On July 4th, the 2024 World Artificial Intelligence Conference (WAIC) and High-Level Conference on Artificial Intelligence Global Governance officially kicked off, showcasing the cutting-edge concepts and achievements in artificial intelligence technology to the world. Fang Lei, Chairman of DataCanvas, was invited to attend the opening ceremony and delivered a keynote speech at the "Towards AGI: Renewal of Large Models and Industrial Empowerment" forum, sharing the latest insights on the intelligent computing industry.

The "Towards AGI: Renewal of Large Models and Industrial Empowerment" forum, as a significant theme forum of WAIC 2024, was hosted by the China Academy of Information and Communications Technology. Positioned as a "Top-tier Event for AI Large Models," the forum invited renowned academicians and influencers from the field of artificial intelligence both domestically and abroad. Leaders from industry giants such as Alibaba Cloud, Huawei, and DataCanvas, as well as experts from top universities and research institutions, gathered with key figures from regulatory bodies, national laboratories, and international organizations, to explore new potentials for productivity driven by large models.

Fang Lei, Chairman of DataCanvas, delivers a speech

Invited to deliver a keynote on "Intelligent Computing Operating System: Embracing a New World Defined by Software," Fang Lei evangelized the necessity of the intelligent computing operating system for intelligent computing centers, showcasing the breakthroughs and boundless possibilities of artificial intelligence foundational software in the era of computing economy.

Intelligent Computing Operating System: Constructing the Consumption Ecosystem for Computing Power, A "New Solution" for the Era of Universal Computing Power

As a new blueprint for AI infrastructure construction in China, intelligent computing centers are accelerating their construction process under unprecedented cohesion within the AI ecosystem. With the widespread experience of universal large models and the explosion of industry-specific large models, how to efficiently deliver better computing resources to computing power users while unlocking the economic value of intelligent computing has become a multifaceted challenge faced by current intelligent computing centers.

The flexible needs of computing power users are gradually increasing, and the efficiency of GPU usage urgently needs to be improved. Intelligent computing centers cannot precisely match consumption demands merely through bare-metal provision of "measurable, divisible, and convertible" computing power. Amidst the trend of hardware homogenization, AI software that operates on the hardware becomes the "new solution" for intelligent computing. The intelligent computing operating system, as an innovative AI software product of the intelligent computing era, is not just a traditional bridge between hardware and software, but a critical link for constructing the consumption ecosystem of upstream and downstream computing power and aiding the development of the AI industry.

So, what are the challenges faced by a high-performance intelligent computing operating system, and what should its capabilities be?

Fang Lei introduced that the new round of fundamental computing transformation is triggered by the rise of large models, and matching the demand for high-speed flexible algorithms with the relatively fixed bottom-layer hardware becomes the core challenge of the intelligent computing operating system. The new intelligent computing operating system faces complex challenges such as difficulty in managing heterogeneous computing resources, insufficient capabilities in large-scale cluster computing power scheduling, underutilized resources, lack of AI acceleration optimization, and high barriers to training and fine-tuning AI large models. Therefore, the development of this highly engineered operating system still has a high threshold, requiring mature, standardized AI products to directly address pain points.

Fang Lei, Chairman of DataCanvas, delivers a speech

Moreover, with the rapid development of the AI industry, the accelerated elimination of information asymmetry, and differing from the single way of acquiring computing power through the purchase or rental of "bare metal," future computing power will serve consumers in a universally "accessible and user-friendly" form. Constructing a sustainable ecosystem for consuming upstream and downstream computing power requires the joint development of AI foundational hardware, AI foundational software, and industry-specific large-model applications, with the intelligent computing operating system, as AI foundational software, playing a pivotal revolutionary role in the middle layer.

Alaya NeW Intelligent Computing Operating System: Full-Stack AI Innovation, Driving AI Technology Prosperity

Leveraging DataCanvas's decade-long accumulation of AI software product prowess and industry understanding in artificial intelligence platforms, the "general + industry" large-model matrix, and cutting-edge Agent intelligent systems, the Alaya NeW intelligent computing operating system, with its integrated innovation from top to bottom, directly addresses pain points, accelerating the development of intelligent computing centers.

• NeW Architecture: A new hardware-software integrated architecture for AI

Alaya NeW, with its advanced heterogeneous management technology, fully supports the computing power of GPUs from various manufacturers. Optimized network collective communication algorithms configured through high-performance IB and RoCE network architectures, high-performance storage facilities designed specifically for large models, as well as the Serverless elastic high-performance computing architecture, provide fully transparent resource scheduling and management, allowing users to focus solely on core AI training and inference tasks.

• NeW Kernel: Advanced AIDC OS system kernel

Alaya NeW achieves intelligent cross-center computing power scheduling, setting up both full-function Kernel and lightweight Kernel to fully manage different types of computing centers and clusters. For large model tasks, Alaya NeW provides dedicated scheduling algorithms and strategies, further optimizing scheduling through capabilities like fault-aware scheduling, topology-aware scheduling, GANG scheduling, and dynamic fair scheduling, thus enhancing computing power availability comprehensively.

• NeW Storage: Storage facilities designed for large models

Alaya NeW matches the needs of large models with a powerful cross-center data platform NeW Dingo. Through multi-center storage, specialized compression algorithms for corpora, materialized views of file systems, the fusion of file systems and vector databases, native security strategies, it not only achieves 70%-90% space savings but also 50% network IO savings, while corpus processing speed improves by up to 10 times.

• LM Infra: Large model infrastructure

Alaya NeW provides high-performance model training and inference services out of the box (achieving a 100% improvement in training efficiency, a 50% increase in GPU utilization, and a 4x boost in inference speed through algorithm acceleration, compilation optimization, memory optimization, and communication acceleration), secure high-performance private model repositories, and dynamic model inference services with shared resources, becoming high-performance, high-availability large model software infrastructure.

• NeW Agent: Team Up on Demand

Alaya NeW is equipped with a powerful Foundation Agent framework, fully supporting knowledge management, data analysis, and multi-modal intelligent system development. With a complete Agent development toolchain, highly compatible and open to the open-source community, it supports one-click Agent deployment and provides a flexible, elastic runtime environment, accelerating the application of diversified Agents in various industry business scenarios.

Fang Lei pointed out that Alaya NeW can be described as a pioneering and innovatively integrated creation of frontier artificial intelligence technology. Alaya NeW provides an unprecedented interaction model, presenting each layer of AI capabilities from infra to Agent in a convenient interactive operating system form. Alaya NeW is no longer just simple AI software for managing, scheduling, and optimizing computing power; it represents one or a set of Agents. Users can achieve complex AI tasks such as large model training through simple interactive commands, essentially providing a personalized "digital intelligence expert team". Alaya NeW is breaking the barriers of large model and Agent applications through innovative interactions.

In his opening address, Premier of the State Council, Li Qiang, pointed out that China has always actively embraced intelligent transformations, vigorously promoting the innovative development of artificial intelligence. DataCanvas's intelligent computing operating system products are further proof of the company's over decade-long commitment to innovations in AI foundational software development and services. In the future, the company will use the Alaya NeW intelligent computing operating system as a lever to accelerate the construction of a robust and sustainable intelligent computing industry ecosystem with AI ecological partners, contributing the power of independent AI innovation to the rapid development of China's intelligent computing economy!