This supercomputer is a game-changer for researchers

A man standing in front of a series of machines that comprise the UF HiPerGator supercomputer.

The University of Florida is home to one of the 25 fastest supercomputers in the world. Erik Deumens is the director of research computing at UF.

Welcome to From Florida, a podcast where you’ll learn how minds are connecting, great ideas are colliding and groundbreaking innovations become a reality because of the University of Florida. 

The University of Florida is home to one of the world’s fastest supercomputers. It’s called HiPerGator and in this episode of From Florida, Erik Deumens explains how its speed and capacity is making a difference in what researchers are able to do. Produced by Nicci Brown, Brooke Adams, Emily Cardinali and James L. Sullivan. Original music by Daniel Townsend, a doctoral candidate in music composition in the College of the Arts.

For more episodes of From Florida, click here.


Nicci Brown: Welcome to From Florida, where you learn how minds are connecting, great ideas are colliding and groundbreaking innovation is becoming a reality because of the University of Florida. I'm your host, Nicci Brown.

The University of Florida is home to one of the 25 fastest supercomputers in the world and today we're going to talk about how UF has differentiated itself among research universities with its investments in computing power.

Our guest today is Erik Deumens, who is the director of research computing at UF and is also a faculty member in UF's Department of Chemistry. Erik received his doctorate in computational nuclear physics from the University of Brussels in Belgium.

He has been at the forefront of numerous advances in his field — electron nuclear dynamics, ENDyne software and the Super Instruction Assembly Language approach to programming massively parallel computers — which I think in simple terms means he knows how to get a lot of computers to work together!

Welcome, Erik, it's a delight to have you here with us today.

Erik Deumens: Thank you very much for having me.

Nicci Brown: So, let's start at the beginning. What is HiPerGator? Can you give us a brief history about the vision that made it a reality on our campus?

Erik Deumens: Well, HiPerGator is now an integral part of the University of Florida and it started in 2011 when the University of Florida hired a new CIO. And he had this vision to build a much more robust, high-performance computing infrastructure. And we built a data center, which is where we now have everything on the east campus.

And we got a supercomputer. Once the data center was ready, we put in a supercomputer. Then we had a contest for the name and finally we found something that really sounded well —  HiPerGator. And it's sort of spelled in a funny way because it stands for High-Performance Gator. At the University of Florida, everything, of course, has to be Gator!

And the HiPerGator was such a good name that when we got a new version of it, that was the first one in 2013, we got a new version in 2015, we couldn't imagine coming up with a name that was as good. So that's why we just called it HiPerGator 1 and HiPerGator 2.0, and now this last year we had HiPerGator 3.0.

And then something extra happened that was even more exciting. The University of Florida under the leadership of the provost and the Board of Trustees wanted to do this big AI initiative. And one of our donors, Chris Malachowsky, who is an alumnus of the University of Florida, was convinced to donate something for a new building — the Chris Malachowsky Data Science building, which is being constructed.

And to celebrate that, he visited the campus one more time. And at that time, he asked the CIO, "I would like to see your data center." So, the CIO and I accompanied him to the data center and at that time, the data center had a big machine room and it was half full. It had HiPerGator 1 in it and HiPerGator 2.0.

And when he saw that he said, "Oh, you guys have room for a supercomputer." And somehow, he must have been thinking, although he's excited about donating to the University of Florida for a building, being a computer person, as one of the founders of NVIDIA, of course, he was very excited about computers.

And he had been thinking probably about giving the University of Florida a computer, but he didn't know that we would be sort of ready to receive it. Because if somebody gives you a $50-million machine that requires 1.6 megawatts of electricity to run, you have to have the staff and the power and the room to do it.

It's not just something like, "Oh, here it is. Here are the keys" and off you go. So that was the beginning of this massive activity in 2020. And the goal was to have everything ready by January of 2021, which we did.

So now this last year, we have been using this new machine for teaching and all kinds of interesting things. So that's the quick story of HiPerGator. And now the latest version is HiPerGator 3.0 and HiPerGator AI. HiPerGator AI is the name of the machine that Chris Malachowsky and NVIDIA donated. It was a partnership donation between Chris Malachowsky and half from the company.

Nicci Brown: Terrific. So just so that people are aware, when we say how HiPerGator is spelled, it's capital H-I, capital P-E-R, and then of course, Gator. Erik, can you help us understand what we mean when we say supercomputer? How does that relate to what we know?

Erik Deumens: A supercomputer is really a computer of supersize. Just compare it to moving gasoline. In your car, you have a gas tank. So, you're moving gas around to drive, it's like your laptop or your cell phone. And then to bring gas to gas tanks, you have the tanker trucks. But then in order to bring gasoline from one country to another, we use oil tankers and they basically have the capacity to move massive amounts of oil in one transition, whereas if you wanted to do that with little boats or trucks or train tanks, then it would take a lot longer and it would be more complicated. So, a supercomputer is some computer that has the capability to process massive amounts of data in a reasonably short time.

Nicci Brown: So again, it sounds like it's the capacity we have, but also the speed which is such a key factor here.

Erik Deumens: Yes, that is very important. So, the supercomputer is sized so that it can handle massive amounts of data and bring a result in a short amount of time. If you wanted to do this with regular computers or by human effort, to process that same amount of data would take so long that it's no longer useful. Just think of predicting the weather. People could look at and do all of  the calculations by hand and then they would be able to say, "Okay, here's the weather report for... Oh, well, actually, that's now a year ago because it took us a year to do the calculation." That is not useful.

Nicci Brown: Can you take us behind the curtain a little bit and tell us what actually the staff does when they're working with HiPerGator?

Erik Deumens: Yeah. So, the research computing team at the University of Florida consists of two types of people. We have one group and they're called the system-facing people. And they're the ones who care and feed the machine, make sure that everything is working, that it doesn't overheat, that the software is working. And then there's the researcher-facing people. And they are the people who help the researchers get their problem done.

And we do all kinds of tasks from answering simple questions, "Well, I'm trying to find this file and I cannot find it" or "I run this software and it immediately dies." And then we give them a training session or we have some documentation all the way from, "Well, I have this question about how I've been trying to find a particular pattern in this data. How do I go about it?"

And then we provide something that's more like a consulting service where we help them define the question, what is available software that they might possibly use and then guide them through the steps of setting it up, provide some training and let them be successful in answering their scientific question.

Nicci Brown: So earlier this year, HiPerGator was named one of the most powerful supercomputers in the world, No. 2 in the U.S. and No. 3 worldwide for higher education. Who gave this recognition? Who are the people who can say this is what it is?

Erik Deumens: Well, there is something called the Top 500 list of supercomputers and it was created in 1989 or something. And it's actually associated with the International Conference of Supercomputing that has been held also since that time. So, the way this works is there is a particular computational benchmark that you have to run. And then that gives you a particular speed. It's sort of like running the marathon and the one who does it the fastest, that's how you rank.

So the Top 500 computers are listed like that. And Supercomputing ’21 was in June, and that's where HiPerGator was listed as No. 22 of the fastest supercomputers in the world. But it includes everything. It includes industry computers, education, research labs. So that's where we then look and say, well, if you look at only universities, then you get to these numbers that you just mentioned.

Nicci Brown: So, I think it's fair to say that having a computer with this level of power is a game-changer for a research university. Can you explain in everyday terms just how powerful HiPerGator is? How does it compare to, say a regular desktop or laptop computer?

Erik Deumens: Well, I can make you a small sort of calculation of how powerful this computer is. If you look at a full length, high-definition movie, two hours, it takes about four gigabytes to store it. And then you take 2,800 of those high-definition movies and you ask it an AI relevant question, like, "Well, who are all the actors in all these movies and when do they appear, and how many roles and how many lines do they have?"

Well, then HiPerGator AI can watch all those 2,800 movies and at the end give you a report and do all of that in one second. Now, if a person wants to watch all these movies, then it would take about 700 eight-hour working days. And I know the millennials now like to watch videos at twice the speed, so they can do it in 350 eight-hour working days, but it's still a whole year compared to HiPerGator AI doing it in one second.

Nicci Brown: It's mind-blowing when you really think about it and put it that way. So, what are some of the research projects that are making use of HiPerGator's power and how is HiPerGator impacting that work?

