Flipping the Flops and Reading the Top500 Tea Leaves

By Tiffany Trader

November 13, 2017

The 50th edition of the Top500 list, the biannual publication of the world’s fastest supercomputers based on public Linpack benchmarking results, was released from SC17 in Denver, Colorado, this morning and once again China is in the spotlight, having taken what is on the surface at least a definitive lead in multiple dimensions. China now claims the most systems, biggest flops share and the number one machine for 10 consecutive lists. It’s a coup-level achievement to pull off in five years, disrupting 20 years of US dominance on the Top500, but reading deeper into the Top500 tea leaves reveals a more nuanced analysis that has as much to do with China’s benchmarking chops as it does its supercomputing flops.

PEZY-SC2 chip at ISC 2017 –click to enlarge

Before we thread that needle, let’s take a moment to review the movement at the top of the list. There are no new list entrants in the top ten and no change in the top three, but the upgraded ZettaScaler-2.2 “Gyoukou” stuck its landing for a fourth place ranking. Vaulting 65 spots, the supersized Gyoukou combines Xeons and PEZY-SC2 accelerators to achieve 19.14 petaflops, up from 1.68 petaflops on the previous list. The Top500 authors point out that the system’s 19,860,000 cores represent the highest level of concurrency ever recorded on the Top500 rankings.

Gyoukou also had the honor of being the fifth greenest supercomputer. Fellow ZettaScaler systems Shoubu system B, Suiren2 and Sakura, placed first, second and third respectively (see perf-per-watt numbers below). Nvidia’s DGX SaturnV Volta system, installed at Nvidia headquarters in San Jose, Calif., was the fourth greenest supercomputer.

Nov. 2017 Green500 top five — click to enlarge
Nov. 2017 Top500 top 10

Another upgraded machine, Trinity, moved up three positions to seventh place thanks to a recent infusion of Intel Knights Landing Xeon Phi processors that raised its Linpack score from 8.10 petaflops to 14.14 petaflops. Trinity is a Cray XC40 supercomputer operated by Los Alamos National Laboratory and Sandia National Laboratories.

Sunway TaihuLight datacenter (Wuxi, China)

China still has a firm grip on the top of the list with 93-petaflops Sunway TaihuLight and 33.86-petaflops Tianhe-2, the number one and and two systems respectively, which together provide the new list with 15 percent of its flops. Piz Daint, the Cray XC50 system installed at the Swiss National Supercomputing Centre (CSCS) remains the third fastest system with 19.6 petaflops. With Gyoukou in fourth position, the fastest US system, Titan, slips another notch to fifth place, leaving the United States without a claim to any of the top four rankings. Benchmarked at 17.59 petaflops, the five-year-old Cray XK7 system installed at the Department of Energy’s Oak Ridge National Laboratory, captured the top spot for one list iteration before being knocked off its perch in June 2013 by China’s Tianhe-2. This is the first time in the list’s 24-year history that the US has not held at least a number four ranking.

Although China has enjoyed number one bragging rights for nearly four years, this is the first list that it also dominates by both system number and aggregate performance share as well. China has the most installed systems: 202 compared to 159 on the last list, while US is in second place with 144 down from 169 six month ago (Japan ranks third place with 35, followed by Germany with 20, France with 18, and the UK with 15.). Aggregate performance is similar: China holds 35.3 percent of list flops, and the US is second with 29.8 percent (then Japan with 10.8 percent, Germany with 4.5 percent, UK with 3.8 percent and France with 3.6 percent).

Based on these metrics, undoubtedly some publications will proclaim China’s supercomputing supremacy, but that would be premature. When China expanded its Top500 toehold by a factor of three at SC15, Intersect360 Research CEO Addison Snell remarked that it wasn’t so much that China discovered supercomputing as it discovered the Top500 list. This observation continues to hold water.

An examination of the new systems China is adding to the list indicates concerted efforts by Chinese vendors Inspur, Lenovo, Sugon and more recently Huawei to benchmark loosely coupled Web/cloud systems that strain the definition of HPC. To wit, 68 out of the 96 systems that China introduced onto the latest list utilize 10G networking and none are deployed at research sites. The benchmarking of Internet and telecom systems for Top500 glory is not new. You can see similar fingerprints on the list (current and historical) from HPE and IBM, but China has doubled down. For comparison’s sake, the US put 19 new systems on the list and eight of those rely on 10G networking.

Top500 development over time–countries by performance share. US is red; China is dark blue. Click to enlarge.

Not only has the Linpacking of non-HPC systems inflated China’s list presence, it’s changed the networking demographics as the number of Ethernet-based machines climbs steadily. As the Top500 authors note, Gigabit Ethernet now connects 228 systems with 204 systems using 10G interfaces. InfiniBand technology is now found on 163 systems, down from 178 systems six months ago, and is the second most-used internal system interconnect technology.

Snell provided additional perspective: “What we’re seeing is a concerted effort to list systems in China, particularly from China-based system vendors. The submission rules allow for what is essentially benchmarking by proxy. If Linpack is run and verified on one system, the result can be assumed for other systems of the same (or greater) configuration, so it’s possible to put together concerted efforts to list more systems, whether out of a desire to show apparent market share, or simply for national pride.”

Discussions of list purity and benchmarking by proxy aside, the High Performance Linpack or any one-dimensional metric has limited usefulness across today’s broad mix of HPC applications. This truth, well understood in HPC circles, is not always appreciated outside the community or among government stakeholders who want “something to show” for public investment.

