The head of Nvidia announced that the corporation will shift its chip manufacturing focus towards designs better suited for AI, as this represents the future. Huang stated that the company’s fresh hardware will be truly transformative.
The world of artificial intelligence is entering a new era. This declaration was made by Nvidia CEO Jensen Huang during his address at the annual GPU Technology Conference (GTC) in San Jose, California, as reported by The Wall Street Journal.
“This is the destiny of AI. This is where AI is heading. It is engineered specifically for executing inference tasks, for this scope of workload,” Huang remarked during the unveiling of the company’s new apparatus. He further emphasized that a crucial turning point in inference processes has been reached: “That is the secret sauce right there.”
Inference (derived from the English word “inference”) signifies the stage where an AI system deploys its already trained model to examine novel data and subsequently formulate decisions.
For instance, the recognition of faces within a photo uploaded to a social media platform constitutes an instance of inference. Chatbots, demand forecasting systems, and AI-driven vehicle control algorithms operate based upon this identical principle.
In San Jose, Huang introduced Nvidia’s latest flagship offering, which he asserted will cause disruption in the field of inference. This involves the new Vera Rubin server system and LPU (Language Processing Unit) chips from the startup Groq, which are integrated into Nvidia Groq 3 LPX racks.
Nvidia’s conventional graphics processing units (GPUs) were typically not regarded as optimal for inference because they demand substantial power and lack sufficient memory capacity for models to access the massive datasets upon which their training occurred, as explained by the WSJ. Consequently, the new devices, leveraging the Vera Rubin platform and the Groq processor operating in tandem, are projected to boast 500 times the high-speed memory volume of the preceding Hopper generation GPUs, thereby resolving the memory scarcity challenge.
Huang projected that Nvidia anticipates generating $1 trillion in sales from its Blackwell and Rubin model chips by the close of 2027.
The Blackwell chip was unveiled last October as the core component of the compact DGX Spark supercomputer. The Blackwell GPU was paired with a 20-core Grace CPU. This unit is capable of delivering up to 1,000 trillion AI operations per second.
About the Author
