The GPU began as a graphics accelerator, designed to recover the triangle and shadow pixels. Then they trained deep nets. Now they run large models behind climate research, protein design, material discovery and automatic lab work. What happens when these engines start planning, reasoning and working in the real world? Hyperscale and HPC’s vice -president Ian Book has been a close view of very few people about this arc.
With CUDA, NVIDIA turned GPU into a software platform, which enables parallel programming that mostly supports deep learning and HPC workload. Under the leadership of the book, NVIDIA’s focus has moved from chips to the full system that considers the data center as the AI factory. Blackwell platform, rack scale nvlink design, liquid cooling, and high -speed networking are all parts of the same push: make AI training and estimation fast, cheap and easy to operate.
It reaches science, health care and industry. The book points to the agent and physical AI for health care, logistics, and industrial workflows, and the matter of science domains where the matter of speed and sincerity: digital biology pipelines, Earth 2 climate imitation, and robotics tools for independent experiments. We talked about a milestone with the book that changed his theory as to what GPU could do, change the system level design, and the next difficult problems. He said it is:
First, congratulate you on your choice as an Avary person of 2025. As the head of the CUDA architect and now NVIDIA’s fast computing business unit, you have seen the GPUs going to Generative AI in learning deeply from graphics. Which milestone has a new look to your own ideology about what GPUS AI can do, and where do you expect the next big shift to come from?
Relevant
