Accelerate everything! Huang Renxun: Blackwell is now in production, launching the next generation AI platform Rubin in 2026, equipped with HBM4

Author: Zhu Xi

Source: Hard AI

On Sunday, June 2nd, NVIDIA CEO Jensen Huang delivered a major speech on the eve of the COMPUTEX Technology Conference. COMPUTEX Taipei International Computer Show is about to be held in Taiwan, China. NVIDIA, as a major beneficiary of the artificial intelligence wave, has become the world's highest market value chip manufacturer, attracting market attention with every move amidst investors' frenzy.

Here's an interesting tidbit: Old Huang strolling through the night market in Taipei, being treated like a star, netizens exclaiming: This is the era of tech CEOs!

Jensen Huang spoke at length on Sunday, and if we were to summarize his latest speech in one sentence, NVIDIA's official website provided the most concise and powerful summary: Accelerate everything.

Huang Renxun stated that NVIDIA plans to upgrade its AI accelerators annually and announced the launch of the Blackwell Ultra chip in 2025, as well as plans to introduce a next-generation platform named Rubin in 2026. NVIDIA also introduced new tools and software models.

NVIDIA sees the rise of generative AI as a new industrial revolution. Huang Renxun said:

Generative AI is reshaping industries, bringing new opportunities for innovation and growth.

Today, we are at the forefront of a major transformation in the computing field. The intersection of artificial intelligence and accelerated computing will redefine the future.

The future of computing is accelerating. With our innovations in artificial intelligence and accelerated computing, we are breaking through possible limits and driving the next wave of technological progress.

He reiterated a theme he raised a year ago at the same event, pointing out that companies without AI capabilities will be left behind.

Regarding Huang Renxun's emphasis on innovation, netizens were more straightforward: Others are not innovating.

In his speech, Huang Renxun stated, "We expect NVIDIA to play a significant role as AI technology transitions to personal computers." Currently, NVIDIA hopes to expand its customer base beyond just a few cloud computing giants. Previously, most of NVIDIA's revenue came from these cloud computing giants. As part of expanding its customer base, Huang Renxun expects more companies from various industries, as well as government agencies, to adopt AI.

Huang Renxun emphasizes NVIDIA's cost reduction capabilities

Huang Renxun said, "We are seeing computational expansion." With the exponential growth in the amount of data that needs to be processed, traditional computing methods cannot keep up, and only through NVIDIA's accelerated computing approach can costs be reduced:

As the expansion speed of central processing units (CPUs) slows down and eventually comes to a halt, we should accelerate the acceleration of every compute-intensive application, and every data center will definitely be accelerated. Accelerating computation is very wise, and this is common knowledge.

NVIDIA achieves acceleration for compute-intensive applications by adding dedicated auxiliary processors to CPUs.

Because these two processors can work in parallel, they are both autonomous and have more independent values. We can accelerate 100 units of time to 1 unit of time, which is incredibly fast.

Huang Renxun emphasizes NVIDIA's cost reduction capability - NVIDIA is reducing the cost of turning data into intelligence. "Accelerating computation is sustainable computation." He explained how the combination of GPUs and CPUs can achieve up to 100 times acceleration while only increasing power consumption by three times, with a performance per watt 25 times higher than using CPUs alone. "The more you buy, the more you save."

Huang Renxun claims that using NVIDIA's technology can save 98% of costs, reduce energy consumption by 97%, and calls this "CEO math," which, although not precise, is correct.

NVIDIA's cost reduction capability is so powerful that fans ask, give me a reason why customers suddenly don't want these chips:

Blackwell Chip Now in Production

Huang Renxun stated that NVIDIA's acceleration platform is in full production, whether through AI personal computers and consumer devices equipped with numerous NVIDIA RTX features, or through enterprises building and deploying AI factories using NVIDIA's full computing platform. NVIDIA's Blackwell chip is now in production.

Regarding the Blackwell chip, Huang Renxun said, "Even at the peak of Moore's Law, the computing power that Blackwell can achieve is incredible. Every time we increase computational intensity, costs decrease. What we are doing is training a GPT-4 model with 200 trillion parameters and 80 trillion tokens with Blackwell's outstanding computing power, reducing energy consumption to 1/350."

In response, netizens expressed that NVIDIA has no strong competitors and can only compete with Moore's Law, how lonely invincibility is:

Annual Rhythm

Huang Renxun introduced the plan for the annual upgrade cycle of chip products, stating that NVIDIA intends to upgrade AI accelerators/AI chips every year. He reiterated the previously mentioned annual rhythm, "**Our company has an annual rhythmOur basic idea is very simple: to build the entire data center scale at the pace of one year, break it down and sell it to you in the form of components, and push everything to the limit of technology."

Huang Renxun announced plans including Blackwell Ultra and subsequent chips:

Blackwell Ultra will be launched in 2025.

The next-generation platform developed for 2026 is called Rubin. The Rubin platform, which is the next generation of the upcoming Blackwell platform, will make its debut.

The upcoming Rubin AI platform will use HBM4, which is the next generation of high-bandwidth memory. Wall Street News website note: High-bandwidth memory is crucial and has become a bottleneck in the production of AI accelerators. The main producer, SK Hynix, is almost sold out by 2025.

Huang Renxun revealed that the Rubin platform will feature a new GPU, a new CPU based on Arm architecture, as well as advanced networking with NVLink 6, CX9 SuperNIC, and X1600 integrated InfiniBand/Ethernet switches. Huang Renxun did not provide more detailed specifications about the upcoming products.

Industry Partners with NVIDIA to Build AI Factory

NVIDIA originally started out by selling desktop gaming cards, a background that has come in handy as computer manufacturers embrace more AI features. Microsoft and its hardware partners are showcasing new laptops with AI-enhanced features at the Computex conference, branded as Copilot+. Most of these upcoming devices are based on a new processor provided by NVIDIA's competitor, Qualcomm, which allows them to last longer on a single charge.

While these devices can handle basic AI functions, adding NVIDIA's graphics cards will significantly boost their performance and bring new features to popular software like games. NVIDIA stated that PC manufacturers like Asus are offering these computers.

NVIDIA, in collaboration with several top PC manufacturers worldwide, has released a system "Array" supported by the NVIDIA Blackwell architecture, equipped with Grace CPUs, NVIDIA networking devices, and infrastructure to support enterprises in building AI factories and data centers, driving the next wave of generative artificial intelligence breakthroughs.

According to Huang Renxun, companies such as Asus, Gigabyte, Supermicro, ASRock Rack, Ingrasys, Inventec, Pegatron, QCT, Wistron, and Wiwynn, are providing cloud computing, empowerment, and edge artificial intelligence systems that will be equipped with NVIDIA's GPUs and networking components. Huang Renxun bluntly stated, "The next industrial revolution has already begun."To help software developers bring more new features to PCs, NVIDIA is also providing tools and pre-trained AI models: they will handle complex tasks, such as deciding whether to process data on the machine itself or send it to a data center via the internet.

New Design of MGX Platform

NVIDIA has also released a new design for server computers based on its chips. NVIDIA's MGX modular reference design platform now supports Blackwell, including the GB200 NVL2 platform designed for large language model inference, retrieval-augmented generation (RAG), and data processing.

Any server system builder can use these reference designs to save development time while ensuring design and performance consistency. Companies like HP and Dell are using NVIDIA's MGX platform to bring products used by enterprises and government agencies to market faster.

Even NVIDIA's competitors AMD and Intel are leveraging MGX, combining their processors with NVIDIA's chips. AMD and Intel also support the MGX platform and plan to introduce their own CPU host processor module designs for the first time.

Next-Generation Network Spectrum-X

Products previously announced by NVIDIA, such as Spectrum X for networks, are now widely available and widely adopted. Jensen Huang announced the annual plan to release Spectrum-X products to meet the demand for high-performance Ethernet networks for artificial intelligence.

NVIDIA's Spectrum-X is the first Ethernet structure built for artificial intelligence AI, with network performance improved by 1.6 times compared to traditional Ethernet structures. It accelerates the processing, analysis, and execution of AI workloads, thereby speeding up the development and deployment of AI solutions. Spectrum-X could scale to millions of GPUs by 2026.

CoreWeave, GMO Internet Group, Lambda, Scaleway, STPX Global, and Yotta are the first batch of AI cloud service providers adopting Spectrum-X, aiming to bring ultimate network performance to their AI infrastructure.

Analysts say that NVIDIA's Spectrum product line is the flagship product of its Ethernet solutions, especially valued by the company's top management. It is predicted that the Spectrum product line could reach a revenue scale of tens of billions of dollars in the current year. NVIDIA follows a business strategy of "InfiniBand first, then Ethernet," prioritizing the promotion of higher-profit InfiniBand switches when launching GPUs, followed by Ethernet products.

NVIDIA's NIM and AI Assistants

NVIDIA's Inference Microservices (NIM, referred to by Jensen Huang as "AI in a box") have also made further appearances. These microservices are a set of middleware and models that can help companies launch AI services faster without worrying about underlying technologiesNVIDIA announced that it will provide free access to NIM products, and companies deploying these services will need to pay NVIDIA for their usage.

With NVIDIA's NIM, 28 million developers worldwide can now easily create generative AI applications.

NIM also enables enterprises to maximize their infrastructure investments. For example, running Meta Llama 3-8B in NIM can generate up to three times more AI tokens on accelerated infrastructure compared to not using NIM. NVIDIA's nearly 200 technology partners, including Cadence, Cloudera, Cohesity, DataStax, NetApp, Scale AI, and Synopsys, are integrating NIM into their platforms to accelerate the deployment of generative AI applications in specific areas such as copilots, code assistants, and digital humanization.

Hugging Face has also started offering NIM, with the first release being Meta Llama 3. Huang Renxun stated, "Today we just released a fully optimized Llama 3 on Hugging Face, where you can try it out. You can even take it with you. So you can run it in the cloud, in any cloud, download it, put it in your own data center, and provide services to your customers."

NVIDIA's RTX AI personal computers, powered by RTX technology, aim to completely change consumer experiences through over 200 RTX AI laptops and more than 500 AI-driven applications and games.

NVIDIA introduces G-Assist and ACE digital human technology NIM software for digital humans:

• NVIDIA announced a demonstration of RTX-driven AI assistant technology called Project G-Assist, showcasing context-aware assistance for PC games and applications.
• The RTX AI toolkit and the newly launched PC-based NIM inference microservices for the NVIDIA ACE digital human platform highlight NVIDIA's commitment to AI accessibility.

NVIDIA has also introduced a small language model (SLM) for RTX-accelerated APIs for Microsoft Windows Copilot Runtime. Microsoft and NVIDIA are collaborating to help developers introduce new generative AI capabilities into their Windows native and web applications through simple APIs. These capabilities include RTX-accelerated SLM, making RAG functionality running on devices part of Windows Copilot Runtime.

For the aforementioned applications, netizens summarized that **the AI software factory NIM is further integrated with CUDA to introduce more digital human cases. Pre-trained AI models are packaged and optimized through CUDA. Working in conjunction with ACE and SLM technologies, realistic digital humans are created, and a series of benchmark cases are introduced, such as games, nurses, customer service representatives, tutors, and more**

NVIDIA Robotics Technology

NVIDIA is leading the $50 trillion industrial digital transformation, with various industries embracing autonomous operations and digital twin technology - these virtual models enhance efficiency and reduce costs. Through its developer program, NVIDIA provides access to NIM to promote AI innovation.

Huang Renxun emphasized the importance of robots and AI in future development:

The next wave of AI is Physical AI, which is real-world AI that understands the laws of physics and can work around us.

Compared to embodied intelligence, Physical AI requires stronger simulation capabilities to complete robot reinforcement training and learning in a virtual world.

Robotics technology is already here. Physical AI is already here. This is not science fiction. This is really, really exciting.

All factories will become robotic factories. These factories will coordinate robots, and robots will manufacture products with robotic capabilities.

Huang Renxun promotes digital twin technology, which can be used in NVIDIA's virtual world called Omniverse:

To demonstrate its potential, Huang Renxun showcased a digital twin of Earth called Earth 2 and how it helps with more complex weather pattern modeling and other intricate tasks.

Huang Renxun pointed out that manufacturers like Foxconn are using these tools to plan and operate factories more efficiently. He demonstrated how Foxconn uses NVIDIA's Omniverse, Isaac, and Metropolis to create digital twins, combining visual AI and robot development tools to enhance robotic facilities.

NVIDIA's Isaac platform provides developers with a powerful toolkit for building AI robots, including Autonomous Mobile Robots (AMRs), industrial robotic arms, and humanoid robots, supported by AI models and supercomputers like Jetson Orin and Thor. Huang Renxun also emphasized Isaac's role in improving factory and warehouse efficiency, with global giants like BYD Electronics, Siemens, Teradyne Robotics, and Intrinsic adopting its advanced AI models and libraries.

NVIDIA's IGX platform is designed specifically for edge computing, enabling data processing locally without relying on centralized cloud servers. The platform aims to meet strict regulatory and security standards, particularly crucial in the healthcare industry where patient safety and data protection are paramount. The IGX platform is suitable for scenarios requiring high reliability and real-time processing, including diagnostic imaging in medical technology, robot surgeries, patient monitoring, as well as automation robots and intelligent manufacturing systems in industrial applicationsCompanies like ADLINK, Advantech, ONYX, etc., are collaborating with NVIDIA to develop and provide solutions based on the IGX platform.

Netizens commented: Huang Renxun unveiled his robot army, no need to panic (for now)!

Conclusion

Throughout the entire speech, some netizens felt that Lao Huang did not say anything new. However, he used the same Bible preaching, with many generative AI videos, music, and speech. In the end, he emotionally connected disciples (X) and partners, calling on everyone to move towards a bright future of AI together.