The giant supercomputer Summit — based at Oak Ridge National Lab’s Leadership Computing Facility (OLCF) —  lives on, at least for one more year supported by a new program, SummitPLUS. Scheduled for shutdown earlier this year, the IBM-built supercomputer has been extraordinarily productive. It was the fastest supercomputer in the world sitting atop the Top500 list in 2018 and 2019, and it remains the fifth fastest based on its November 2023 showing.

Summit ‘s impressive record includes being recognized in two different 2020 Gordon Bell Prize winning efforts (work simulating SARS-CoV-2 virus’ spike protein activity and work modeling and testing of the DeePMD-kit software) and was a Special Gordon Prize finalist in 2019. It was also used by Google in an experiment to demonstrate quantum supremacy, and although that work has engendered controversy, Summit performed as just as promised. This is hardly a full list.

Today, of course, Summit’s OLCF successor, Frontier, the first U.S. exascale computer, has the limelight and bears the Top500 crown.

Now, SummitPLUS is extending the life of Summit, by supporting more than 100 research projects to be run on Summit this year before final shutdown at the end of 2024. Last week, HPCwire had a chance to talk to Bronson Messer, director of science at OLCF, about one of the projects — an effort to better understand the feasibility of quantum computing-HPC system integration and to build a benchmarking suite, Q-Stone, better assess QC-HPC integrations.

But unsurprisingly, Messer didn’t want to neglect SummitPLUS more broadly. Messer’s job is perhaps one of the most exciting and taxing among the big system sites. His charge is to ensure all of OLCF’s computers and scientific instruments are working at full-tilt on worthwhile science projects. Let’s not forget that includes Frontier.

Even as Summit approached sunset, Messer and colleagues thought it could do more. “You know, we’re constantly running a computer in production and looking ahead to the next computer; we have a finite amount of space as everybody does. So it was very much our plan to shut the computer down. But it’s been such a productive scientific instrument that we negotiated with IBM and Nvidia and got a sixth year of support for the machine,” said Messer.

“That took some doing and because it took a considerable amount of effort, we got to the end of the calendar year or close to the end of the calendar year and realized, oh, my gosh, we need to figure out what we’re going to do with the machine for the coming year because we have not included in any of our allocation programs for 2024. So we basically stood up an allocation program and reviewed close to 200 proposals in about four weeks. Yeah, which is the most ridiculously crazy thing that I’ve been involved with so far.”

The lives of these colossal machines are both amazing and short, in the 5-6 year range generally.

Talking about Summit, Messer said, “We have utilization rates well in excess of 80%. So we’ve been running those GPUs full out for five and a half years now. Asking a lot of them. They’re built to last for a while, but they don’t last forever. So it’s a perishable resource to some extent.

“It’s all about maintenance. It’s all about making sure that if changes go into libraries that we get reasonable updates in a reasonable amount of time. There’s hardware failing every day on the machine, right, and we run it hard. That happens on many machines. But you can tell that Summit has been working hard for a long, long time.”

He doesn’t think yet another year is in the cards for Summit and said, “I cannot say for sure, but I think it would be a minor miracle if we extended operations.”

In the meantime, SummitPLUS is a big win for many researchers. “We’ve got people jumping on the machine and using it like gangbusters. You know, the other thing we did with Summit that I didn’t mention is we didn’t allocate all the time available through SummitPLUS. We are also a resource provider for NAIRR, the National Artificial Intelligence Research Resource, and easily the biggest right now, during the pilot; we’re offering more time on Summit than any other resource provider currently. So I think we’ll have projects from that program appearing on the machine in the very near future.”

Messer said they don’t normally release the details of all allocations on the machine, but said “I can tell you some stats:

1) “The quantum project we discussed is actually the smallest on the allocation list with 20k node-hours. That doesn’t mean it’s the least important! It’s just ‘right-sized’ for the work to be done at this point. All SummitPLUS projects enjoy the same priority on the machine.

2) “The largest projects have allocations a little more than 10x this size, with the largest allocation going to FUS155 from Noah Reddell from ZAP Energy (300k node-hours).

3) “The average allocation size is about 175k node-hours.”

He did provide the list of projects on SummitPLUS and that list is included below.

It’s notable to recall OLCF’s ground-breaking history. Remember Titan, Summit’s predecessor, which was decommissioned in 2019. Titan was one of the first large scale heterogenous systems. OLCF program director Buddy Bland at the time recalled, “Choosing a GPU-accelerated system was considered a risky choice.”

The Top500 List organizers reported in 2012:

“When the 40th edition of the list was released at the start of SC12, the No. 1 position was claimed by Titan, a 560,640 processor system with a Linpack performance of 17.6 petaflop/s. Oak Ridge National Laboratory’s Titan is a Cray XK7 system that relies on a combination of GPUs and traditional CPUs to make it the world’s most powerful supercomputer. Each of Titan’s 18,688 nodes contains an NVIDIA Tesla K20 GPU along with a 16-core AMD Opteron 6274 CPU processor, giving the system a peak performance of more than 27 petaflops. Titan also has more than 700 terabytes of memory.

“Titan’s use of GPUs also points the way for future scientific supercomputers. Because GPUs provide high-performance and energy-efficient computing power, they will allow supercomputing systems to become ever more powerful while avoiding the obstacles inherent in growing size and power consumption.”

Now, as Messer pointed out, “We rode the crest of hybrid CPU-GPU computing, for large scale computing, we were at the top of that wave. And now my gosh, it’s the world, literally the world at this point.”

FYI, this article began as an intent to discuss the QC-HPC integration project being run as part of SummitPLUS, but it seemed a stretch to squeeze the two together given Messer’s enthusiasm for the SummitPLUS program generally and Summit’s remarkable history. HPCwire will post an article on the SummitPLUS QC-HPC integration project shortly.

Stay tuned

Top image: Bronson Messer, director of science, OLCF, with the IBM-built Summit supercomputer, was taken in 2019 when Summit was atop the Top500 List.

LIST SUMMITPLUS PROJECTS