The Deep500 – Researchers Tackle an HPC Benchmark for Deep Learning

By John Russell

January 7, 2019

How do you know if an HPC system, particularly a larger-scale system, is well-suited for deep learning workloads? Today, that’s not an easy question to answer in the sense there is no widely agreed-upon benchmark or reference architecture for comparing DL performance across systems. A group of researchers led by Tal Ben-Nun and Torsten Hoefler of ETH Zurich has set out to develop Deep500 – a benchmarking suite, reference architecture, and, yes, contest – to provide a meaningful assessment tool for deep learning capabilities on HPC platforms.

“We would like the community to establish good benchmarking practice for large-scale deep learning scientific computing workloads, while still taking the correctness of the end result (i.e., accuracy and generalization) into account,” Hoefler told HPCwire recently. “That is why we created Deep500. This benchmark should be insightful to HPC researchers and supercomputer vendors, but also to the users in computational science who use deep learning as an inference tool. We should thus avoid common pitfalls that accompany measuring performance alone.”

SC18 Papers Chair Torsten Hoefler with ETH Zurich. Image courtesy of SC18.

Hoefler and Ben-Nun have set up Deep500.org and organized a well-attended Deep500 BOF at SC18, which featured an expert panel, and served as further outreach to the HPC community. How quickly Deep500 will take shape is uncertain. The researchers are working with other prominent researchers from academia and industry; for example Satoshi Matsuoka, director of Riken Center for Computational Science, and Pradeep Dubey, Intel Fellow and director of Intel’s Parallel Computing Lab, were among the panelists.

Their idea is to create a useful tool not just a vanity trophy. That message came through in force when Matsuoka described in some detail the struggle his group faced in developing procurement criteria for Japan’s AI Bridging Cloud Infrastructure (ABCI) which is intended specifically to handle artificial intelligence workloads. (ABCI was number seven on the most recent Top500).

“If you ever procure a machine, the benchmark will have to be very precise, because if there is any wiggle room, people will cheat. I’m not blaming anyone. People will improvise. Find loopholes. This has always been an issue with benchmarks,” said Matsuoka at the BOF.

“The immediate next step in Deep500,” Ben-Nun told HPCwire, “is to construct a meta-review of how deep learning is used by the scientific community, so that the workload types and their properties are clear. We are organizing a monthly meeting with leading researchers and interested parties from the industry. The meetings are open and posted on the Deep500 website (https://www.deep500.org/). Following that, the next step is to establish a steering committee for the benchmark. It is imperative that we fix the ranking and metrics of the benchmark, as the community is undecided right now on several aspects of this benchmark. We intend to make considerable progress this year, reconvene at SC19.”

Ben-Nun and Hoefler suggest the HPC community response to the Deep500 concept has been positive; there is, they say, widespread recognition of the need for such a benchmark, but there are also significant challenges and differing views on how to solve them. Here are their thoughts on four prominent issues provided to HPCwire recently by email:

  • Appropriate Datasets. “First, public deep learning datasets originate from fields such as computer vision, and as an HPC community we must establish datasets that are relevant to scientific computing, including data types and dimensions. Second, image and speech processing datasets are fixed, which makes deep learning a strongly scaling problem and thus inherently limit a supercomputing benchmark, as well as fosters specialization of techniques to specific datasets. Therefore, we need to consider synthetically-generated, relevant datasets as well.”
  • Metrics. “As for metrics, there have been several opinions on the matter: is throughput the main ranking criterion, or is test accuracy and overall time-to-solution important? We are open to input from the community, and encourage anyone who is interested to join the meetings.”
  • Allowable Methods. “The next open question is the algorithms and methods we allow the competitors to use. This is especially important, since many of the recent advances in distributed deep learning relate to modifying the original synchronous SGD algorithm. Approaches such as gradient quantization, asynchrony, and sparsification currently dominate the field, as the robustness of deep learning still yields satisfactory results. Openness to different methods is important, but as traditional benchmarks measure hardware, they usually leave little room for changing the algorithm itself. To cope with this landscape it is important to know what kind of learning strategies are important for scientific computing and are used for learning at scale.”
  • Verification. “Last but not least is the issue of verification. How do we ensure that a result of the benchmark is correct? As opposed to HPL, HPCG, and Graph500, where the result is known, deep learning problems define loss functions and accuracy metrics, whose values vary due to the problem definition (stochasticity, datasets) and applied techniques. As evident in conferences such as NeurIPS (formerly NIPS), reproducibility is very important to the ML community, and we would like to guarantee it in Deep500.”

There are, of course, many (old and new) tests and datasets being used ad hoc to assess machine learning and deep learning capabilities of systems. Most are fairly narrow. One new effort – the MLPerf benchmark suite for assessing training and inference performance introduced last May – has attracted considerable support and recently released its first round of results (see HPCwire article, Nvidia Leads Alpha MLPerf Benchmarking Round.)

As you might expect the BOF conversation was wide-ranging; it covered everything from why such a benchmark is needed, what attributes it should encompass, to strategies for combating inevitable efforts to “beat” the test.

When Matsuoka suggested the new benchmark “should measure through-put and not time-to-solution” an audience member quickly challenged that idea; he recalled an earlier effort focused on throughput “but time-to-solution was so bad that everyone dropped it,” and added that the clever use of cache allowed cheating throughput. Matsuoka agreed but emphasized modern benchmarks need to be scalable for use on different size machines which can influence time to solution.

Clearly much work remains. Regardless, Deep500 efforts are forging ahead. Hoefler, Ben-Nun, and colleagues plan to post a new paper – A Modular Benchmarking Infrastructure for High-Performance and Reproducible Deep Learning – quite soon (Jan. or Feb.) on arXiv, and they whetted the BOF audience appetite with the slide below.

Along with Hoefler and Ben-Nun, panelists included: Dubey; Todd Gamblin (Center for Applied Scientific Computing, Lawrence Livermore National Lab); Tom Gibbs (Nvidia); Thorsten Kurth (National Energy Research Scientific Computing Center, LBNL); Matsuoka; and Jidong Zhai (Tsinghua University). A few of the brief BOF presentations re available on the Deep500 website.

Much of the conversation covered familiar ground but all of it was fascinating. Besides discussion of what a new benchmark should include there were a few snapshots of current DL practices and expectation based on a literature search and a fairly recent NERSC’s ML user survey. Presented below are a few slides from BOF panelists (click on slides to enlarge).

CURRENT DL TRENDS FROM BEN-NUN/HOEFLER
Ben-Nun and Hoefler presented a few snapshots from their literature review – “more than 100 papers” – of DL practices. Perhaps not surprisingly, the GPU use jumped dramatically in the past few years and now dominate. You may also find their paper, Demystifying Parallel and Distributed Deep Learning: An In-Depth Concurrency Analysis, of interest. Below are three of their slides.

NERSC ML USER SURVEY
Kurth of reviewed results of NERSC’s ML user survey which shows and interesting mix of tools being used with most models still run on relatively few CPUs/GPUs. He emphasized the importance of using open exchange formats to accommodate the rapidly changing/evolving ML framework landscape. Here are three of Kurth’s slides.

WHAT’S NEEDED IN DL BENCHMARK – INTEL
Intel’s Dubey emphasized the need for developing a benchmark that scales well, includes TCO, and can act as a proxy for “real-world, forward-looking” applications. Below are four of his slides.

 

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!

U.S. Quantum Director Charles Tahan Calls for NQIA Reauthorization Now

February 29, 2024

(February 29, 2024) Origin stories make the best superhero movies. I am no superhero, but I still remember what my undergraduate thesis advisor said when I told him that I wanted to design quantum computers in graduate s Read more…

pNFS Provides Performance and New Possibilities

February 29, 2024

At the cusp of a new era in technology, enterprise IT stands on the brink of the most profound transformation since the Internet's inception. This seismic shift is propelled by the advent of artificial intelligence (AI), Read more…

Celebrating 35 Years of HPCwire by Recognizing 35 HPC Trailblazers

February 29, 2024

In 1988, a new IEEE conference debuted in Orlando, Florida. The planners were expecting 200-300 attendees because the conference was focused on an obscure topic called supercomputing, but when it was announced that S Read more…

Forrester’s State of AI Report Suggests a Wave of Disruption Is Coming

February 28, 2024

The explosive growth of generative artificial intelligence (GenAI) heralds opportunity and disruption across industries. It is transforming how we interact with technology itself. During this early phase of GenAI technol Read more…

Q-Roundup: Google on Optimizing Circuits; St. Jude Uses GenAI; Hunting Majorana; Global Movers

February 27, 2024

Last week, a Google-led team reported developing a new tool - AlphaTensor Quantum - based on deep reinforcement learning (DRL) to better optimize circuits. A week earlier a team working with St. Jude Children’s Hospita Read more…

AWS Solution Channel

Shutterstock 2283618597

Deep-dive into Ansys Fluent performance on Ansys Gateway powered by AWS

Today, we’re going to deep-dive into the performance and associated cost of running computational fluid dynamics (CFD) simulations on AWS using Ansys Fluent through the Ansys Gateway powered by AWS (or just “Ansys Gateway” for the rest of this post). Read more…

Argonne Aurora Walk About Video

February 27, 2024

In November 2023, Aurora was ranked #2 on the Top 500 list. That ranking was with half of Aurora running the HPL benchmark. It seems after much delay, 2024 will finally be Aurora's time in the spotlight. For those cur Read more…

Royalty-free stock illustration ID: 1988202119

pNFS Provides Performance and New Possibilities

February 29, 2024

At the cusp of a new era in technology, enterprise IT stands on the brink of the most profound transformation since the Internet's inception. This seismic shift Read more…

Celebrating 35 Years of HPCwire by Recognizing 35 HPC Trailblazers

February 29, 2024

In 1988, a new IEEE conference debuted in Orlando, Florida. The planners were expecting 200-300 attendees because the conference was focused on an obscure t Read more…

Forrester’s State of AI Report Suggests a Wave of Disruption Is Coming

February 28, 2024

The explosive growth of generative artificial intelligence (GenAI) heralds opportunity and disruption across industries. It is transforming how we interact with Read more…

Q-Roundup: Google on Optimizing Circuits; St. Jude Uses GenAI; Hunting Majorana; Global Movers

February 27, 2024

Last week, a Google-led team reported developing a new tool - AlphaTensor Quantum - based on deep reinforcement learning (DRL) to better optimize circuits. A we Read more…

South African Cluster Competition Team Enjoys Big Texas HPC Adventure

February 26, 2024

Texas A&M University's High-Performance Research Computing (HPRC) hosted an elite South African delegation on February 8 - undergraduate computer science (a 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…

Apple Rolls out Post Quantum Security for iOS

February 21, 2024

Think implementing so-called Post Quantum Cryptography (PQC) isn't important because quantum computers able to decrypt current RSA codes don’t yet exist? Not Read more…

QED-C Issues New Quantum Benchmarking Paper

February 20, 2024

The Quantum Economic Development Consortium last week released a new paper on benchmarking – Quantum Algorithm Exploration using Application-Oriented Performa Read more…

Training of 1-Trillion Parameter Scientific AI Begins

November 13, 2023

A US national lab has started training a massive AI brain that could ultimately become the must-have computing resource for scientific researchers. Argonne N 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 Wins SC23, But Gets Socked by Microsoft’s AI Chip

November 16, 2023

Nvidia was invisible with a very small booth and limited floor presence, but thanks to its sheer AI dominance, it was a winner at the Supercomputing 2023. Nv 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…

Analyst Panel Says Take the Quantum Computing Plunge Now…

November 27, 2023

Should you start exploring quantum computing? Yes, said a panel of analysts convened at Tabor Communications HPC and AI on Wall Street conference earlier this y Read more…

Royalty-free stock illustration ID: 1675260034

RISC-V Summit: Ghosts of x86 and ARM Linger

November 12, 2023

Editor note: See SC23 RISC-V events at the end of the article At this year's RISC-V Summit, the unofficial motto was "drain the swamp," that is, x86 and Read more…

China Deploys Massive RISC-V Server in Commercial Cloud

November 8, 2023

If the U.S. government intends to curb China's adoption of emerging RISC-V architecture to develop homegrown chips, it may be getting late. Last month, China 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…

Leading Solution Providers

Contributors

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…

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…

Chinese Company Developing 64-core RISC-V Chip with Tech from U.S.

November 13, 2023

Chinese chip maker SophGo is developing a RISC-V chip based on designs from the U.S. company SiFive, which highlights challenges the U.S. government may face in 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…

Royalty-free stock illustration ID: 1182444949

Forget Zettascale, Trouble is Brewing in Scaling Exascale Supercomputers

November 14, 2023

In 2021, Intel famously declared its goal to get to zettascale supercomputing by 2027, or scaling today's Exascale computers by 1,000 times. Moving forward t 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…

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…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire