In this special feature from The Exascale Report, Mike Bernhardt looks at our current progress towards the future milestone of supercomputing 1018 flops.
As thousands of global HPC stakeholders prepare to assemble in Denver for the 25th anniversary of the SC conference, SC13, the staff of The Exascale Report is thrilled to present an upbeat discussion on the topic of Exascale.
Over the past three and a half years, we’ve tried to present a balanced perspective of community opinion by covering both the positive and negative discussions of exascale.
We’ve praised the worthwhile efforts and creative spirit of many community leaders, while also stepping up to ‘tell it like it is’ when politics or lack of leadership were threatening to bring an entire ecosystem to a dysfunctional halt.
In hindsight, perhaps we all expected too much.
In The Exascale Report article from July 2010 titled DARPA and UHPC: Jump Starting a Revolution, (written by John Kirkley), we noted that DARPA was looking for a 1,000X increase in computing capabilities by 2015. In that same issue, we had numerous references to the widening chasm between two camps on what it would take to get us there. With much praise from our readers, The Exascale Report placed a spotlight on the discussion of Evolution vs. Revolution.
That was then. This is now.
Here we are, almost three and a half years later, and it doesn’t look like we’ll hit that 1,000x performance improvement milestone by the year 2015. Most industry experts feel that 2020 or 2022 is a more realistic timeframe. And truth be told, there are still arguments, albeit mostly semantic, around the discussion of a revolutionary vs. evolutionary approach being the best path to take us to exascale-level computation.
The HPC Battlefield
Flashback: 25 years of SC conferences
I remember well my first exposure to the HPC community. I was working with the VLIW company, Multiflow Computer, when I attended Supercomputing ’88 in Orlando. There were only 1,400 attendees that year, (as a contrast to the size of SC today), but I was hooked on HPC from that first event. One thing that I observed back then – and every year since then, is that this has never been and likely never will be a community lacking in ideas – or opinions.
Over the past quarter century, Gallium Arsenide and CMOS prophets took strong positions, as did the opposing camps of SIMD and MIMD, Open Source and Proprietary, RISC vs. CISC, NUMA vs UMA, Vector Supercomputers vs. Killer Workstations, and Supercomputers vs. Clusters.
These battles often rolled into the technology media and became key discussion points at the SC conferences, going back to the very first keynote in 1988 when Seymour Cray’s presentation was titled, “What’s This About Gallium Arsenide?”
Progress in the HPC space has been steady and impressive, but unfortunately, has created an environment of high expectations. I think at times, it’s difficult for some of us to see and appreciate important stepping stones of progress when they aren’t wrapped in holiday paper and presented as major milestones.
The 2013 Reality Check: It’s not unanimous, but it is a majority opinion. We are making progress.
A small number of leaders in the HPC community rate our 2012-2013 progress toward exascale as poor. They cite lack of leadership, inadequate funding, and no solid direction. But based on our recent survey, this view is in the minority.
The results of our recent Exascale Progress poll tell a different story. Taking an average score from our more than 200 survey respondents, the vast majority of HPC stakeholders believe the Exascale Progress Meter has moved two points – and in the right direction.
This is indeed good news. The emerging exascale community has spoken. We are making progress.
It is worth noting that a number of respondents cited their optimism around the appointment of our new Secretary of Energy, Ernest Moniz. His support of exascale beyond some initial statements is still to be seen, but many community leaders seem to have a renewed confidence that we have a supportive voice on the President’s cabinet.
Using the SC Conference as our frame of reference, we’ve had significant challenges over the past two years as Washington politics directly impacted the productivity of our nation’s scientists and researchers – cutting budgets, eliminating travel, and creating environments that made collaboration a painful process. We’ve had concerns for the past several conferences regarding attendance at SC, but for the most part, it was unfounded. Department leaders were able to get a respectable number of key researchers and scientists to the conferences – through dogged determination and creative accounting.
Through these very difficult times, the HPC community held strong and pushed forward, driven by passion, determination, and the deep-rooted understanding that scientific discovery and economic progress can’t simply be put on a back shelf.
If the HPC community has been able to thrive, even conservatively, during the most dysfunctional times, imagine what we can do with strong leadership. Realistically, we may not be able to convince Congress or the White House to allocate the level of funding that is necessary to drive a national technology agenda at a competitive pace, but this community, the emerging exascale community, will push forward. We are making progress.
The real core of the HPC community is the people – and this is a many, many-core community.
With our first exascale progress meter back in 2011, the community rated our progress as 5.5 on a scale from 1 to 20. Most people voiced frustration aimed at funding agency leadership, along with confusion around an exascale roadmap and which path to take.
Today, that is different. While a few people still feel that way, the majority of the community is encouraged – largely because of the new DOE leadership, but also because of the tireless efforts of a few organizations and key individuals.
Our poll reflects the opinions of almost 400 users (non-vendors) throughout the HPC community. On our scale of 1-20, scores ranged from a low of 3 to a high of 13. The average came out to 9.5. That score equates with ‘good’ progress.
The general perception of our forward progress on the path to exascale has improved 4 points since our first poll in 2011.
When we conducted our recent poll to determine community perception of exascale progress, we also asked our readers if any organizations or individuals stood out and deserved to be recognized for their leadership and contribution to keeping the HPC community strong and moving forward. We solicited nominations to recognize one organization or one individual as a Vanguard of HPC.
The response was extremely positive. We received 25 nominations from almost 400 people. Of those 25 nominations, six individuals are in a very tight race to see who will be the recipient of the inaugural HPC Vanguard Award.
These six have already risen to the status of legends in HPC, but only one, as determined by a vote of HPC community peers and colleagues will be selected for the recognition of HPC Vanguard.
The Exascale Report will be announcing the winner of the first HPC Vanguard Award at SC13 on Monday, November 18th.
We expect to see a prototype exascale operating system sometime in 2016. We have new leadership at the helm of the U.S. Department of Energy. There is renewed enthusiasm and hope among many HPC stakeholders, and we anticipate a very exciting SC13 conference.
So, as thousands of us gather at the Colorado Convention Center and the many fine watering holes of Denver to celebrate the 25th Anniversary of the SC conference, I feel there is a lot for which we should be thankful. We are making progress.
The six finalists the HPC Vanguard Award – individuals who have been nominated by a community vote for having played a key role in moving the exascale progress meter forward, are:
• Pete Beckman, Argonne
• Bill Dally, NVIDIA
• Jack Dongarra, University of Tennessee
• Alan Gara, Intel Corporation
• Bill Gropp, University of Illinois (Champaign-Urbana)
• Thomas Sterling, Indiana University
Courtesy of The Exascale Report, this story appears in The Print ‘n Fly Guide to SC13 in Denver. We designed this 24-page Guide to be an in-flight magazine custom tailored for your journey to the Mile-High city at SC13.
- The Fast Data Imperative – an interview with Mellanox CTO Michael Kagan
- Interview: Conference Chair Bill Gropp on the 25th Anniversary of SC
- Technical Marketing in the Age of HPC
- Local’s Guide to Restaurants in Denver
- Move Over, HPCers: Another Wave of Immigrants is Hitting Your Shores by IDC’s Steve Conway
- SC 25th Anniversary – The Complete History of Keynotes
- Local’s Guide to Bars and Entertainment in Denver
- Brian Sparks and Scot Schultz on the Secrets of Technical Marketing
- An Update to the Exascale Progress Meter
- City Guide: SC13 Comes to Denver, Colorado
- Sci-Fi Original: The Observer Effect by Rich Brueckner