Release date：2025-02-22Author source：KinghelmViews：29
According to Huang, the reason why generative AI can communicate with people about text and images is that the text and images have been digitized and understood. Next, many things, including proteins, genes, and brain waves, can achieve digital ——. As long as you understand its structure, or abstract a specific pattern, you can understand its connotation, you can do digital. This is the generative AI revolution.
Undoubtedly, to achieve this future, NVIDIA now has to do is continue to promote generative AI, not only should have enough powerful hardware support, but also provide developers with AI development tools, actively guide those who don't understand how to introduce AI to productivity and enterprise processes of individuals and enterprises, and NIM is clearly NVIDIA future will generate AI further cash technology, is its popularity, consolidate AI ecological, one of the important products.
NVIDIA NIM is an inference microservice that provides models in the form of optimized containers and is deployed in the cloud, data center, or workstation. In other words, NIM provides a simple, standardized way, will generate AI added to the application, AI developers don't have to struggle with multiple models with different functions to generate text, images, video, voice, etc., also don't need to waste time on the complexity of model deployment, reliability and performance optimization, significantly improve work efficiency at the same time, also can be more focus on the business logic.
With NVIDIA NIM, developers can easily build generative AI applications for copilots, chatbots, etc., cutting the time required from weeks to a few minutes. NIM also enables companies to maximize their infrastructure investments. For example, when running Meta Llama 3-8B in NIM, the latter produces up to 3 times more generative AI tokens on the accelerated infrastructure than without NIM. This enables businesses to substantially increase efficiency, using the same computing infrastructure to generate more results.
The next generation of AI-enabled autonomous machines and robots are also an important embodiment of the accelerated landing of "generative AI". After all, "AI can understand the laws of physics and fight alongside humans" is becoming a reality, not just in science fiction.
According to the latest report from UBS, in the coming years, as robots gradually transform from automated to autonomous (AI with self-learning ability), the human perception of robots will be overturned. Related data show that only the global mobile robot market is expected to grow 9 times, from $13 billion in 2021 to more than $123 billion in 2030, it is no wonder that Huang Renxun on multiple occasions described robot computer as "a new type of computer", can transfer all mobile machine to the virtual world.
NVIDIA Isaac The robot platform has a complete set of NVIDIA acceleration libraries, AI base models, and simulation technologies, such as:
Based on the Omniverse, the Isaac Sim is the most realistic robot simulator ever, and the goal is to make the robot unable to tell whether it is in a simulated environment or in the real world;
Support the robot operating system (ROS) developer community, making it easier for ROS developers to build high-performance AI robots on the Jetson platform, so as to achieve a complete "end-to-end machine learning loop" closed loop;
Isaac Manipulator Simplify the development of an AI manipulator or manipulator to seamlessly sense, understand, and interact with the environment; Isaac Perceptor provides multi-camera 3D viewing capabilities for AI-based autonomous mobile robots.
The Isaac ecosystem now has more than 700 companies and partners, a number that has increased fivefold over the past four years. Byd electronics, Siemens, Thai and Alphabet company Intrinsic global more than a dozen leading robot industry, is accelerating NVIDIA Isaac library, based on physics simulation and AI model integrated into its software framework and robot model, to improve the work efficiency of factories, warehouse and distribution center, make the robot human colleagues work more safely, and make the robot perform repetitive or super precision task intelligent assistant.
With the acceleration of digital transformation in various industries, the rise of generative AI, the development of 3D data interoperability, the progress of advanced graphics, and the improvement of simulation computing capabilities from the edge to the cloud, all provide new opportunities for the digitization of enterprise physical processes.
Today, "$50 trillion worth of industry in competing for digital" —— because it not only enables enterprises to create accurate conform to the real physical laws of digital assets, also can through accurate 3D simulation and the construction of physical product environment, help the entire supply chain, production chain and all participants in the industry chain to achieve cost savings and efficiency. When the 3D virtual world is connected with the real physical world, enterprises will be able to use AI to conduct continuous monitoring, constantly optimize the digital twin and physical twin, and promote enterprise operations to a more efficient and intelligent direction.
On the other hand, the core goal of industry users actively building digital twins is to improve operational efficiency and reduce costs, which is expected to save customers billions of dollars. In addition, once the industrial digital twin is built, it will provide an ideal virtual environment for robotics and AI training, and ensure that the AI trained in the virtual environment can be effectively implemented in the real world, realizing the seamless connection between the physical world and the digital world.
As to all walks of life to create and operate digital twin application platform, NVIDIA Omniverse has extended to the world's largest industrial software ecosystem, help industrial enterprises to accelerate the development of software definition products, promote the autonomous mobile vehicles, autonomous vehicles, humanoid robot, intelligent storage and wisdom city application process. The emergence of Omniverse Cloud API will bring a new acceleration to the training, simulation and deployment of AI-based digital twin simulation workflows, further promoting the digital transformation of industry.
In a presentation at the COMPUTEX keynote in June, Huang showed how Foxconn, one of the world's largest electronics manufacturers, developed the digital twin of its factory on the NVIDIA Omniverse platform. First, Omniverse helps Foxconn's team optimize the device layout of the operation process, and the AI cameras monitor worker safety through the NVIDIA Metropolis. Then, Foxconn can use the factory digital twin as a virtual training environment to simulate, test and validate its autonomous mobile robot (AMR) built based on the NVIDIA Isaac Perceptor acceleration library, as well as the AI robot robotic arm supported by the NVIDIA Isaac Manipulator AI model.
Omniverse The application of the digital twin is far beyond industry, automotive, manufacturing, media, construction, energy, scientific computing simulation and many other industries will benefit from this. In the field of carbon emissions, through the technology to convert carbon dioxide into a liquid state and inject it underground, users are able to create underground digital twin models to ensure the safety and efficiency of carbon dioxide storage. The application of this technology could reduce carbon dioxide emissions equivalent to 1,000 vehicles a year.
Earth-2, a global climate model for weather research, combines 3D interactive technology with extensive data training to enable users to simulate and analyze weather trends and atmospheric rivers in digital twin environments, thus improving prediction accuracy for extreme weather events such as hurricanes.
The AI-level digital twinning is also interesting. For example, through the digital twin model of AMR car and station, combined with AI Agents technology, optimize the path planning and real-time response to emergencies. Such applications demonstrate the powerful potential of combining digital twinning with AI and the important role of the NVIDIA Omniverse platform in connecting the physical and digital world.
In fact, the factory digital twin is a necessary part of the future development of productivity, and its essence is a narrow meta-universe. Based on the virtualization development needs of the future economy, GPU graphics rendering and AI assistance are the construction of the basic and core capabilities.
NVIDIA Is the largest player in the AI chip market, and has an absolute leading share in GPU and CUDA. But beyond that, NVIDIA has a range of technologies that can be used to expand the workload on multiple GPUs and systems. These include its on-chip and encapsulation interconnections, NVLink for GPU to GPU communication in servers, Infiniband beyond the pod, and the Spectrum™ -X Ethernet network platform for connecting to the wider infrastructure, which is also seen by the industry as part of the NVIDIA's "invisible Moat."
Spectrum-X is the world's first Ethernet network platform built for AI, which is essentially an end-to-end soft and hardware solution, rather than just a network card or switch. The platform is equipped with SN5600 Ethernet switch and BlueField-3 SuperNIC, through dynamic routing technology and congestion control technology to maximize the bandwidth and noise isolation, can improve the generative AI network performance over the traditional Ethernet network platform of the sixfold,
The reason for AI is that AI workloads are unique and generate a lot of network data streams that can have a huge impact on the network. If these workloads are run on an existing cloud infrastructure built on a traditional Ethernet, they can cause severe congestion, increased latency, and unfair bandwidth allocation, leading to inadequate utilization of the system's GPU resources.
NVIDIA We expect Spectrum-X to generate billions of dollars in revenue within a year, so it promises to launch new Spectrum-X products every year, to provide customers with higher bandwidth, more ports, stronger software capabilities and programmable capabilities, and continuously improve leading AI Ethernet network performance.
To meet the diverse accelerated computing needs of data centers worldwide, NVIDIA also provides computer manufacturers with an MGX reference architecture so that they can build more than 100 system design configurations in a fast and low-cost manner. More than 90 currently published or under development systems from more than 25 partners have used the MGX reference architecture, a significant increase from 14 systems from six partners last year. By adopting MGX, the development cost and time are reduced by 3 / 4 and 2 / 3, respectively.
At the same time, leading companies are rapidly using Blackwell to empower their AI businesses. Representative cases include Chang Memorial Hospital in Taiwan planning to use Blackwell computing platforms to promote biomedical research and accelerate imaging and language applications to improve clinical workflow and enhance patients' feelings; Foxconn is planning to use Grace Blackwell to develop intelligent solutions for AI EV and robot platforms, as well as growing language-based generative AI services to create a more personalized experience for its customers.
As a new computing stack, generative AI is fundamentally changing the way humans and machines work, live their lives, and the way in which information is acquired, and is transforming the traditional industry into an industry that produces digital intelligence. But generative AI is still in the early development, the market is far from mature stage, chip companies only "underlying hardware" is obviously not enough, interconnection, storage, across the interconnection, antipyretic, communication, and then extended to the system level, to build up the ecological level generative AI era competitors the most difficult obstacles.
Copyright © Shenzhen Kinghelm Electronics Co., Ltd. all rights reservedYue ICP Bei No. 17113853
