Oden Institute Offers Intensive CUDA Programming Course

Following the success of the CUDA programming for NVIDIA GPUs short course offered in 2022, the Oden Institute for Computational Engineering and Sciences at The University of Texas at Austin offered the same course, now in an intensive one week format, to graduate students, postdocs, and all members of the Oden Institute community. 

The course, which took place in March, was led by Mike Giles, who holds a chair in Scientific Computing at the Mathematical Institute at the University of Oxford. Giles is a pioneer in the development of algorithms designed for GPU computing and has made many fundamental contributions to the field. 

In his class, Giles shows students the ropes of NVIDIA's CUDA (Compute Unified Device Architecture) platform to harness the immense computational power of NVIDIA graphics processing units (GPUs) for general-purpose computing tasks. This approach allows programmers to offload parallelizable computations from the CPU to the GPU, significantly accelerating the execution of tasks ranging from scientific simulations to machine learning algorithms. 

The importance of CUDA programming lies in its ability to unlock unprecedented performance gains, enabling researchers and developers to tackle increasingly complex problems with greater efficiency. By leveraging the parallel processing capabilities of GPUs, CUDA programming opens up new avenues for innovation across diverse fields, from scientific research and engineering simulations to artificial intelligence and data analytics.

Drawing from his extensive expertise, Giles led participants through a comprehensive exploration of CUDA programming, empowering them to harness the full potential of NVIDIA GPUs. This intensive week-long course builds upon the resounding success of Giles's similar CUDA workshop at the University of Oxford, which has attracted participants from diverse backgrounds including academia, industry, and government laboratories across the UK.

“I started CUDA programming in early 2007 when CUDA had its initial beta release,” said Giles. “I did a little bit of teaching on CUDA in 2008, and then in summer 2009 I gave the first one-week course in Oxford which is the basis for the current course. Apart from a couple of years during the Covid pandemic, I've taught it every year since then, with help from an Oxford colleague in recent years, and I've also given it in various different countries. In total I have probably trained around 1000 people so far.” 

The course consisted of a mixture of lectures and hands-on practices. The first three days alternated between 90-minute lecture sessions and practical exercises, and the latter two days followed a schedule of morning lectures and afternoon practices, which provided extended time for students to delve into application scenarios. Students were also encouraged to read NVIDIA CUDA C Programming Guide and other reference documentations. 

The course received excellent feedback from Oden Institute students. 

“I thoroughly enjoyed the course offered by Dr. Giles through Oden. I found that it provided insights and practical examples motivated by research and industrial applications,” said CSEM graduate student Raghav Pant. “Most memorably for me, Dr. Giles was extremely willing to answer questions and chat about everything from the details of the course material to his experiences over the years working with CUDA programming in his research. It's always special to get the chance to speak directly with a leader in the field.” 

Vignesh Sella, another course attendee and Oden CSEM graduate student, said a significant factor behind the recent AI boom is the increased computational capability that GPUs provide. "As we learned during the course, this capability not only stems from NVIDIA's hardware but their dedication to software (such as CUDA) over the past two decades. I had very little experience in the world of general purpose GPU computing prior to this course and found this experience to be eye-opening. Prof. Giles excellently motivated each of the lecture topics with real world examples ranging from applications in AI, such as large language models and self-driving vehicles, to quantitative finance within investment banks.

Sella added that the guest talk by the director of the Texas Advanced Computing Center (TACC), Dan Stanzione, was a great addition to the course. "Dan gave us a unique peek into the future of high performance computers (HPCs) in the near future. Spoiler alert: they will likely include a heck of a lot more GPUs!"

Reflecting back on the course, Giles wish for students who have taken this course to not only gain a good understanding of the fundamentals of CUDA programming, but also to develop the confidence to apply it to their research. 

“It is really important to me that the course is a combination of lectures and practicals. Only by doing the practicals are you sure you understand all of the theory, and that practice gives you the confidence to go on to bigger programs,” Giles emphasized. 

As the Oden Institute continues to push the boundaries of scientific discovery, Giles's course serves as a cornerstone in empowering individuals to utilize the full potential of GPU computing. With his guidance, students are not only equipped with technical proficiency but also the hands-on practice to apply this skill to their research.