NVIDIA is set to make waves at the upcoming Hot Chips 2024 event, where they will discuss their groundbreaking use of AI in designing next-generation chips.
The company will delve into the intricate process of building some of the most complex products on the planet, specifically focusing on their new Blackwell B200 AI GPU. This marvel of engineering boasts an impressive 208 billion transistors, fabricated using TSMC’s advanced 4NP process node.
A key highlight of NVIDIA’s presentation will be the exploration of generative AI’s role in creating optimized Verilog code. Verilog is a specialized language used to describe circuits in code form, essential for designing and verifying processors.
NVIDIA’s Agentic AI LLM application takes center stage in this process. This application acts as a powerful tool for accelerating Computer Aided Engineering (CAE) by generating Verilog code. This generated code brings numerous advantages:
Inside NVIDIA’s Blackwell B200 AI GPU, the 208 billion transistors (resulting from merging two reticle-limited GPUs) are packed onto the TSMC 4NP process node. This chip also features 192GB of HBM3E memory with an impressive 8TB/sec of memory bandwidth.
Furthermore, the Blackwell B200 boasts a remarkable 1.8TB/sec bi-directional NVLink bandwidth, facilitated by the high-speed NVLink-C2C Link connecting it to the Grace CPU.
NVIDIA’s demonstration at Hot Chips 2024 is poised to be a significant event in the world of AI and chip design, showcasing how the latest advancements in AI are being harnessed to create even more powerful and efficient computing technologies.
