컴퓨팅에서 처리해야 할 데이터가 기하급수적으로 빠르게 증가하는 반면 이를 처리해야 하는 프로세서의 연산은 데이터의 증가 속도를 따라잡기 못하고 있다. 이에 젠슨 황 CEO는 컴퓨테이션 인플레이션이 발생하고 있다면서 데이터센터의 비용과 전력량이 증가하고 있다고 설명했다.
▲A scene from NVIDIA Computex 2024. While the CPU scaling curve is gradually becoming flatter, the data growth curve is becoming steeper.
Early release of next-generation Rubin platform equipped with HBM4
Omniverse, Accelerated Computing + AI = Data Visualization
Breaking the MS alienation theory, RTX-based copilot AI PC
Support for acceleration of data centers, AI factories, and robot factories
While the amount of data that needs to be processed in computing is increasing exponentially, the computational power of the processors that need to process it is not keeping up with the rate at which the data is increasing. In response, CEO Jensen Huang explained that computational inflation is occurring, and that the cost and power consumption of data centers are increasing.
NVIDIA CEO Jensen Huang raised expectations at Computex 2024 by unveiling a new semiconductor roadmap with a one-year cycle to overcome the challenges facing computing and semiconductors amid the spread of artificial intelligence.
■ Next-generation Rubin platform following Blackwell
The Rubin platform, which was first unveiled, is the successor to the soon-to-be-released Blackwell platform. Rubin features advanced networking with HBM4, a new GPU, a new Arm-based CPU called Vera, NVLink 6, the CX9 SuperNIC, and the X1600 converged InfiniBand/Ethernet switch.
Jensen Huang began his speech by saying that Nvidia is reducing the cost of turning data into intelligence.
“Accelerated computing is sustainable computing,” he stressed, explaining that the combination of GPUs and CPUs can deliver up to 100x speedups while increasing power consumption by only three times, and can achieve 25x higher performance per watt than using CPUs alone.
■ Emphasis on data visualization and digital twins through omnibus
▲NVIDIA CEO Jensen Huang demonstrated Earth-2 during his keynote speech at Computex 2024 . / (Photo: Nvidia)
“Over the past 60 years, we’ve seen just a few key technologies transform the computing industry, and now we’re at another moment of transformation,” said Jensen Huang, CEO, in his keynote. “These two technologies—accelerated computing and artificial intelligence running in an omniverse—will fundamentally transform the computing industry.”
NVIDIA highlighted data federation and visualization via NVIDIA Omniverse, showcasing a number of simulation demos including NVIDIA Earth-2. In particular, Earth-2 is a full-stack, open platform that accelerates climate and weather predictions with interactive, high-resolution simulations, using CorrDiff, a generative AI model that downsizes kilometer-scale climate simulations.
We also demonstrated AI tools from NVIDIA Inception Startup Program members built on NVIDIA NIM and NVIDIA accelerated computing. NIM, an inference microservice that delivers models as optimized containers, can be deployed in the cloud, in a data center, or on a workstation.
Over 200 technology partners, including Cadence, Cloudera, Cohesity, DataStax, NetApp, Scale AI, and Synopsys, are integrating NIM into their platforms, accelerating the deployment of generative AI for domain-specific applications such as Copilot, Code Assistant, and Digital Human Avatar. Hugging Face is said to currently offer NIM starting from Metarama 3.
■ GeForce RTX-based AI PC launch prospects
▲NVIDIA RTX AI PC / (Photo: NVIDIA)
NVIDIA has announced the launch of NVIDIA RTX AI PCs powered by RTX technology. More than 200 RTX AI laptops are expected to be released and are expected to support Microsoft's Windows Copilot.
MS and NVIDIA are working together to enable developers to bring new generative AI capabilities to Windows native or web apps with easy API access to the RTX-accelerated SLM. The RTX-accelerated SLM supports RAG features running on the device as part of the Windows Copilot Runtime.
The RTX AI Toolkit and the newly available PC-based NIM inference microservice for the NVIDIA ACE Digital Human Platform enable AI accessibility. NVIDIA has unveiled Project G-Assist, a demo of its AI assistant technology that demonstrates contextually aware support for PC games and apps.
■ Data Center/AI Factory Construction Solution Released
Jensen Huang, CEO, announced that ASRock Rack, ASUS, Gigabyte, Ingrasys, Inventec, Pegatron, QCT, Supermicro, Wistron, and Wiwin are offering cloud, on-premise, embedded, and edge AI systems using NVIDIA GPUs and networking.
The NVIDIA MGX modular reference design platform includes the GB200 NVL2 platform for large language model (LLM) inference, retrieval-augmented generation (RAG), and data processing, and supports Blackwell.
Leveraging the high-bandwidth memory performance provided by the NVLink®-C2C interconnect and the dedicated decompression engine of the Blackwell architecture, data processing speeds are up to 18x faster and energy efficiency is improved by 8x over x86 CPUs.
To address all types of applications, we will offer a wide range of products, from single GPUs to multi-GPUs, from x86-based processors to Grace-based processors, and from air-cooling to liquid-cooling technology.iv>
AMD and Intel are supporting the MGX architecture with plans to offer their own CPU host processor module designs for the first time.
“The next industrial revolution has begun,” said CEO Jensen Huang. “Companies and countries are working with NVIDIA to transform trillions of dollars of traditional data centers into accelerated computing and to build a new type of data center—the AI factory—to produce a new product: artificial intelligence.”
■ NVIDIA Industrial AI Accelerates Robot Factory Digitalization
▲NVIDIA, AI and Omniverse to support accelerated digitalization of robot factory industry / (Photo: NVIDIA)
Major Taiwanese electronics manufacturers, including Delta Electronics, Foxconn, Pegatron, and Wistron, announced that they are transforming their factories into more autonomous facilities with new reference workflow technology from NVIDIA.
Manufacturers have leveraged these workflows to build digital twins for real-time factory layout simulation. This allows manufacturers to optimize space, processes and efficiency without having to make costly physical plant changes.
Jensen Huang presented an example of how Foxconn used NVIDIA Omniverse, Isaac, and Metropolis to create a digital twin that combines vision AI and robotics development tools to improve robotics equipment.
Global electronics companies are integrating NVIDIA’s autonomous robotics into their factories, using simulations in Omniverse to test and validate new AI for the physical world, including more than 5 million pre-programmed robots worldwide.
“Every factory will be robotized,” Jensen Huang explained. “Factories will coordinate robots, and robots will produce robotized products.”
“With the advent of AI for manufacturing, every factory is becoming increasingly autonomous, driven by the transformative impact of generative AI and digital twin technologies,” said Deepu Tala, vice president of Robotics and Edge Computing at NVIDIA. Tala emphasized how NVIDIA Omniverse, Metropolis and Isaac solutions can deliver operational efficiency and cost benefits.
.jpg)
