News on June 28th reveals that, during the developer conference of US data platform, Snowflake, NVIDIA founder Jensen Huang started a discussion about "how to bring generative AI to enterprise users". In the discussion, Huang shared his views.
Huang insinuated that a fundamental change is occurring in computing platforms, the first of its kind in six decades. He warned that acquiring a large number of CPUs in the coming year will not result in increased computational throughput. He advocated for an acceleration in computation, suggesting the use of NVIDIA's accelerated computing platform.
In Huang's perspective, concepts such as Central Processing Units, IO subsystems, DMA controllers, virtual memory, and multitasking can be traced back to the 1960s. Over past six decades, these concepts have aided users in continuously extending CPU capabilities. However, Huang believes that we have now reached a point where these strategies have become redundant.
Highlighting the shift in trends, Huang emphatically stated that computation has encountered a fundamental shift. He strongly suggested that CPU acquisition strategies have become obsolete.
Adding to his previous views, Huang mentioned that although NVIDIA GPUs come with a higher price tag, the direct use of these models essentially results in a price reduction of 95%. He said, "If you go to any cloud, you will see that NVIDIA GPU is the most expensive computing entity there. However, you will find that the speed at which we deliver our services is as if you are getting a 95% discount."
