Doing a keynote before the Computex starts and during the evening at Taiwan's top university stadium is something nobody had ever tried before. However, Nvidia CEO Jensen Huang is determined to flex his muscles to steal the spotlight from competitors by announcing a new platform and AI PC wins ahead of the big event.
Despite the heavy rain, National Taiwan University's stadium was fully packed with thousands of enthusiastic fans taking selfies and posting photos on social media. Not only were all the Nvidia supply chain partners such as Wistron chairman Simon Lin, Quanta chairman Barry Lam, Inventec chairman Sam Yeh, Foxconn chairman Young Liu, and SuperMicro CEO Charles Liang in the stadium, Cher Wang of HTC, Daniel Tsai of Taiwan Mobile under the Fubon Group, and Minister of Economic Affairs J.W. Kuo also attended the event.
Revealing the plan to call its next-generation platform "Rubin" for the first time and saying new products will be launched at a one-year rhythm, Huang signaled the relentlessness of the AI arms race, and he already pulled it off with all the supply chain partners in Taiwan behind him to back up the capacity. Clearly, Huang wants his competitors to see how Nvidia is winning in terms of speed, influence, and applications.
"We are in a new Industrial Revolution, and the US$3 trillion IT industry is about to create something that can directly serve US$100 trillion of industry, no longer just an instrument for information storage or data processing. But at a factory to generate intelligence for every industry." Huang said, "What started with accelerated computing led to generative AI, and now an industrial revolution."
Jensen Huang revealed for the first time that the Blackwell Ultra requires eight HBM3e, and the Rubin Ultraq requires HBM4. Andrew Lu, a senior semiconductor analyst, commented that if Nvidia accelerated iteration to 3nm, and both GPUs are using TSMC processes, it will indeed prevent AMD from obtaining enough 3nm capacity, no wonder there are rumors that AMD is looking for Samsung's help. AMD has denied the rumors and said it continues cooperating with TSMC on the 3nm process.
Credit: DIGITIMES
In the 98-minute speech, Huang demonstrated how Nvidia is enabling the new Industrial Revolution, yet softly brushed aside competitor's criticism over the "insane" cost of Nvidia equipment.
"By 100 times speeding up the computation, you only increase the power by about a factor of three. And you increase the cost by only about 50%. We do this all the time in the PC industry," explained Huang, admitting computation inflation is not sustainable, but said Nvidia's CUDA software augments a CPU offload and accelerates the compute that a specialized processor can do much, much better, implying using Nvidia products to accelerate computation and lower the power consumption is money well spent.
Besides showing off the Blackwell platform and NVLink technology already introduced at the GTC in March, Huang also presented Nvidia NIM for generative AI inferencing.
By 2026, Nvidia's Spectrum X CX9 1600G SuperNIC will scale up to millions of GPUs. In the keynote, Huang also announced that there are already over 200 models of laptops that are Nvidia RTX AI PCs -- with a speed of 700 TOPS, able to do 7X generative AI tasks at the edge.
Credit: DIGITIMES
Credit: DIGITIMES
Huang also impressed the audience with images of future AI trends that expounded that the next wave of AI is physical AI.
"Everything is going to be robotic. All of the factories will be robotic. The factories will orchestrate robots. And those robots will be building products that are robotic robots interacting with robots, building robotic products," said Huang, showing video clips that robots are all moving by themselves in the Omniverse, where simulations were implemented through digital twins. "Generative Physical AI can learn skills using reinforcement learning from physics feedback in a simulated world. In these simulation environments, robots learn to make decisions by performing actions in a virtual world that obeys the laws of physics."
Although the Nvidia Earth-2 digital twin simulation model for weather predictions and disaster prevention presented in GTC was presented again here in Taipei, this time it was Jensen Huang's AI digital twin speaking in Chinese to narrate.
At the end of the event, a video of appreciation was dedicated to Taiwan, saying that Taiwan is the bedrock of the AI Revolution. "It's Taiwan that helped us realize a vision. Countless partners lifted us every step of this long journey, from accelerated computing, computer graphics, and scientific research to AI. Every chip and every computer described stories of hard work and pursuit of perfection. You are the unsung heroes, the pillars of the world."
AMD chair Lisa Su, Intel CEO Pat Gelsinger, and Qualcomm CEO Cristiano Amon are all coming to give their keynote speeches at Computex Taiwan 2024, from June 3-7 at the Nangang Exhibition Center. Jensen Huang arrived in Taiwan one week early to visit the supply chain partners. He also attracted media attention by inviting tech leaders to local Taiwanese restaurants and getting the 92-year-old TSMC founder Morris Chang to the night market for the first time in Chang's life.
