SHENZHEN, CHINA – September 20, 2024 -Huawei Cloud today unveiled CloudMatrix, its AI-native cloud infrastructure architecture, at the Huawei Connect 2024 conference. This groundbreaking platform is designed to meetthe surging demand for large-scale, efficient, stable, and flexible compute power in the AI era.
The demand for AI compute power is outpacingthat of general-purpose computing, making the construction of elastic and efficient, diverse compute infrastructure critical for AI development. CloudMatrix addresses this challenge by seamlessly integrating and pooling resources like CPUs, NPUs, DPUs, storage, and memory into adistributed, peer-to-peer, fully interconnected architecture. This everything-is-pool-able, everything-is-equal, everything-is-combinable approach allows for the evolution from individual compute power to a matrix-basedapproach, delivering a powerful AI compute engine for customers.
Key Features of CloudMatrix:
- Everything is Pool-able: Huawei provides robust general-purpose and intelligent computing through CPUs and NPUs. Huawei Cloud’s distributed QingTian technology breaks the limitations of single servers, enabling the pooling and flexible configuration ofdiverse resources within servers, including CPUs, NPUs, DPUs, storage, and memory. This results in a large-scale, tightly coupled, diverse compute pool architecture, significantly enhancing compute power.
- Everything is Equal: CloudMatrix utilizes the high-bandwidth ScaleUp network, transitioning from traditional Ethernet to sharedbus Ethernet. This ensures high-speed interconnection of pooled compute power and, through adaptive topology awareness, overcomes the limitations of linear clusters. By combining the ScaleOut network for traditional compute clusters with the ScaleUp high-bandwidth network, Huawei Cloud has increased resource interconnection bandwidth by an order of magnitude. The new network protocol enablesunified identification, access, and message communication mechanisms across physical servers. This facilitates the unified pooling of diverse resources like CPUs, NPUs, storage, and memory across servers. The fully equal connection of resources enables both horizontal and vertical scaling, breaking the linear limitations of computing clusters. This can improve the linearity of large model trainingclusters to 95% and achieve minute-level recovery from network failures.
- Everything is Combinable: CloudMatrix allows for the on-demand combination of diverse compute resources and flexible adjustment based on application needs. Building upon the unified, diverse compute logic pooling and high-speed, fully equal interconnection, theYaoGuang intelligent cloud brain provides unified modeling and flexible scheduling and combination of cloud resources, offering them to applications on demand. This covers the training and inference of both large and small models and can be applied to various scenarios, including data analysis and media content generation.
Huawei Cloud announced that the new generation of Ascend AI cloudservices based on CloudMatrix will be launched by the end of the year.
Industry Impact:
The launch of CloudMatrix marks a significant step forward in AI infrastructure development. By providing a highly scalable, efficient, and flexible platform, Huawei Cloud empowers businesses to accelerate their AI adoption and innovation. This platform will be crucial fordriving advancements in various sectors, including:
- Artificial Intelligence Research: CloudMatrix provides the necessary compute power for researchers to develop and train advanced AI models, pushing the boundaries of AI capabilities.
- Data-Driven Businesses: Businesses across industries can leverage CloudMatrix to analyze vast datasets, extract valuable insights, andoptimize operations.
- Emerging Technologies: CloudMatrix fuels the development of cutting-edge technologies like autonomous driving, robotics, and personalized medicine, accelerating their adoption and societal impact.
The launch of CloudMatrix underscores Huawei Cloud’s commitment to providing a comprehensive and innovative AI platform that empowers businesses and researchers to harness thetransformative power of AI. This platform is poised to play a critical role in shaping the future of AI development and its impact on various industries and aspects of human life.
Views: 0