Fastest Academic Supercomputer Enters Full Production at TACC, Just in Time for Hurricane Season

By Tiffany Trader

September 3, 2019

Frontera, the NSF supercomputer installed at the Texas Advanced Computing Center (TACC) in June, passed its formal acceptance last week and is now officially launched. The Dell EMC-built supercomputer spans 8,008 Mellanox HDR connected Xeon Platinum 8280 nodes, capable of 23.5 Linpack petaflops, making it the fastest academic supercomputer in the world. Frontera joins more than a dozen other TACC systems, including Stampede2, which is now the second fastest academic system in the U.S.

Ahead of today’s unveiling ceremony, which brought representatives from Intel, IBM, Nvidia, Mellanox, DDN and the National Science Foundation (NSF) to the University of Texas at Austin, HPCwire spoke with TACC Director Dan Stanzione about the science that’s already being enabled on Frontera, as well as other details of the $60-million NSF leadership computing award (announced last year), including two GPU-powered subsystems that are currently undergoing testing in preparation for production-readiness within the next couple of months.

TACC’s Frontera system, featuring 8,008 Dell EMC PowerEdge server nodes

The primary Frontera system is powered by Intel’s highest-bin SP Cascade Lake processors (the Xeon Platinum 8280s), interconnected with HDR-100 links to each node, and full 200 Gbps HDR links between leaf and core switches. Data Direct Networks contributed the main storage system (50+ PB disk, 3PB of flash, 1.5/TB sec of I/O capability). Frontera’s 90 racks leverage direct liquid cooling technology from CoolIT, enabling about twice as many racks in the same footprint as the recently retired Stampede1. Running Linpack (the system entered the June 2019 Top500 list at number five), Frontera’s Xeons never got above 50C so won’t be thermally constrained; unlike Stampede2, which hit temps of up to 85C on its Linpack run, Stanzione shared.

In August, TACC fielded two Frontera subsystems. One is a cluster comprised of 360 Nvidia Quadro RTX 5000 GPUs submerged in liquid coolant filled racks developed by GRC. The liquid cooling specialists–formerly known as Green Revolution Cooling–are also based in Austin and TACC deployed the first GRC prototype outside of the company’s lab back in 2009. The new six tank system (internally, it’s called Maverick) uses Mellanox HDR-100 networking and provides 4 petaflops of peak single-precision performance.

Stanzione revealed that the Quadro RTX nodes were brought in under a new single-precision academic program that Nvidia is launching that provides CUDA licensing support and maybe even some nice discounts for RTX family parts for academic sites (we are tracking down additional details and will report further soon).

TACC’s new Longhorn: an IBM Power9 cluster with 448 Nvidia V100 GPUs

TACC is also deploying an IBM Power9 subsystem with 448 V100 GPUs (in the 4:1 GPU-to-CPU configuration) offering a peak aggregate output of 3.5 double-precision petaflops. The InfiniBand-EDR connected cluster reprises the Longhorn name at TACC (the original Longhorn, a GPU cluster outfitted with Quadro FX 5800 parts, was decommissioned in 2014). Frontera’s two GPU subsystems will target artificial intelligence, machine learning, and molecular dynamics research.

In the coming months, Frontera will also be integrating with cloud providers Amazon, Google, and Microsoft to provide researchers access to a range of emerging computing technologies and long-term storage. “Cloud is not an either/or decision, they don’t really do the same thing,” Stanzione said at the Rice Oil and Gas conference in March, referring to traditional on-prem HPC and cloud. “We are bringing in the cloud to use the things the cloud does well on, publishing data, and access to composable workflows, things like natural language processing. And also to play with the latest gear, FPGAs, Tensor Processors, etc., and see what the users want.”

Since entering early science operations the first week of July, Frontera has supported 39 science projects from across the NSF space and Stanzione expects that number to expand to about 100. As NSF’s leadership-class “tier 1” system, the follow-on to Blue Waters at the National Center for Supercomputing Applications, Frontera’s mission is to provide open science projects with significant system time; some projects are at the scale where they require 5 percent of the machine’s time in a year, for example. Smaller projects will be kept on Stampede2.

Frontera will run for at least five years; 80 percent of cycles are selected through a competitive process, while 20 percent are discretionary. Frontera will support high-impact science in nearly every domain. Early workloads include astrophysics, quantum chemistry, photovoltaic materials research, machine learning – and just this past week hurricane modeling.

TACC has been working with the storm research community for about a decade and has discretionary allocation time for just this kind of emergency, seasonal workload. “They call us when there’s a storm, and usually, we’re able to let them on,” said Stanzione. “Obviously, these forecasts are much less useful two or three days from now. So we just give them a chunk of time, wherever, wherever we have the space.”

As Hurricane Dorian was still gaining strength over the Atlantic last week, TACC provided a discretionary allocation for researchers Clint Dawson (University of Texas at Austin) and Jason Fleming (University of North Carolina), to carry out storm surge forecasting on the hurricane that reached category-5 status on Sunday using both Stampede2 and Frontera.

The researchers were able to get their code up in running on Frontera in about 10 minutes due to how similar Frontera is to Stampede2 with respect to its software stack and architecture — and they achieved twice the speed on the same number of nodes, Stanzione told us. Some of this speedup is attributable to the fact that they had a fairly quiet system to run on, so less contention for I/O and other resources. More generally, Stanzione is seeing a 10-15 percent performance uplift moving from Skylake to Cascade Lake, with up to a 30 percent improvement on codes that aren’t memory bandwidth bound.

The Intel security mitigations (that address side channel attack vulnerabilities, Spectre and Meltdown, et al.), haven’t impacted TACC very significantly, according to Stanzione. With the Skylake-based Stampede2 nodes, at most they saw up to a 10 percent hit on data intensive workloads with a lot of small I/O, but for the majority of floating point bound simulation codes, the performance penalty was around 1-2 percent. With the new hardware mitigations on Cascade Lake, TACC is not able to conduct direct with-and-without comparisons, but given codes are hitting the projected performance targets, Stanzione believes that if there is an impact, it’s minor.

TACC will also be deploying a small fleet of Optane nodes as part of the Frontera project: 16 quad-socket nodes filled with the 256 GB Optane Persistent Memory Modules — that’s six per socket, 24 per node, for an aggregate of more than 98 TB of Optane memory.

TACC’s engineers will be experimenting with the Optane DIMMs in both “I/O mode” and memory mode, using it for in-memory database applications, and to boost fault tolerance for check pointing. “We haven’t decided if we like it better as a fast storage node or a sort of a slow but really large memory node because it’s sort of a hybrid between storage and memory in a lot of ways. So we’ll have a hundred terabytes, 16 of these broad-socket nodes, and we’ll see what users want to do with it,” said Stanzione.

A phase-2 NSF leadership system that is 10x the capacity of Frontera is on the roadmap for 2024 deployment, which will be after the DOE Frontier (ORNL) and El Capitan (LLNL) machines are up and running. “Five year planning is more in the style that DOE usually gets to do,” Stanzione has said. This longer lead time allows TACC and its NSF collaborators to gather data and thoughtfully plan what makes sense for the second half of the 2020s. They’ll have a BOF at SC this year to start getting feedback and will begin working on a conceptual design next year, drawing on the lessons learned from Frontera and its subsystems, and as always listening closely to their end user scientific community, whom Stanzione continuously references throughout our nearly hour-long conversation.

This community-needs focus came up again when we talked about the decision to go with a straight x86 machine (plus smaller GPU subsystems), even though in the U.S. at least extreme-scale architectures have been trending toward fat GPU-node systems. (See five DOE CORAL awards, Perlmutter at NERSC, Big Red 2 at Indiana University, to name a few. There are of course notable exceptions to the fat-node approach, especially internationally, e.g., SuperMUC-NG at LRZ and the upcoming post-K Arm “Fugaku” system at RIKEN. For the record, Stanzione is intrigued by Arm and the Fujitsu post-K chip, specifically.)

“It helps that in the DOE space, most of the codes are C++. They can compel the developers to go the direction they want to go, and they use templating libraries like Cocos and Raja to get performance out of those,” said Stanzione. “But, you know, if I told our community to do that, first of all, right now, none of the codes would work. And, you know, I don’t have two years to get to it, either. Most of those systems [that you mentioned] are around the 2021 timeframe. Assuming they can make the software switch — and they have lots of good reasons to think that they can — their community is just very different.”

In a bit of friendly posturing, Stanzione also said he expects that despite those [big GPU] systems having higher spec’d performance (peak and Linpack) and higher funding levels, there will be a set of codes for which Frontera will outrun systems like Summit and Sierra. “We’ll put our productivity per dollar up against any of them,” said Stanzione, “and I think there’ll be a set of problems, particularly some of the big adaptive mesh, partial differential equations type models, where I bet we can outrun them in just raw speed because those codes just aren’t really friendly to those fat nodes. If you take away the GPUs, then Summit’s essentially a 3,000 node machine of plain-old CPUs and we have a lot more (nodes) than that. So, I think we will find some problems over the next two or three months, where Frontera is the fastest machine in the world.”

Read more about the science Frontera will be enabling at: https://fronteraweb.tacc.utexas.edu/

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire