We're in Taipei to catch NVIDIA's Computex 2024 keynote address. AI would be decades behind where it is without NVIDIA's single-minded focus on accelerating it, and supporting the most fundamental research in AI, to bring it where it is now—in a position to drive a whole new industrial revolution of the kind the Internet did some 30 years ago. While NVIDIA's AI acceleration focuses on hard iron AI GPUs for the data-centers, the company hasn't lost sight of the hundreds of millions of gamers with a GeForce RTX GPU that's capable of accelerating AI faster than any NPU could by far. The company has set sights on the AI PC. It's already been running AI-based applications for gamers for over 5 years now, with DLSS, but has been consistently releasing AI-based apps, such as the most recent ChatRTX. The company is looking to make gamers' lives even more enjoyable with the magic of AI. Join us in this live blog as we unwrap NVIDIA's latest toys for gamers.10:59 UTC: And we're live!

11:09UTC: The show is running a little late.

11:12 UTC: NVIDIA CEO Jensen Huang takes stage at the NTU Sports Center stadium.

11:13 UTC: NVIDIA's relationship with Taiwan is inseparable, all its partners are there.11:14 UTC: AI marks a restart of the computer industry: Jensen.

11:15 UTC: Graphics, simulations, and AI, are at the soul of NVIDIA, brought to life in Omniverse.11:17 UTC: AI is the next chapter in computing.11:20 UTC: Two fundamental things happening: the processor performance scaling slowed, but computing needs rising exponentially. If data keeps scaling and performance doesn't, we'll experience computation inflation (happening now). CPU scaling is never coming back to past trajectories, so we should accelerate everything, and make computing smarter with AI.11:22 UTC: NVIDIA's fusion of CPU and accelerators offer 100x speedups for just 3x power and 1.5x power.11:24 UTC: NVIDIA's acceleration pays for itself when implemented at scale.

11:25 UTC: The more you buy, the more you save.11:26 UTC: AI acceleration's biggest hurdle is rewriting software for the new hardware. This is why NVIDIA invests so much on software.11:28 UTC: TSMC is implementing cuLITHO, accelerated chipdesign.

11:29 UTC: cuOPT is the best route-planning algorithm based on AI.

11:29 UTC: cuQUANTUM emulates quantum computers using CUDA and cloud computing.

11:30 UTC: cuDF is a data-processing accelerator that's widely used in the industry.

11:31 UTC: NVIDIA has 350+ such libraries. Now accelerated Google Colab.11:32 UTC: CUDA has achieved a virtuous cycle.11:33 UTC: NVIDIA cracked the chicken-egg problem of accelerated computing through its hundreds of domain libraries.

11:34 UTC: The emergence of AI was made possible by reducing the cost of computing.

11:37 UTC: Earth 2 digital-twin. One of the world's most ambitious computing project.11:38 UTC: NVIDIA's first contact with AI was in 2012 with CUDA, with Alexnet. This is where NVIDIA decided to understand the foundation of deep-learning and device accelerators for it.11:43 UTC: NVLink and Tensor cores are now 10 years old, and the Mellanox acquisition came together .

11:44 UTC: By 2017, transformers and GPTs came to being.11:45 UTC: November 2022 saw the big moment: ChatGPT.11:46 UTC: AI was all about perception and detection. GPT was the first generative AI.11:49 UTC: We have arrived at a generative AI era. Ai tokens are a new commodity. A token is a fungible bit of generated information.11:51 UTC: AI has changed every single layer of computing.11:53 UTC: The future of content is generated, because it consumes less energy.

11:54 UTC: NIMS: NVIDIA inference microservices. This will be as significant as packaged software became after Microsoft democratized the PC.11:55 UTC: The basic model of NIMS.11:57 UTC: NIMS build AI driven applications.12:00 UTC: Every company in the future will have a large collection of NIMS and assemble applications.12:02 UTC: Digital humans are natural-looking, realistically rendered AI agents.12:03 UTC: Digital humans' applications transcend gaming, NVIDIA Ace is a suite of digital human NIMS.12:06 UTC: Every GeForce RTX GPU has tensor cores, and there are over 100 million of them.12:08 UTC: The next generation of AI needs to be physically based: Jensen.12:11 UTC: Blackwell implements the 2nd generation of transformer acceleration.12:14 UTC: Blackwell AI GPU production board.12:15 UTC: Blackwell makes computation zoom far ahead of Moore's Law.12:16 UTC: With great processing power comes great power savings.12:18 UTC: 17000 Joules/token for Pascal (3 tokens per word). 0.4 Joules/token with Blackwell.12:20 UTC: MGX modular system.12:20 UTC: NVLink makes 72-GPU Blackwell DGX possible, thanks to a new generation switching chip.12:22 UTC: DGX is a whole big GPU. 72 GPUs held together with an NVLINK spine, operates at 10 TB/s.12:25 UTC: Scaleout of DGX sees the benefits of Infiniband to Ethernet.12:27 UTC: Spectrum-X 1600 is a smart switch for connecting millions of GPUs in the datacenters.12:30 UTC: Jensen announces the next generation successor to Blackwell, say hello to Blackwell Ultra.12:38 UTC: Say hello to Rubin, successor to Blackwell.12:38 UTC: This is the NVIDIA Blackwell platform.12:39 UTC: One day, everything that moves will be robotic.12:42 UTC: Omniverse is a robot gym, where robots can learn how to interact with the world.12:47 UTC: Various Taiwan-based ODM giants are building digital twins of their factories.12:53 UTC: The next generation of real world robots will be humanoid robots.13:01 UTC: Thanks all folks.