Berkeley Lab’s John Shalf Ponders the Future of HPC Architectures

By Kathy Kincade

June 27, 2019

Editor’s note: Ahead of John Shalf’s well-attended and well-received “high-bandwidth” keynote at ISC 2019, Shalf discussed the talk’s major themes in an interview with Berkeley Lab’s Kathy Kincade. 

What will scientific computing at scale look like in 2030? With the impending demise of Moore’s Law, there are still more questions than answers for users and manufacturers of HPC technologies as they try to figure out what their next best investments should be. As he prepared to head to ISC19 in Frankfurt, Germany, to give a keynote address on the topic, John Shalf – who leads the Computer Science Department in Lawrence Berkeley National Laboratory’s Computational Research Division – shared his thoughts on what the future holds for computing technologies and architectures in the era beyond exascale. ISC took place June 16-20; Shalf’s keynote was on Tuesday, June 18.

What was the focus of your keynote at ISC?

What the landscape of computing, in general, is going to look like after the end of Moore’s Law. We’ve come to depend on Moore’s Law and to really expect that every generation of chips will double the speed, performance, and efficiency of the previous generation. Exascale will be the last iteration of Moore’s Law before the bottom drops out – and the question then is, how do we continue? Is exascale the last of its kind, or are we going to embark on a first-of-its-kind machine for the future of computing?

How long have you been thinking/talking about what’s next for HPC after Moore’s Law?

Where we are now is really the second shoe dropping. I got involved in the Exascale Computing Initiative discussions back in 2008, but actually, my interest in this predates exascale. Back in 2005, David Patterson’s group at UC Berkeley was talking about it in the Parallel Computing Laboratory, and we spent two years there in discussion and debate about the end of  Dennard’s scaling. Ultimately we published “The Landscape of Parallel Computing Research: A View from Berkeley,” which was the prediction that parallel computing would become ubiquitous on account of clock frequencies no longer scaling at exponential rates. This was followed closely by the DARPA 2008 Exascale report that set the stage for the Exascale Initiative for HPC. So the end of Dennard’s scaling was the first shoe to drop, but we always knew that the second shoe would drop fairly soon after the first. And the second shoe dropping means we can’t shrink transistors at all anymore, and that is the real end of Moore’s Law. Exascale is addressing the mass parallelism from the first shoe dropping, and I’ve been concerned about the second shoe dropping during the entire 10-year ramp-up to the Exascale Computing Initiative and subsequent Project, as were many others who were involved in writing the View from Berkeley report and the DARPA 2008 report.

How is the slowing of Moore’s Law already affecting HPC technologies and the industry itself?

We are seeing already procurement cycles stretching out so that the replacement of machines is happening at a slower pace than it has historically. Eric Strohmaier at Berkeley Lab has been tracking the replacement rate on the TOP500 very closely, and he has seen a noticeable slowdown in system replacement rates. I’ve also heard from our colleagues in industry that this is a troubling development that will affect their business model in the future. But we are also seeing these effects in the mega datacenter space, such as Google, Facebook, and Amazon. Google has actually taken to designing its own chips, specialized for particular parts of their workflow, such as the Tensor Processing Unit (TPU). We will probably see even more specialization in the future, but how this applies to HPC is less clear at this point – and that’s what I would like to get people thinking about during my keynote.

Is the lithography industry experiencing a parallel paradigm shift?

Yes, the lithography industry is also being affected, and something’s going to need to change in the economics for that industry. What we have seen in the past decade is that we’ve gone from nearly a dozen leading-edge fabs down to two. Global Foundries recently dropped out as a leading-edge fab, and Intel has had a huge amount of trouble getting its 10nm fab line off the ground. So clearly there are huge tectonic shifts happening in the lithography market as we speak, and how that will resolve itself ultimately remains unclear.

Do we have to start imagining an entirely new computing technology development and production process?

I think the way in which we select and procure systems is going to have to be revisited. While using user application codes to run benchmarks to assess the performance and usability of emerging systems is a great way for us to select systems today that use general purpose processors, it doesn’t seem to be a very good approach for selecting systems that might have specialized features for science. In the future, we need to be more closely involved in the design of the machines with our suppliers to deliver machines that are truly effective for scientific workloads. This is as much about sustainable economic models as it is a change in the design process. The most conventional or even the most technologically elegant solution might not survive, but the one that makes a lot of money will. But our current economic model is breaking.

Looking ahead, I see three paths going forward. The first is specialization and better packaging – specialization meaning designing a machine for a targeted class of applications. This has already been demonstrated in the successful case of the Google TPU, for example. So that is the most immediate path forward.

Another potential path forward is new transistor technology that replaces CMOS that is much more energy efficient and scalable. However, we know from past experience that it takes about 10 years to get from a lab demo to a production product. There are promising candidates, but no clear replacements demonstrated in the lab, which means we are already 10 years too late for that approach to be adopted by time Moore’s Law fails. We need to dramatically accelerate the discovery process in that area through a much more comprehensive materials-to-systems co-design process.

The third approach is to explore alternative models of computation such as quantum and neuromorphic and other, related approaches. These are all fantastic, but they are really expanding computing into areas where digital computing performs very poorly. They aren’t necessarily replacement technologies for digital general purpose computing; they are merely expanding into areas where digital isn’t very effective to start with. So I think these are worthy investments, but they aren’t the replacement technology. They will have a place, but how broadly applicable they will be is still being explored.

What about the development of new chip materials – what role might they play in the future of HPC architectures?

New materials are definitely part of the CMOS replacement. It’s not just new materials; fundamental breakthroughs in solid-state physics will be required to create a suitable CMOS replacement. The fundamental principle of operation for existing transistor technology cannot be substantially improved beyond what we see today. So to truly realize a CMOS replacement will require a new physical principle for switching, whether electrical, optical, or magnetic switching. A fundamentally new physical principle will need to be discovered and that, in turn, will require new materials and new material interfaces to realize effective and manufacturable solutions.

Are there any positives when you look at what is happening in this field right now?

Yes, definitely there are positives. We believe the co-design processor is going to require not just software and hardware people to collaborate, it is going to require this collaboration to go all the way down into the materials and materials physics level. And for the national laboratories, this is a great opportunity for us to work closely with our colleagues in the materials science divisions of our respective laboratories. I work at a national laboratory because I’m excited by cross-disciplinary collaboration, and clearly, that is the only way we are going to make forward progress in this area. The recent ASCR Extreme Heterogeneity and DOE Microelectronics BRNs show strong interest by DOE in this deep co-design and collaborative research that is really needed in this space. So to that extent, it is kind of an exciting time.

When you think about the future of HPC and supercomputing architectures and technologies, what do you imagine they will look like 10 years from now?

I think we’re going to have smaller machines that are more effective for the workflows they target. For three decades we have become used to ever-growing, larger and larger machines, but that doesn’t seem to be the winning approach for creating effective science in the post-exascale and post-Moore era.


About the Author

Kathy Kincade is a science & technology writer and editor with the Berkeley Lab Computing Sciences Communications Group.


Article courtesy Berkeley Lab; Feature image credit: ISC High Performance.

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!

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans to release an AI accelerator with heavy in-memory processing Read more…

ASC24 Student Cluster Competition: Who Won and Why?

June 18, 2024

As is our tradition, we’re going to take a detailed look back at the recently concluded the ASC24 Student Cluster Competition (Asia Supercomputer Community) to see not only who won the various awards, but to figure out Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch that D-Wave’s brand of analog quantum computing (quantum Read more…

Apple Using Google Cloud Infrastructure to Train and Serve AI

June 18, 2024

Apple has built a new AI infrastructure to deliver AI features introduced in its devices and is utilizing resources available in Google's cloud infrastructure.  Apple's new AI backend includes: A homegrown foun Read more…

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and implementation of artificial intelligence (AI) tools, while the Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are many interesting stories, and only a few ever become headli Read more…

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch Read more…

Shutterstock_666139696

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and imp Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are Read more…

Labs Keep Supercomputers Alive for Ten Years as Vendors Pull Support Early

June 12, 2024

Laboratories are running supercomputers for much longer, beyond the typical lifespan, as vendors prematurely deprecate the hardware and stop providing support. Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

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…

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…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then 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…

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…

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…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

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…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

GenAI Having Major Impact on Data Culture, Survey Says

February 21, 2024

While 2023 was the year of GenAI, the adoption rates for GenAI did not match expectations. Most organizations are continuing to invest in GenAI but are yet to Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir 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…

A Big Memory Nvidia GH200 Next to Your Desk: Closer Than You Think

February 22, 2024

Students of the microprocessor may recall that the original 8086/8088 processors did not have floating point units. The motherboard often had an extra socket fo Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire