings
Effectively Using Cloud Platforms and AI Resources
Difficult aspects of AI development include scarce andexpensive computing power and ineffective development. Consider developers, enterprises,and research institutes as examples. They often expect large-scale computing clusters for training processes, but purchasing and maintaining these is expensive.
ModelArts customers receive high-performance, scalable GPU resources and pay only for what they use. ModelArts leverages synergies between
HUAWEI CLOUD hardware (high-end servers, network, and storage) and its software and algorithms to increase computing speed. Its distributed systems and improved algorithms bring excellent user experience. ModelArts has become more accessible as an AI requirement with fast turnaround and excellent performance.
On March 21st, the HUAWEI CLOUD AI Market was launched to create an AI developer community, and further the development and implementation of fast AI. It provides AI modelling, API-enabled transactions, competition examples, and other sharing capabilities. This marketplace caters to educational institutions, application developers, and solution integrators. HUAWEI CLOUD AI Model Market is the China's first platform to publish and provide subscription to AI models.
The one-stop ModelArts AI development platform is already becoming vital to healthcare, smart manufacturing, smart city, construction, autonomous driving, parking, and many other industries. It has helped companies such as Kingmed Diagnostics, Glodon, and Cloudansys implement their AI applications.
By the end of 2018, HUAWEI CLOUD had launched 56 AI services, with a total of 159 functions, and had explored more than 200 projects applying AI to enterprise-specific scenarios. HUAWEI CLOUD is committed to becoming the new engine driving upgrade of diverse industries.