At GTC 2020 held on December 15 (online, of course), NVIDIA Chief Scientist Bill Dally gave a keynote address focusing on the progress of his AI research team, where he special emphasis name of the founder and CEO of NVIDIA, Jen-Hsun Huang, and asserts that now what prevails is Huang’s law, but what is this law?
Huang’s Law describes the development of AI in GPUs
NVIDIA Chief Scientist Bill Dally currently holds more than 120 patents, and it is not for nothing that he was Head of the Computer Science Department at Stanford University before joining the NVIDIA team in the ‘year. 2009. He is currently responsible for research in the areas of AI, ray tracing and broadband interconnection.
At the aforementioned conference, Dally used three projects to describe how his research team of over 200 people successfully implemented “Huang’s Law,” and he boldly predicted that the GPU will double its artificial intelligence performance every year.
To achieve this, NVIDIA has developed a special tool called Magnet. It is a reasoning accelerator in Artificial Intelligence which can reach a reasoning speed of 100 TOPS / W (100,000 million operations per watt of energy consumed) according to the first simulation tests that were carried out with it, a speed much higher than any AI processor today.
MAGNet adopts a number of new technologies to coordinate and control the flow of information through the device to minimize data transmission, which is the most consuming power link for chips today. In addition, NVIDIA also investigated a type of optical link much faster than it can replace current electrical connections; Using dense wavelength division multiplexing technology, it is expected to achieve data transmissions of more than 1 TB / s with 1 nanometer chips, which would be 10 times faster than what is available with current technology.
Bill Dally also showed an example (which you can see in the image below) of a new system model. NVIDIA DGX which can integrate more than 160 GPUs using this same type of optical links.
Of course, NVIDIA seems very, very optimistic in formulating this “Huang’s Law” that they are talking about, but the truth is that no one can deny that in recent years they have put a lot of effort and money into the implementation. new artificial intelligence technology, They seem determined that this is the future of the business. We will see where everything is.