“Actual system effectiveness is getting more difficult to compare, as the industry swings back toward specialized hardware,” Snell commented. “Just because one architecture outperforms another on one benchmark doesn’t make it the best choice for all workloads. This is particularly challenging for mixed-workload research environments trying to serve multiple domains. 88 percent of all HPC users say they will need to support multiple architectures for the next few years, running applications on the most appropriate systems for their requirements.”

Chip technology – click to expand (Source: Top500)

There has been stagnation on the list for several iterations and turnover is historically low. Neither Summit or Sierra (the US CORAL machines, projected to achieve 150-200 petaflops) nor the upgraded Tianhe-2A (projected 94.97 petaflops peak) made the cut for the 50th list as had been speculated. While HPC is seeing a time of increased architectural diversity at the system and processor level, the current list is less diverse by some measures. To wit, of the 136 new systems on the list, Intel is foundational to all of them (36 of these utilize accelerators*). So no new Power, no new AMD (it’s still early for EPYC) and nothing from ARM yet. In total 471 systems, or 94.2 percent, are now using Intel processors, up a notch from 92.8 percent six months ago. The share of IBM Power processors is at 14 systems, down from 21 systems in June. There are five AMD-based systems remaining on the list, down from seven one year ago.

Nvidia’s New SaturnV Volta system. Click to enlarge.

In the US, IBM Power9 systems Summit and Sierra are on track for 2018 installation at Oak Ridge and Livermore labs (respectively), and multiple other exascale-focused systems are in play in China, Europe and Japan, showcasing a new wave of architectural diversity. We expect there will be more exciting supercomputing trends to report on from ISC 2017 in Frankfurt.

*Breakdown of the 36 new accelerated systems: 29 have P100s (one with NVLink, an HPE SGI system at number 292 (Japan)), one internal Nvidia V100 Volta system (#149, SaturnV Volta); one K80-based system (#267, Lenovo); two Sugon-built P40 systems (#161, #300), and three PEZY systems (#260, #277, #308). Further, out of the 36, only the internal Nvidia machine is US-based. 30 are Chinese (by Lenovo, Inspur, Sugon); the remaining five are Japanese (by NTT, HPE, PEZY).

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!

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to communities across the globe. As climate change is warming ocea Read more…

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical claims. A paper published on July 10 by researchers in the U. Read more…

Belt-Tightening in Store for Most Federal FY25 Science Budets

July 15, 2024

If it’s summer, it’s federal budgeting time, not to mention an election year as well. There’s an excellent summary of the curent state of FY25 efforts reported in AIP’s policy FYI: Science Policy News. Belt-tight Read more…

Peter Shor Wins IEEE 2025 Shannon Award

July 15, 2024

Peter Shor, the MIT mathematician whose ‘Shor’s algorithm’ sent shivers of fear through the encryption community and helped galvanize ongoing efforts to build quantum computers, has been named the 2025 winner of th Read more…

Weekly Wire Roundup: July 8-July 12, 2024

July 12, 2024

HPC news can get pretty sleepy in June and July, but this week saw a bump in activity midweek as Americans realized they still had work to do after the previous holiday weekend. The world outside the United States also s Read more…

Nvidia, Intel not Welcomed in New Apple AI and HPC Development Tools

July 12, 2024

New Mac developer tools will leverage Apple's homegrown chips, limiting HPC users' ability to use parallel programming frameworks from Intel or Nvidia. Apple's latest programming framework, Xcode 16, was introduced at Read more…

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to com Read more…

Shutterstock 1886124835

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical c Read more…

Shutterstock 2203611339

NSF Issues Next Solicitation and More Detail on National Quantum Virtual Laboratory

July 10, 2024

After percolating for roughly a year, NSF has issued the next solicitation for the National Quantum Virtual Lab program — this one focused on design and imple Read more…

NCSA’s SEAS Team Keeps APACE of AlphaFold2

July 9, 2024

High-performance computing (HPC) can often be challenging for researchers to use because it requires expertise in working with large datasets, scaling the softw Read more…

Anders Jensen on Europe’s Plan for AI-optimized Supercomputers, Welcoming the UK, and More

July 8, 2024

The recent ISC24 conference in Hamburg showcased LUMI and other leadership-class supercomputers co-funded by the EuroHPC Joint Undertaking (JU), including three Read more…

Generative AI to Account for 1.5% of World’s Power Consumption by 2029

July 8, 2024

Generative AI will take on a larger chunk of the world's power consumption to keep up with the hefty hardware requirements to run applications. "AI chips repres Read more…

US Senators Propose $32 Billion in Annual AI Spending, but Critics Remain Unconvinced

July 5, 2024

Senate leader, Chuck Schumer, and three colleagues want the US government to spend at least $32 billion annually by 2026 for non-defense related AI systems.  T Read more…

Point and Click HPC: High-Performance Desktops

July 3, 2024

Recently, an interesting paper appeared on Arvix called Use Cases for High-Performance Research Desktops. To be clear, the term desktop in this context does not 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…

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…

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_1687123447

Nvidia Economics: Make $5-$7 for Every $1 Spent on GPUs

June 30, 2024

Nvidia is saying that companies could make $5 to $7 for every $1 invested in GPUs over a four-year period. Customers are investing billions in new Nvidia hardwa 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…

AMD Clears Up Messy GPU Roadmap, Upgrades Chips Annually

June 3, 2024

In the world of AI, there's a desperate search for an alternative to Nvidia's GPUs, and AMD is stepping up to the plate. AMD detailed its updated GPU roadmap, w 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…

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…

Leading Solution Providers

Contributors

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…

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…

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…

IonQ Plots Path to Commercial (Quantum) Advantage

July 2, 2024

IonQ, the trapped ion quantum computing specialist, delivered a progress report last week firming up 2024/25 product goals and reviewing its technology roadmap. 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…

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…

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…

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…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire