Customer care
+34 913 67 09 92
Follow us

Nvidia GTC 2024: Key AI and Automotive Advances

The Evolution of Nvidia Drive

The Nvidia GPU Technology Conference (GTC) has been the foremost GPU conference for over a decade, now also becoming a pivotal event for AI hardware-software-deployment. At GTC 2024, although automotive content was less prominent, the importance of the technologies presented was significant, highlighting the announcement of the Blackwell GPU being part of Nvidia’s Drive Thor centralized computer for autonomous vehicles.

Nvidia Drive is Nvidia’s computing platform for developing advanced driver-assistance systems (ADAS) and autonomous vehicles (AVs). Introduced at CES 2015, the platform has evolved through multiple generations, each offering substantial improvements in performance and capabilities.

Nvidia Drive Generations Overview:
  1. Drive CX and Drive PX: Utilized Maxwell microarchitecture, focusing on digital cockpits and ADAS applications, with 256 or 512 CUDA cores.
  2. Drive PX2: Launched in 2016, featuring Pascal GPU architecture and 12 64-bit Arm CPUs, used by Tesla for its Autopilot in battery electric vehicles.
  3. Drive PX Xavier: Introduced in 2017 with Volta microarchitecture, positioned for L3 and L4 vehicles.
  4. Drive PX Pegasus: Based on Turing architecture, Nvidia’s first automotive product with AI functionality, providing a 10× performance increase over PX2.
  5. Drive AGX Orin: Introduced in 2019 and based on Ampere architecture from 2020, continues to be used for ADAS and L3-L4 vehicles, with 2,048 CUDA cores and 17 billion transistors.

Development of Drive Thor

Drive Thor, based on the Blackwell GPU architecture and Arm Neoverse V3 CPU, represents a significant advancement over Drive Orin. With 12× more transistors and 60× higher performance, Drive Thor uses 4-bit floating-point (FP4) calculations, greatly accelerating AI models and reducing power consumption. Additionally, it incorporates a reliability, availability, and serviceability (RAS) engine to identify potential faults and minimize downtime, providing detailed diagnostic data for maintenance planning.

The AI Transformer model, crucial for large language models (LLM), can be leveraged by Drive Thor to address automotive challenges. Blackwell’s Decompression Engine accesses large memory amounts in the Nvidia Grace CPU, accelerating database queries essential for AI LLM platforms.

Drive Thor Customers

Known Drive Thor customers as of April 2024 include various automotive companies and AV startups. More customers are expected as enhanced vehicles are launched. Nvidia’s Inference Microservices (NIM) facilitates GPU-centric software availability, attracting developers and boosting AI applications in the automotive industry.

Nvidia Omniverse and Its Impact on Automotive Development

Nvidia’s Omniverse platform, integrating Universal Scene Description (USD) technology, creates virtual 3D worlds. Announced at GTC 2024, Omniverse Cloud APIs extend its reach, becoming a leading platform for industrial digital twin applications. Digital twins are crucial for automotive development, enabling virtual creation, testing, and simulation of products.