Exascale Watch: El Capitan Will Use AMD CPUs & GPUs to Reach 2 Exaflops

By John Russell

March 4, 2020

HPE and its collaborators reported today that El Capitan, the forthcoming exascale supercomputer to be sited at Lawrence Livermore National Laboratory and serve the National Nuclear Security Administration (NNSA), will use AMD’s next-gen ‘Genoa’ Epyc CPUs and Radeon Instinct GPUs and deliver 2 exaflops (peak double-precision) performance, a 30 percent increase over the original spec. The new system, expected to be put into service in 2023, will be 10x faster than Summit, the fastest publicly-ranked supercomputer in the world today (Top500, November 2019).

The choice of AMD processor technology had not been made when the Department of Energy first announced the ~$600 million El Capitan procurement last August. Cray, now part of HPE, was announced as the prime contractor as was selection of its Shasta architecture. More detail on the CPU/GPU selections along with a few other system elements were presented in a media pre-briefing this week given by Bronis de Supinski, CTO, LLNL, Steve Scott, SVP, senior fellow, and CTO, HPE, and Forrest Norrod, SVP and GM, datacenter and embedded systems group, AMD.

HPE, through Cray, has been the big winner so far in the U.S. Exascale sweepstakes, obtaining contracts for all three systems – Aurora, with an Intel CPU/GPU pair; Frontier, with another AMD CPU/GPU pair, and El Capitan, which we now know will also feature AMD processors and AMD accelerators. After re-entering the HPC server market with its Epyc line of CPUs in 2017, AMD at first treaded lightly in pairing Epyc with Radeon GPUs in high-end servers. That clearly has changed.

Steve Scott, HPE/Cray

Talking about the delayed processor selections for El Capitan, Scott said, “The strategy that they’ve (DoE/LLNL) used – and increasingly others are using it as well – is to choose the system architecture, start getting that in place, and then make the processor choice as late as possible in the process. By doing that you can you get better visibility. Your headlights are illuminating farther in the future. The processor roadmaps have continued to improve and by making that decision later you tend to get a better result in the end, and that’s exactly what happened here.”

He declined to specify the process technology for the new processors but there’s been speculation Genoa will be fabbed on a 5nm process. We may know more soon, “We’ll be unpacking more the details in those parts as time goes by,” said AMD’s Norrod, “Our next disclosure on Genoa will, quite frankly…we’ll say a little bit about that at our financial analyst day which is this Thursday.” That’s tomorrow.

El Capitan’s primary mission is within NNSA’s advanced simulation and computing program, which uses simulations to certify the country’s nuclear stockpile is safe, secure and reliable. “To provide that certification we require complex simulations and as the nuclear stockpile ages, the complexity of the simulations only increases and need to be able to use larger and larger systems,” said de Supinski.

El Capitan will leverage HPE’s Shasta architecture which is at the core of the most recent refresh of the HPE/Cray advanced scale product line. Other core components include a new software stack, new Slingshot interconnect technology, and new storage system.

“[The] new software stack that provides a much more dynamic cloud like environment for hybrid workflows,” said Scott. “It has open documented APIs between the software components. It has a management system that’s built with redundant microservices and managed as a Kubernetes cluster, and has robust container support to allow users to take any workload that runs anyplace and run under this system as well.” A systems monitoring framework will run underneath the stack to optimize performance and help predict failures.”

Calling it a future-proof design, Scott said “We’ve designed it to accommodate a wide diversity of processors, different amounts of power, different types of processing, different physical sizes of the processor and memory system. And we’ve given it the power and the cooling headroom to handle processors that are headed again in the years ahead up to the kilowatt power levels.”

El Capitan will be liquid cooled and have an energy budget between 30-to-40 megawatts with expectations it will end up closer to 30MW than 40MW according to Scott. Slingshot and high performance Ethernet comprise the planned system interconnect. The planned storage system is HPE’s new ClusterStor E1000, which Scott said is, “a highly flexible tiered storage system using flash and hard drive partitions That allows you to individually optimize for performance as well as capacity and then does intelligent tiering of data between the partitions and this attaches directly to the slingshot interconnect, which helps take out cost and complexity and latency.”

Shasta Compute Blade

Specific performance specs were generally not disclosed. The new AMD CPU will use the Zen4 core which is reportedly on schedule for launch in 2021. The new CPU-GPU pairing (A-plus-A in AMD parlance) will leverage AMD’s Infinity fabric 3.0 to deliver memory coherency. The detailed node structure and number of nodes for El Capitan were not discussed in the pre-briefing but the official press release characterized the architecture as, “using accelerator-centric compute blades (in a 4:1 GPU to CPU ratio, connected by the 3rd Gen AMD Infinity Architecture for high-bandwidth, low latency connections) to increase performance for data-intensive AI, machine learning and analytics needs by offloading processing from the CPU to the GPU.”

As is the case generally in heterogeneous architectures, accelerators handle most of the work and require efficient IO. Norrod said, “We have next generation memory and IO subsystems that can provide non-blocking access to memory, non-blocking access to IO, and ensure that the full power of the Zen4 CPU engine and the Radeon Instinct GPU engines.”

He said the new GPU is optimized for high performance computing and machine intelligence applications. “It has extensive mix precision operations to optimize that deep learning performance, as well as [the ability] to provide peak single and dual precision performance with more traditional HPC applications. It does embody a next generation of high bandwidth memory (HBM) memory on package to provide the memory bandwidth and capacity that’s so critical to again feed the beast (GPU).”

While many data analytic workloads look quite different from high performance simulation, Scott said, “It turns out AI is one of the workloads that shares a lot in common with high performance simulation. Typically, the granularity or the precision that you use for the computations is quite a bit different. Most of AI is done at 16- or, or 32-bit precision, whereas, most of the scientific simulation is done at 64-bit precision. But modern processors like the AMD GPUs can take their function units and run them either in 64-bit mode or in 16-bit mode or 32-bit mode depending upon the particular computation. [To do that] you need a strong interconnect and need very high memory bandwidth which it shares in common with scientific workloads.

“We find the combination of the CPU and GPU with flexible precision, married with very high memory bandwidth and interconnect bandwidth and storage bandwidth to be well suited for both simulation and AI workloads and we can use all the compute nodes in the system to bring to bear to either those workloads,” Scott said.

Interestingly, AI is not currently a top priority at LLNL.

De Supinski said, “We’re doing a lot of research and development at Livermore exploring how we can bring [AI] to bear our simulations. Whereas we need a certain accuracy, deep learning models are probabilistic and so you can often be good enough with lower precision operations whereas we have to be able to understand where the errors are and where they are becoming larger because of the reduced precision and then be able to bring some mechanism in to increase precision and accuracy required.”

At the pre-briefing, a question was asked about El Capitan’s ability to use non-GPU accelerators. Scott said while GPUs are currently the AI accelerator of choice, many users are looking at alternatives and that El Capitan’s system architecture is “designed to accommodate that kind of heterogeneous mix.”

De Supinski noted LLNL is using an unclassified system, Lassen, a sister machine to the classified Sierra system, to learn more about emerging AI accelerators. “We’re actively exploring ways of adding purpose-built machine learning accelerators to that system. I would anticipate that the mechanism by which we’re doing that is available entirely in El Capitan; that is we can add additional nodes to the system that are designed specifically for that purpose. We will see how things go with our exploratory studies on Lassen. If they go well, we will be very likely to engage HPE in helping us figure out how we can exploit that.”

AMD, HPE and LLNL are collaborating on software tools for El Capitan. Part of the plan is to leverage AMD’s ROCm framework to take advantage of  “coherent acceleration in the OpenMP environment as well as other environments” according to Norrod.

Scott said, “As part of this procurement, the Department of Energy has provided additional funds beyond the purchase of the machine to fund non-recurring engineering efforts and one major piece of that is to work closely with AMD on enhancing the programming environment for their new CPU-GPU architecture.” Work is ongoing by all three partners to take the critical applications and workloads forward and optimize them to get the best performance in the machine when El Capitan is delivered.

De Supinski emphasized, “This is a collaborative process particularly for the software. Some of the software for the system is being developed at Lawrence Livermore in addition to the applications. For instance, we very much expect Spack, which is an open source management package [to able to run] on the new system.”

One interesting feature which was not discussed in the briefing but was mentioned in the official announcement is El Capitan’s planned use of optical data transmission.

According to the release, “HPE is expanding its partnership with LLNL to actively explore HPE optics technologies, a computing solution that uses light to transmit data, to feature in the DOE’s El Capitan. HPE’s optics technologies stem from R&D efforts related to PathForward, a program backed by U.S. DOE’s Exascale Computing Project. HPE developed and demonstrated breakthrough optics prototypes that integrate electrical-to-optical interfaces to enable broad use in future classes of system interconnects. Together, HPE and LLNL are exploring ways to integrate these optics technologies with HPE’s Cray Slingshot for DOE’s El Capitan to transmit more data, more efficiently. This approach aims to improve power efficiency, reliability and ability to cost-effectively increase global system bandwidth.”

Optical data transmission is a hotbed of research with many companies aggressively seeking practical implementation.

El Capitan will become NNSA’s fastest computer and greatly enhance NNSA’s ability to run 3D simulations quickly instead of 2D simulation. LLNL is managing the new system for the NNSA and has developed emerging techniques that allow researchers to create faster, more accurate models for primary missions across stockpile modernization and inertial confinement fusion (ICF), a key aspect of stockpile stewardship.

Link to the official announcement: https://www.llnl.gov/news/llnl-and-hpe-partner-amd-el-capitan-projected-worlds-fastest-supercomputer

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