An AI Class That is Exclusive and Needed Now More Than Ever

A famous movie quote included the phrase, “…what I do have are a very particular set of skills.” The same can be said by Associate Professor Andrew Sohn in Ying Wu College of Computing’s (YWCC) department of computer science, whose course, CS 485: Intro to GPU Cluster Programming, may well be – technically - a class like no other in the U.S. due to his work with NASA, Lawrence Berkeley National Laboratory, as well as his venture capital-backed startup company in Silicon Valley.

According to Sohn, there are some schools teaching GPU (graphics processing unit) courses, which involve programming a single machine with Nvidia GPU(s), but not programming a cluster of many machines with GPUs to solve a single problem, called MPI (message passing interface) +CUDA (compute unified device architecture) programming. Only Stanford University seems to come close, offering an introduction to parallel computing using MPI, OpenMP, and CUDA. But he is not sure if the course teaches MPI and CUDA separately or together. Even then, Stanford’s is a graduate level course. Sohn’s is offered to YWCC students during their senior year as well as graduate students.

“I have a unique set of skills,” he affirmed.  Many of these came from his work programming “hundreds of machines” in MPI to solve real-world problems as a NASA as well as an NSF Faculty Fellow. Add instruction in standard GPU programming, and the combination transpires to deliver a course in which NJIT students are clamoring to get a seat.

“It’s two skills, and from what I’ve researched on the Web, nobody else seems to be teaching this in one class,” he continued. The reason is simple: it is difficult to duplicate the process, given that he doesn’t teach theory, but programs in real-time in class. “If I don’t program myself, I don’t teach” asserts Sohn who received the coveted Van Houten Award for Teaching Excellence by the NJIT Alumni Association in 2018.

GPUs were originally designed for gaming, but during the 90’s and early 2000’s, experimentation on building AI models at the time using GPU powered AI resulted in processing neural networks, in particular, convolutional neural networks, for specific tasks that were 1,000 x faster.

For context, Large Language Models (LLMs), such as ChatGPT, are trained using billions of Nvidia GPU cores, which is why the company was ranked #1 in the world in early December 2024 ahead of Microsoft, Apple and Google. Nvidia chips are the foundation of AI, and Facebook and Microsoft have announced multi-billion-dollar investments in Nvidia GPUs for training their GenAI LLM models.

This was why Elon Musk, ranked #1 on the list of 100 most powerful people in the December 2024 – January 2025 issue of Fortune, took #2 ranked Jensen Huang, CEO and founder of Nvidia, out to dinner. People go to Musk, not the other way around.

GPU cluster programming programs Nvidia chips, and with a chip shortage, this “particular set of skills” is in high demand more than ever before. One need not wonder why Sohn’s class is as well.

However, to be successful in the course, students must have strong system skills and be able to freely manipulate Linux boxes. The course is not for the faint of heart as the old adage goes.

CS 485: Intro to GPU Cluster Programming has been offered for the third time as a special topics class during the spring 2025 semester. With the demand for chips increasing, along with the interest from students, this course has the potential to grow along with the technology it is helping to transform - and YWCC, under the guidance of Andrew Sohn, an author of two upcoming books on the Linux kernel, is uniquely poised to offer it exclusively at NJIT.

The course also becomes ever more important for students in computing amidst the industry race for global dominance. The introduction of DeepSeek cost Nvidia $589 billion in market capitalization, the largest loss on a single day in the history of the stock market (Monday, 1/27/2025). This once-in-decades, not century, event only reaffirms that more use of GPUs is central to the race for AI in general, and artificial general intelligence in particular, as DeepSeek demonstrated that many organizations can also enter the race of AI. It's no longer the sole realm for industry giants such as Microsoft, Google, Facebook, Amazon, Apple, etc.