Table of Contents
Announcement of <a href="http://www.world-today-news.com/nvidia-is-said-to-be-developing-a-new-geforce-rtx-50-using-3nm-process-technology-which-is-rumored-to-be-twice-as-fast-as-the-current-rtx-40/" title="NVIDIA is said to be developing a new GeForce RTX 50 using 3nm process technology, which is rumored to be twice as fast as the current RTX 40.”>Blackwell GPU-based ultra-high-performance AI supercomputing platform
AI accelerator ‘Maia’ and eco-friendly solution ‘Corbat’
[이코노믹데일리] Microsoft (MS) unveiled AI infrastructure equipped with NVIDIA’s latest Blackwell GPU, setting a new standard for the AI supercomputing era. At the annual developer conference ‘Ignite 2024’ on the 20th, ‘Azure ND GB200 V6’ was introduced and announced that it would provide the highest performance ever by connecting hundreds of thousands of GPUs through a network.
Azure ND GB200 V6 is a custom cloud server for modern large-scale language models (LLM) and other AI applications. This system is designed around NVIDIA’s GB200 superchip and boasts excellent data processing speed by connecting two Blackwell GPUs and Grace CPUs via NV Link-C2C. This achieves programming simplification along with powerful performance that can handle trillions of parameters.
Microsoft CEO Satya Nadella introducing Azure ND GB200 V6 with NVIDIA Blackwell.[사진=MS]
CEO Satya Nadella emphasized, “Azure ND GB200 V6 will set a new standard for AI technology in the cloud by dramatically increasing the speed of large-scale AI model learning and inference.” This system provides up to twice the performance improvement and enhanced data security compared to existing systems, and is suitable for highly sensitive fields such as healthcare, finance, and national defense.
At the site, ‘Azure HBv5 VM’, jointly developed with AMD, was also unveiled. This platform is based on AMD’s latest EPYC 9V64H processor and is equipped with high-bandwidth memory and high-performance Zen 4 cores, providing speeds up to 35 times faster than before. In particular, it is optimized for tasks that require a high level of computational power, such as weather modeling and aerospace simulation.
MS explained that HBv5 VM uses NVIDIA InfiniBand networking to connect numerous GPUs and provides up to 8 times higher performance than on-premise systems.
MS also announced its own designed AI accelerator ‘Maia’ and Arm-based processor ‘Cobalt’. Maia is the first custom accelerator designed to process AI workloads in the Azure cloud and is currently working on Azure OpenAI inference. Corbat is a solution that reduces power consumption by 40% and maximizes energy efficiency.
“Maia plays a key role in processing Microsoft’s most impactful services,” said CEO Satya Nadella.
Meanwhile, MS is focusing on strengthening its security infrastructure in addition to advancing AI technology. The hardware security module (HSM) released this time is dedicated hardware for data encryption and key management and is applied throughout Azure data centers. Additionally, Azure Boot DPU (ADP) alleviates cloud network load, reduces power consumption by one-third and improves performance by up to four times.
CEO Satya Nadella said, “Maximizing cost-performance ratio and improving efficiency are the core of the cloud environment,” adding, “We are providing the optimal environment through continuous innovation with industry partners.”
What are the expected impacts of the Azure ND GB200 V6 on the future of AI supercomputing?
1. Guest 1: As an AI expert, what do you think about the announcement of the Azure ND GB200 V6, a platform powered by NVIDIA’s latest Blackwell GPU? How will it impact the AI supercomputing industry and what specific advantages does it offer over existing systems?
Guest 2: As a technology journalist, I’m curious about the significance of the collaboration between Microsoft and NVIDIA in developing this ultra-high-performance AI supercomputing platform. Can you discuss the technical aspects of the Blackwell GPU and how it contributes to the overall performance of the system? Additionally, what challenges did they face during the development process, and what were some key decisions made to ensure its success?
2. Guest 1: The introduction of the AI accelerator ‘Maia’ and eco-friendly solution ‘Cobalt’ seems to be a strategic move by Microsoft. Could you explain the role of ‘Maia’ in processing Microsoft’s most impactful services, and why it’s important for the company to focus on energy efficiency with ‘Cobalt’?
Guest 2: As an observer of the tech industry, I’m interested in the integration of security features into Azure ND GB200 V6. Could you elaborate on the hardware security module (HSM) and Azure Boot DPU (ADP), and how they contribute to enhancing data encryption, key management, and network performance? Furthermore, what measures are being taken to ensure the security of sensitive information in this age of rapidly advancing AI technology?
3. Guest 1: The announcement of Azure HBv5 VM based on AMD’s latest EPYC processor indicates a shift towards heterogeneous computing. How does this benefit Microsoft’s customers, and what are some other ways in which the company is embracing diverse hardware architectures to enhance the performance of its cloud solutions?
Guest 2: As a business analyst, I’m impressed by the focus on maximizing cost-performance ratio and improving efficiency in the cloud environment. Can you provide examples of how Microsoft is achieving this through its continuous innovation with industry partners like AMD and NVIDIA? Additionally, what challenges does the company face in balancing cost, performance, and sustainability in
