Full of guests and friends! NVIDIA officially announced that DRIVE Thor will use the latest BlackWell architecture GPU

It’s all about generative AI.

When DRIVE Thor was released in 2022, ChatGPT was not yet popular, and large language models (LLM) were not well known. Now that Transformers, large language models and generative AI have become industry standards, cars have also begun to adopt the route of large models. Therefore, at GTC 2024, NVIDIA announced that NVIDIA DRIVE Thor will use the latest Blackwell architecture GPU.

The Blackwell GPU architecture is equipped with six transformative accelerated computing technologies that will drive breakthroughs in data processing, engineering simulation, electronic design automation, computer-aided drug design, quantum computing and generative AI, which NVIDIA sees as Emerging industry opportunities. The platform enables organizations around the world to build and run real-time generative AI on trillion-parameter large language models (LLMs) at up to 25 times lower cost and energy consumption than previous generation products.

Wu Xinzhou, vice president of NVIDIA's Automotive Division, said: "Accelerated computing has brought transformative breakthroughs to areas such as generative AI, which is redefining autonomous driving and the global transportation industry. DRIVE Orin remains the first choice for today's smart vehicles. AI in-vehicle computing platform, at the same time, we see leading companies in the mobility field having the foresight to include NVIDIA DRIVE Thor in their next-generation AI automotive product roadmaps."

Car companies are supporting

In addition to the previously announced Li Auto and Ji Krypton, DRIVE Thor on GTC has welcomed the participation of multiple car companies.

Many well-known electric vehicle companies, including BYD, GAC Aion, its high-end luxury brand Haopin, and Xpeng, have announced that they will adopt DRIVE Thor.

In addition, Nuro, Plus, Waabi, WeRide, and Lenovo Car Computing have all added support for DRIVE Thor.

Detailed explanation of BlackWell

At GTC, the six transformative technologies of the BlackWell architecture were announced in detail. Blackwell can implement AI training and real-time LLM inference on models with up to 10 trillion parameters.

The world's most powerful chip: Blackwell architecture GPU has 208 billion transistors and is manufactured using a specially customized double photolithography limit size 4NP TSMC process. Through 10 TB/s inter-chip interconnection, the GPU die is connected into a unified GPU.

Second-generation Transformer engine: Thanks to new micro-tensor scaling support and NVIDIA’s advanced dynamic range management algorithms integrated into the NVIDIA TensorRT-LLM and NeMo Megatron frameworks, Blackwell will implement computing with new 4-bit floating-point AI inference capabilities. Forces and model size are doubled.

Fifth-generation NVLink: To improve the performance of trillion-level parameter models and hybrid expert AI models, the latest generation of NVIDIA NVLink provides a breakthrough 1.8TB/s bidirectional throughput for each GPU, ensuring seamless communication between up to 576 GPUs. Seam high-speed communication to meet the needs of today's most complex LLM.

RAS Engine: GPUs based on the Blackwell architecture include a dedicated engine designed to ensure reliability, availability, and serviceability. In addition, the Blackwell architecture adds multiple chip-level capabilities that leverage AI preventive maintenance to run diagnostics and predict reliability-related issues. This will maximize system uptime and increase the resiliency of large-scale AI deployments, allowing them to run uninterrupted for weeks or even months while reducing operating costs.

Secure AI: Advanced confidential computing capabilities protect AI models and customer data without impacting performance, and support new native interface encryption protocols, which is critical for industries such as healthcare and financial services that place high emphasis on privacy issues.

Decompression engine: A dedicated decompression engine supports the latest formats, delivering extremely powerful data analysis and data science performance by accelerating database queries. In the coming years, data processing that costs enterprises tens of billions of dollars each year will increasingly be accelerated by GPUs.