Erik Deumens: Our vice president for research, Dr. Norton, made a Catalyst Award announcement in the summer of '20. Lots of people applied with ideas of how they could use and how AI could enhance and HiPerGator could help their particular research project. And we had 95 awards that are now working on HiPerGator AI to do their stuff. And it involves graduate students, collaborators all over the world and all led by faculty at the University of Florida.

Some examples are precision medicine, like in dentistry. One of the things that people are interested in is if you want to put an implant, it's very important to very accurately place the hole where you're going to drill that screw. And if that is too close to this little channel in the jaw where the nerve is, then you're going to have constant toothache. So, it's very important. And now this is done by looking at X-rays with people. So, one of the projects would be, “Can we use AI and its image recognition to make that more precise” so that we can make sure that people don't end up with a permanent toothache?

Then there are other things, in agriculture, you basically are trying to find canker. Canker is a big industrial impact to the state of Florida. It's obvious if you take satellite pictures or drone pictures or air pictures, if half of your crop is gone and all the trees are turning a brown color or something, then of course, yeah, it's bad, but then it's usually too late. What you want is you want to see that it's happening in a small location. And that's, again, where AI can look at these thousands of images that come from the satellites and you take drone pictures and you mix them all together just trying to get more and more information — something that humans could not do in a timely manner — and then give a report to the farmers. So those are just two examples and then there's everything in between from engineering and so it's way too long. I could talk for hours about it.

Nicci Brown: Well, it is incredible when you think about just how much AI is integrated across our world and can be integrated across so many different areas. And we know that the University of Florida is engaged in trying to do this across the curriculum so that every student, no matter what their field of study, has some kind of exposure to this technology and how it's transforming our world. So as part of this initiative, we've been hiring a lot of faculty, I think 100 AI faculty, and more than half are on board already, I believe. So, I would imagine the opportunity to be able to engage and use HiPerGator is a huge draw for these faculty members as well.

Erik Deumens: Actually, yes, it is and also it is very impressive. As a matter of fact, this afternoon, I am going to give a tour to a faculty candidate and the dean of the College of Dentistry to show them HiPerGator. And in the past months when we were hiring all these people, I've given many tours to show what our capability is. And as you said, it is a big attraction. There are lots of people who come here because they read about it. And then when they see the infrastructure, it's like, "Well, this is where I can make my research dreams come true because the infrastructure's there, the tools are there." Yes, it's very exciting.

Nicci Brown: And UF is also providing researchers at other institutions with access to HiPerGator. And these include institutions in the Inclusive Engineering Consortium, the entire state university system.

Erik Deumens: Yes, that is correct. So most, if not all, of the universities in the state of Florida are already working with some of their faculty on HiPerGator.

Nicci Brown: Are there other reasons why we're engaging in these partnerships?

Erik Deumens: Well, it's basically the vision of our provost and the Board of Trustees and the governor of making sure that we are creating and training the next generation of citizens that are aware of AI is important.

So, it's mostly focused on teaching and then enabling research, but it's highly focused on making sure that everybody in the country can learn about what is AI, what can it do. Even if you are not going to be a computer science person looking at the nitty-gritty details of AI, you still need to know what are the dangers of AI, how does it work.

Nicci Brown: So how would that benefit Florida? What are some examples that an everyday person living at home in Florida, what are the tangibles that they're going to get out of this?

Erik Deumens: One of the things is that we want to be an attractive place for people to come. And if somebody has an idea … everybody is convinced that this whole AI will generate a lot of startup companies, the University of Florida is very good at nurturing startup companies. So, if people have an idea, then we can help make it into something that turns into a product or a service. And with HiPerGator AI here, people will want to come here and interact with other people who are doing the same thing. So, it'll be good for the economy. It'll create jobs and put the name and fame of the state of Florida out there everywhere.

Nicci Brown: Well, that sounds like a wonderful note to end on. Thank you so much for being our guest today. It's been terrific to have you here.

Erik Deumens: Thank you very much for having me.

Nicci Brown: Listeners, thank you for joining us for an episode of From Florida. I'm your host, Nicci Brown. And I hope you'll return for our next story of innovation from Florida.

UF News January 4, 2022