At the NVIDIA GTC Conference as scheduled, Huang Jenxun released a number of epoch-making products in succession. From the 4nm process process of the super GPU H100 to the 144-core super chip Grace CPU Superchip, from the robot to omniverse, almost everything is at the forefront of the times. The NVIDIA DRIVE system, as a robot in the field of autonomous driving, released the DRIVE Hyperion 9 and drive map at the conference, and DROVE Orin announced that it will be officially released this month.
Wen 丨 wisdom driving network Huang Huadan
In today's world, the expectation of Apple's conference continues to decline, Musk stands on the top of the mountain and is worshiped by everyone, and the only one who can compete with it is Huang Jenxun.
Strictly speaking, in the evening of March 22, for an hour and a half, Huang Renxun released no less than ten new products and solutions with data flicker and technical hard core at the annual white NVIDIA GTC conference.
Simply listing these products can be dizzying.
For example, the super GPU H100 with the new Hopper architecture, based on the 4nm process process, integrates 80 billion transistors, and the number of cores soars to 16896.
For example, the CPU super chip with a total of 144 cores, grace CPU Superchip. Using the latest Arm architecture, it has 1TB/s memory bandwidth, which is higher than Apple's latest M1 Ultra's 800GB/s.
Another example is the new generation of enterprise AI infrastructure DXG H100 and the world's fastest AI supercomputing EOS, as well as Sionna, an AI framework for 6G communication research.
In the driving field that smart driving network is most concerned about, Nvidia released the DRIVE Hyperion 9 and drive map, and announced that the DRIVE Orin will be released this month.
In addition, as one of the world's top technology companies, Huang Jenxun said that NVIDIA is building multiple robot platforms.
In addition to THE DRIVE for autonomous driving, it includes Isaac for manipulation and control systems, Metropolis for autonomous infrastructure, and Holoscan for robotic medical devices.
The workflow of the robotic system is complex, but Jen-Hsun Huang simplifies it into four pillars: collecting and generating truth-value data, creating AI models, simulating using digital twins, and operating robots.
Omniverse, on the other hand, uses digital twins to operate the core of the robotic fleet.
With the development of technology, concepts such as metaversities and robots are becoming a topic that technology companies cannot avoid.
Although intelligent driving is introduced by Huang Ren in a separate section, it is actually part of the robot industry in NVIDIA's product lineup, and even the meta-universe platform Omniverse has become a platform for robot training.
In a word, Huang Renxun made a wish to all of us humans: I have built the infrastructure for you in the robot age.
He officially opened the door to the robotics industry.
For the intelligent driving network, we take intelligent driving as an incision to see what innovationS NVIDIA has brought to the robot industry built on top of autonomous driving technology.
——01——
AI driver NVIDIA DRIVE
NVIDIA DRIVE is a range of products developed by Nvidia specifically for autonomous vehicles, covering everything from cars to data centers. It is a full-stack end-to-end platform and is open to developers, with the option to use the entire platform or a subset of them.
DRIVE Hyperion is an in-vehicle solution that includes sensors, drive AGX for computing, and software tools necessary for powerful autonomous driving and smart cockpit functions.
Also included are NVIDIA DGX for training DNNs for perception and DRIVE Sim for simulation in Omniverse. Drive AV is an autonomous driving application that runs on the in-vehicle Orin computing platform.
In his speech, Jen-Hsun Huang played a video of the latest version of NVIDIA DRIVE's self-driving car.
Visualize what NVIDIA DRIVE can do.
Vehicle routes include highway and urban routes, as well as complex driving scenarios such as congested intersections. The scene seen by the car can be seen in the confidence level rendering view.
NVIDIA DRIVE Chauffeur is an AI-assisted driving platform that can use autonomous driving to reach the address of interest.
In addition to being able to open Drive Pilot and implement basic functions such as navigation, NVIDIA DRIVE Concierge can also identify specific people and answer questions in different modes. Confidence renderings show the vehicle's field of view and allow the driver to know in advance what the autonomous vehicle is going to do next.
In addition, the sensors in drive Sim can also show real-world scenes. AI can find parking spaces and complete parking in real time.
——02——
NVIDIA DRIVE Hyperion 9 is released, and DRIVE Orin is on sale this month
In his keynote speech, Jen-Hsun Huang focused on the release of a new generation of DRIVE Hyperion architecture based on the Atlan computing platform, which will be powered by cars delivered from 2026 onwards.
The DRIVE Hyperion platform features a design that scales across generations, enabling customers to leverage existing assets in future architectures as well.
The next-generation platform will improve the processing performance of sensor data to further enhance safety and expand into the field of fully autonomous driving operation.
The sensor kit for the latest DRIVE Hyperion 9 will contain 14 cameras, 9 radars, 3 lidars and 20 ultrasonic sensors.
Drive Hyperion 9 will process twice as much sensor data as Hyperion 8, further enhancing safety and expanding the field of work on fully autonomous driving.
The DRIVE Hyperion 8 will be installed in Mercedes-Benz from 2024 and to Jaguar Land Rover from 2025 onwards.
DRIVE Hyperion is a platform for mass production of autonomous vehicles.
DRIVE Orin-based AI computing is integrated with a complete sensor suite to accelerate development, testing, and validation on the road to mass production.
With a complete software stack for autonomous driving, as well as driver monitoring and visualization, DRIVE Hyperion is capable of over-the-air updates, adding new features and functionality throughout the vehicle's lifecycle.
Together, DRIVE Hyperion and DRIVE Orin act as the vehicle's nervous system and brain, processing massive amounts of sensor data in real time to safely perceive, plan, and act.
The DRIVE Orin System-on-Chip provides 254 TOPS computing power per second and is the central computer for smart vehicles.
"Electric vehicles have prompted industry to overhaul automotive architectures, and the cars of the future will be highly programmable, evolving from many embedded domain controllers built into highly centralized computing platforms," Huang said. AI and autonomous driving capabilities will be provided in software and will continue to evolve throughout the vehicle's lifecycle. ”
At the GTC keynote, Jen-Hsun Huang proudly announced that BYD will equip its cars that will start production in the first half of 2023 with the DRIVE Orin computing platform.
In addition, new energy vehicle start-ups such as Lucid, Weilai, Ideal Automobile, Xiaopeng, Zhiji Automobile and Feifan Automobile, Jidu, Huaren Express, VinFast, and WM Motor are also developing software-defined vehicles on the DRIVE platform.
These collaborations signal that 20 of the world's leading 30 electric vehicle manufacturers have chosen NVIDIA DRIVE Orin as their preferred AI computing platform.
Huang said drive Orin will go on sale this month.
——03——
Digital twins in the metaverse
NVIDIA DRIVE Map is a multimodal map engine that includes data from cameras, radar, and lidar that can be individually located to each layer of the map. This provides versatility and redundancy for the highest level of security.
Drive Map has two map engines: a truth-value mapping map and a crowdsourced fleet map.
By the end of 2024, NVIDIA expects to map and create digital twins of all major highways in North America, Western Europe and Asia, with a total length of about 500,000 kilometers, Huang said. Millions of passenger cars will continue to expand and update the map.
NVIDIA's plan is to build a digital twin of an Earth-scale autonomous fleet to explore new algorithms and designs and test software before deploying to fleets to maximize safety.
NVIDIA is developing two scenario simulation methods, each recreating the entire world in a different way.
One of them starts with NVIDIA DRIVE Map, a multimodal map engine that creates highly accurate 3D representations of the entire world. Maps are loaded into Omniverse, and buildings, vegetation, and other roadside targets are generated.
Cars reason and locate moving targets, cars, and pedestrians encountered during previous driving, and then put them into digital twins.
Developers can animate or assign AI behavior models to each moving target, and can also apply domain randomization to generate diverse and reasonable challenging scenarios.
The second method uses neurographic AI and Omniverse to convert pre-recorded driving videos into replayable and modifiable driving processes.
The 3D scene is reconstructed first, the system identifies and removes the moving target, and the background is restored. After reconstructing the scene, the developer can change the behavior of existing vehicles or add vehicles that are fully controllable and react realistically in traffic environments.
The regenerated driving process, combined with 3D geometry and physics-based materials, allows developers to re-light the scene, apply physics effects and simulate sensors such as lidar.
This reruns pre-recorded scenes and applies them to closed-loop simulation and testing.
Drive Map and DRIVE Sim contain breakthrough AI outcomes from Nvidia's research efforts, demonstrating the power of Omniverse digital twins in driving the development of self-driving cars.
——04——
Occupy half of the autonomous driving
In addition to automakers, self-driving startups can also develop on DRIVE Hyperion to deliver software-defined vehicles.
During the GTC conference, a number of autonomous driving companies, including Yuanrong Qixing, Wenyuan Zhixing, and Yunji Zhixing, announced that they would be equipped with NVIDIA DRIVE Hyperion solutions.
On March 23, Yunji Zhixing announced that it has reached a deep partnership with NVIDIA, and its latest vehicle-grade mass production (L4 level) autonomous driving computing platform will be equipped with a DRIVE Hyperion solution based on NVIDIA DRIVE Orin SoC (System-on-Chip). Its autonomous driving computing platform will cover the global automatic driving scenario, providing computing power support for vehicles to achieve safe and smooth L4 level automatic driving in complex urban open roads.
In addition, Yo-Run Technology, which takes the skateboard chassis as the core capability, also announced on March 23 that it will use the NVIDIA DRIVE Hyperion AV platform architecture to create its high-performance automotive brain, integrate NVIDIA DRIVE Orin system-on-chip capabilities, and create the industry's first open automotive supercomputing platform with flexible computing power.
Nvidia's official data shows that since April last year, Nvidia's total car order revenue has grown from $8 billion to more than $11 billion over the next six years, covering almost the entire self-driving automotive industry.
From the top electric vehicle manufacturers, to the world-renowned automakers Jaguar Land Rover, Mercedes-Benz and Volvo Cars, to the leading trucking companies Zhijia Technology, Tucson Future and Volvo Autonomous Driving Solutions, as well as advanced self-driving taxi manufacturers AutoX, Didi Chuxing, Xiaoma Zhixing and Zoox.
Just like Musk's pursuit of general artificial intelligence and humanoid robots, NVIDIA not only frequently launches high-performance products that amaze the industry on hardware, but also provides high-performance and high-safety solutions for the industry by combining its self-driving robots with Omniverse.
Nvidia is opening the door to artificial intelligence with a dizzying array of products.
【Pay attention to the video number of intelligent driving, pay attention to the frontier of intelligent driving】