Everything announced at the Nvidia CES event in 12 minutes – video

At CES 2025, Nvidia CEO Jensen Huang kicked off CES, the world’s largest consumer electronics show, with a new RTX gaming chip, updates to its AI Grace Blackwell chip and its future plans to delve deeper into robotics and self-driving cars.

Here is. Our new GForce RTX 50 series, Blackwell architecture, the GPU is just a beast, 92 billion transistors, 4000 peaks, 4 petaflops of AI, 3x higher than last generation Ada, and we need all of it to generate those pixels we’ve I showed you. 380 teraflops of ray tracing so we can calculate the pixels we have to count, calculate the most beautiful image possible and of course 125 teraflops of shading. There is actually a teraflop of concurrent shaders as well as a scalar unit of equal performance. So there are two double shaders, one is for the float, and 0.1 is for the integer. Micron’s G7 memory is 1.8TB/s, twice the performance of our last generation, and we now have the ability to mix AI workloads with PC graphics workloads. One of the amazing things about this generation is that programmable shaders are now also capable of processing neural networks. So shaders are able to carry these neural networks and as a result we invented it. Neural texture compression and neural texture shaders with Blackwell RTX 5070,4090 family performance at 5:49. Impossible without AI, impossible without the 4 vertices, 4 ruptures of the AI ​​tensor cores. Impossible without G7 memories. Well, the performance of the 5070, the 4090, the $549, and that’s the whole family from the $5070 all the way up to the $5090, which is twice the performance of the 4090. First of course, we’re producing a very wide range of availability starting in January. Well, it’s unbelievable, but we were able to put these gigantic performance GPUs into a laptop. This is a 5070 laptop at 1299. This 5070 laptop has the performance of a 4090. So the 5090, the 5090. It’s going to fit into a laptop, a thin laptop. The last laptop’s size was 14.4.9mm. I’ve had the 5080, 5070 TI and 5070. But basically what we have here is 72 Blackwell GPUs or 144 GPUs. This single chip here has a capacity of 1.4 exaflops. The largest supercomputer in the world, and the fastest supercomputer, appeared only recently. This entire room supercomputer recently achieved exaflop plus. This equates to 1.4 exaflops of AI floating point performance. It has 14TB of memory, but here’s the amazing thing is that the memory bandwidth is 1.2 petabytes per second. That’s basically, basically the whole thing. Internet traffic happening now. The entire world’s Internet traffic is processed through these chips, okay? We have a total of 10,130 trillion transistors, and 2,592 CPU cores. A whole bunch of networks and so I wish I could do that. I don’t think I will, those are Blackwells. This is our ConnectX. Network chips, this is the MV connection and we’re trying to pretend to be the backbone of the MV connection, but that’s not possible, okay. And that’s all HBM memories, 1214 terabytes of HBM memory. This is what we are trying to do and this is the miracle, this is the miracle of the black wall system so we fine-tuned it using our experience and capabilities and turned it into the Llama Nemotron range of open models. There are small objects that react in a very, very fast response time, very small uh, they’re what we call Super Llama Super Nemotron, and they’re basically the mainstream versions of your models or your super model, and the super model can be used uh be a teacher model for a whole bunch of other models. It can be an evaluator of the reward model. Uh, a referee for other models to create answers and decide whether it’s a good answer or not, and basically give feedback to other models. It can be distilled in many different ways, basically the guru model, the distillation of knowledge, uh, uh, model, very large, very capable, and all of this is now available online and across Cosmos, the first of its kind in the world. Global foundation model. It is trained on 20 million hours of video. 20 million hours of video focused on physically dynamic things, so dynamic nature, nature subjects, uh, humans, uh, walking, uh, moving hands, uh, manipulating things, uh, you know, things, uh, fast camera movements. It’s really about teaching AI, not about creating creative content, but teaching AI to understand the physical world and through that physical AI. There are several downstream things we can do as a result of this, where we can generate synthetic data to train models. We can effectively extract and transform them to see the beginnings of a robotics model. You can have it generate several realistic and physically plausible scenarios for the future, essentially doing Doctor Strange. Well, you can, because since this model understands the physical world, of course you’ve seen a whole bunch of images generated by this model that understands the physical world, and it can also, of course, do captions and so it can take videos, caption it nicely Extremely, this commentary and video can be used for training. Large linguistic models. Large multimodal language models, so you can use this technology to use this basic model to train bots as well as large language models, and that’s the Nvidia world. The platform has an automatic waterfall model for real-time applications as a publishing model to generate high-quality images. It’s an incredible token that basically learns real-world vocabulary and a data pipeline, so if you want to take all of that and then train it on your own data, this data pipeline because there’s a lot of data involved, we’ve sped everything up at the end to finish it for you. This is the world’s first data processing pipeline that can be accelerated as well as AI accelerated, all part of the Cosmos platform and we’re announcing today. That the universe is open and licensed. It is open and available on GitHub. Well, today we’re announcing that the next generation car processor, the next generation car computer is called Thor. I have one here. Wait a minute. Well, that’s bull. This is Thor, this is a robot computer. This is an automated computer that takes sensors and a huge amount of sensor information, and processes it, you know. There are countless cameras, high-resolution radars, LIDARs, all going into this chip, and this chip has to process all of this sensor, convert it into codes, put it into a transducer, and predict the next path. This AV computer is now in full production. Bull 20 times. The processing power of Orin’s latest generation, which is truly the standard for today’s autonomous vehicles. This is really unbelievable. Thor is in full production. By the way, this robotic manipulator also turns into a full robot and then it could be an AMR, it could be a human or a robot, uh, it could be the brain, or it could be, uh, the manipulator, uh, this this processor is basically a computer Global robotic. Instant Chat GPT. As for general robotics, it is just around the corner. And in fact, all the enabling technologies that I was talking about are. We will make it possible in the next few years to see very rapid breakthroughs, surprising breakthroughs in the field of robotics in general. Now the reason generic robots are important is because robots with tracks and wheels require special environments to accommodate them. There are 3 robots. 3 robots in the world that we can make that do not need green fields. Adaptation to the brown field is perfect. If we can build these amazing robots, we can deploy them in the world we have built for ourselves. These three robots are agent robots and agent AI because you know they are information workers as long as they can accommodate the computers in our offices, it will be great. No. 2, self-driving cars, and the reason for that is because we’ve spent over 100 years building roads and cities. Then number 3, humans or robots. If we had the technology to solve these three problems, this would be the largest technology industry the world has ever seen. This is Nvidia’s latest artificial intelligence supercomputer. It’s finally called Project Digits now, and if you have a good name for it, drop us a line. Ah, this is the amazing thing, this is an artificial intelligence supercomputer. It runs the entire Nvidia AI stack. All Nvidia software works on this. DGX Cloud works on this. This exists, well, somewhere and it’s wireless or you know connected to your computer, it’s even a workstation if you want to and you can access it, you can access it like a cloud supercomputer and Nvidia’s AI runs on it and it’s built on a secret chip. The very thing we’ve been working on is called the GB 110, which is the smallest Grace Blackwell chip we’ve ever made, and that’s the chip inside of it. This is in production. This top-secret chip, which we made in collaboration with the CPU, the gray CPU, was, uh, designed for Nvidia in collaboration with MediaTech. Ah, they’re the world’s leading SOC company, and they worked with us to build this CPU, this CPU core, connect it chip to chip and link to the Blackwell GPU, and this little thing here in full production. Uh, we expect this computer to be available around the May time frame.