GAINESVILLE, Fla. — At the University of Florida, inside a brick building with fake windows, lives a supercomputer called HiPerGator, one of the fastest in the world to harness the power of artificial intelligence.
It's been four decades since TIME magazine put a computer on the front cover of their Jan. 3, 1983 issue. The "Man of the Year" honor was replaced by the "Machine of the Year."
We wanted to see how far computer processing power's advanced over the past 40 years. So, we requested a tour of one of the most advanced computers in academia.
HiPerGator is so powerful the possibilities for researchers are endless.
"There is no way that humans can look at the data and draw proper conclusions," Erik Deumens said.
Deumens is the Senior Director for Research Computing at UF. "It's a tool that helps do what humans can do, but much faster."
Deumens gave ABC Action News reporter Michael Paluska and photographer Reed Moeller a data center tour.
Keeping the supercomputer up and running efficiently is as impressive as the computer. Massive diesel generators are ready to run for five days if there is a power outage, and the system has extensive backup battery power.
"And then from the batteries, we convert by a computer-generated perfect electromagnetic current that is perfect AC, alternating current, and that gets fed to the machines. The result is that all the spikes and the brownouts that you get typically come from the feed from the outside because there are lightning strikes, and people are working on it. And they're making disruptions; all of that is washed out," Deumens said. "These computers get their entire life perfect current. And the temperature is always the same. So, as a result, the machines live very long."
A massive cooling system ensures the data center temperature stays between 61 and 62 degrees. The air recirculates through the facility twice a minute, and the hum of the computers keeps the noise level at a consistent 80 decibels. Still, when researchers harness the supercomputer's power, the noise can be louder than a jet engine. During our tour, we wore headphones to communicate. The margin for error is slim.
"If there's no cold water to keep cooling it, the machine room will overheat in four minutes," Deumens said.
The tech specs will leave any computer geek drooling. According to the University of Florida, there are "1,120 A100 GPUs and 2.5 PB all-flash high-performance storage system. There are 70,000 cores total; 30,000 come with 4 GB of RAM per core, and the 40,000 newest come with 8 GB RAM. The Intel and AMD cores of HiPerGator provide a total of about 2 Petaflops computing speeds as measured by the HPL benchmark. The HiPerGator AI system has an HPL rating of over 16 Petaflops and a theoretical AI performance peak of 700 Petaflops."
The computing power was made possible by a$70-million partnership with NVIDIA.
The best part about HiPerGator is all students and researchers across the state university system can access it.
"I think one of the coolest things about it, though, is it doesn't matter what your major is; it can be astronomy, biology, health, medicine or business, or any of them, even the arts. And our students can use AI to solve problems in their discipline," David Reed, Associate Provost for Strategic Initiatives at UF, said. "We're putting a focus on artificial intelligence here at the University of Florida. The one thing I want students to take away from this is you don't have to be a computer programmer or a mathematician to use artificial intelligence in your work. The applications that we use in AI are getting simpler and simpler."
We sat down with researchers who told us how much data the supercomputer processes can turn what would take a human or slower computer 6 to 10 years to compute into 6 to 10 days.
"It's unimaginable that a Ph.D. student or researcher has to wait six years just to train one model; no research will be done at all," Dr. Ruogu Fang, an Associate Professor in the Biomedical Engineering Department, said. "Only with the power of HiPerGator and all those supercomputers with paralleled multi-GPU computation. Only with them can we actually train those models and advance today's AI research."
Dr. Fang's research focuses on brain health and using trillions of data points to hopefully, one day, find a cure or treatment for Alzheimer's, Parkinson's, and other brain diseases.
"Instead of a doctor looking at the brain using their eyes, we can use computers to look at all those different voxels like millions of voxels in the brain," Dr. Fang said.
"When we're shaving off all of these years with research, and we're breaking it down to days, what does that benefit do for humanity?" Paluska asked.
"The major benefit for humanity is that we can train AI models that can help us to absorb the knowledge, the information from millions, or even billions of data, and to identify the important features or key information so that we can apply them to the new patients so that we can make a diagnosis, designs a precision or individualized treatment," Dr. Fang said.
"This research could save lives because it could give answers to people that are suffering from Alzheimer's and Parkinson's disease, maybe a cure faster than they thought they could get it?" Paluska asked.
"I think that's a goal of our research. It is still in the research stage. And that's what we hope for that only with the power of those supercomputers; we can develop models that validate our hypothesis and show that there is a promise there," Dr. Fang said. "Because today's data is not small data, it's huge. And it can be, can be millions in scale and billions in dimension. So as human beings, we cannot process our source data fast enough and even comprehensively enough because we will forget what we get from the previous data set. And I hope using the combined human knowledge with artificial intelligence, computer computation power; we can accelerate this process to find the cure and the intervention methods that can slow down or even entirely cure this disease."
On a farm in Wimauma at the Gulf Coast Research and Education Center, we met with Dr. Kevin Wang.
Wang is an assistant professor for plant breeding, machine learning, computer vision, and robotics. His camera setup is enough to make even the paparazzi jealous.
"We're capturing images of, you know, multi-angle images of the strawberry breeding trials taking six photos every second," Dr. Wang said. "So there's 3000 cultivars, different varieties of strawberry in this field. The more images we capture, the better 3d model we can build from those images. But ultimately, the goal is we'll capture in video clips."
Dr. Wang feeds all of his data into HiPerGator to create his 3-D model of the strawberries.
"Because we are in a big data era, this one will generate about half a terabyte of data," Dr. Wang said. "We do this a collection weekly. So you can imagine, like, we generate, you know, maybe three to three and a half terabytes of data every week. And then we can't just do that manually. So HiPerGator gives us this power."
Researchers tell me the practical applications of HiPerGator are endless, but what does that mean for humans and computers in the future?
"Should people be scared of this technology or embrace it? Because the technology is pretty incredible," Paluska said.
"The technology is pretty incredible. But really, these machines are very stupid. If you let them do what they do without monitoring them, then they're going to cause damage. And then there's reason to be afraid. But if we just pay attention, and don't let it just go, then I think we'll be fine."
"We've seen in movies about robots taking over, Terminator 2 Judgment Day," Paluska said. "We're at a point now where computers are extremely smart. Is this computer smarter than a human being? Or is it just using our brains and technology to calculate what we need faster?" Paluska asked.
"Yeah, actually, computers are not smarter; we're actually using the wrong words," Deumens said. "Because computer computers are just good at doing the manual, or the sort of repetitive tasks that humans take a lot of time to do. So you still have to steer them. Because these AI machines don't know what they're doing. You just give them some information, and they spit out something else. And they have no clue why one thing is related to something else. That's where the human has to come in and say, 'ah, this is useful information. Oh, this is irrelevant. I'm going to ignore that, but that is useful.' And I can draw a conclusion and give some advice."