I will be in Denver next week for SC19.
As usual, NAG will have a booth (#932) in the exhibition. The booth will have information and handouts on our HPC consulting services (I believe the marketing strapline is "The leading independent and international center-of-excellence in the business and technical aspects of HPC"). The NAG staff on the booth will be happy to discuss our consulting work or software tools.
However, I expect to spend no more than a few minutes near the NAG booth during the whole week. So, where can you find me and what will I be doing?
Each year I have to balance four personalities at SC:
Tuesday, 12 November 2019
Saturday, 9 November 2019
Supercomputers and Jet Engines
We all know that supercomputers are used to design jet engines.
Designing and understanding the performance characteristics of a jet engine would likely take millions of core-hours on large HPC systems.
But, a random fact thrown out over lunch at May's industry HPC leaders group meeting sparked an interesting conversation.
If a jet engine takes many megawatt-hours of supercomputing to design - how many MW does a jet engine create, and thus how many petaflops could a jet engine support if it were the power source?
The HPC leaders of GE, Boeing, ExxonMobil and others drew together their shared knowledge of aerospace and HPC - and some use of google search - to built a fun picture.
But, a random fact thrown out over lunch at May's industry HPC leaders group meeting sparked an interesting conversation.
If a jet engine takes many megawatt-hours of supercomputing to design - how many MW does a jet engine create, and thus how many petaflops could a jet engine support if it were the power source?
The HPC leaders of GE, Boeing, ExxonMobil and others drew together their shared knowledge of aerospace and HPC - and some use of google search - to built a fun picture.
Friday, 8 November 2019
SC19 Tutorials
At SC19, I will be again be leading a full day of tutorial on the business aspects of HPC: "Delivering HPC: Procurement, Cost Models, Metrics, Value, and More".
My co-presenters for SC19 will be Ingrid Barcena Roig, Branden Moore, Dairsie Latimer, and Sierra Koehler.
The tutorials is on Monday 18th November, in room 210-212 of the Denver convention center, 8:30am - 5.00pm.
Please join us to learn more about how to get investment in HPC, how to spend it wisely, and how to measure the impact.
My co-presenters for SC19 will be Ingrid Barcena Roig, Branden Moore, Dairsie Latimer, and Sierra Koehler.
The tutorials is on Monday 18th November, in room 210-212 of the Denver convention center, 8:30am - 5.00pm.
Please join us to learn more about how to get investment in HPC, how to spend it wisely, and how to measure the impact.
Guide to announcements for SC19
(Originally published on my LinkedIn profile: post link)
It's that time of year - yes, the annual fest of press releases and social media deluges in the run up to 'SC' - the primary annual supercomputing conference, held this year in Denver.
Here is a handy guide for vendor PR teams ...
[company] will be at #SC19!
Yes, along with almost everyone else in #HPC world
[company] will be highlighting products at SC19!
As above
[company] will launch new version of our current product in a slightly different shade of grey at SC19!
We had no actual news
However, the HPC centers are just as bad with "news" for the big annual #Supercomputing conference:
Wednesday, 7 November 2018
SC18 preview
I've written my customary preview of SC, which is now published at HPC Wire: https://www.hpcwire.com/2018/11/06/sc18-preview-big-in-dallas/.
Over 10,000 members of the global HPC community will gather in Dallas for the SC18 conference. Even a decent sized team will struggle to attend everything the official program has to offer. On top of this, there will be a plethora of public and private meetings outside the official program, many of which are more valuable than the official program. Plus, there will be the usual flood of press releases, social media blasts, etc.
Out of all of this, what will emerge as the key themes? What are some essential things to do/attend? Read the @hpcnotes SC18 preview to find out!
Over 10,000 members of the global HPC community will gather in Dallas for the SC18 conference. Even a decent sized team will struggle to attend everything the official program has to offer. On top of this, there will be a plethora of public and private meetings outside the official program, many of which are more valuable than the official program. Plus, there will be the usual flood of press releases, social media blasts, etc.
Out of all of this, what will emerge as the key themes? What are some essential things to do/attend? Read the @hpcnotes SC18 preview to find out!
Tuesday, 6 November 2018
SC18 Networking Receptions
Networking Receptions at SC18 Dallas [updated regularly until SC starts]
A huge part of the SC conference (or any HPC conference) is meeting people - from old friends to new contacts. Here is a curated list of networking opportunities (receptions) crowd-sourced from this twitter thread https://twitter.com/hpcnotes/status/1059437643837161474 and other sources:
Sunday 11th
- Spectrum Scale UG and happy hour, 12-4pm
- LLVM & Flang social, 6pm-9pm, Aloft, http://lists.llvm.org/pipermail/llvm-dev/2018-October/127303.html
- Intersect360, 2pm-4pm, Biergarten Restaurant, https://www.eventbrite.com/e/intersect360-research-sc18-networking-reception-tickets-51318644447
- DDN user group reception, 5pm, Old Red Museum of Dallas County History & Culture, following user group meeting https://www.ddn.com/company/events/user-group-sc/
- SC official opening gala, 7pm-9pm, https://sc18.supercomputing.org/presentation/?id=pec122&sess=sess284
- Beowulf Bash, 9pm, Eddie Dean's Ranch, https://beowulfbash.com/
- Women in HPC, 6.30pm-9pm, Cafe Herrera on Lamar (at the Omni Dallas), https://www.eventbrite.co.uk/e/whpcsc18-networking-and-careers-reception-tickets-51886141847
- IBM
- Cray
- DDN, 7pm, Reunion Tower, tickets via DDN booth #3213
- Nimbix, 6pm-10pm, d.e.c. on dragon st, https://www.eventbrite.com/e/nimbix-sc18-lounge-party-registration-50540352555
- OpenACC, 7pm-10pm, Eddie Dean's Ranch, https://www.eventbrite.com/e/9th-openacc-user-group-meeting-sc2018-tickets-50250658071
- Mellanox, 6.30pm, Sheraton, http://www.mellanox.com/sc18/event.php?ls=social-tw&lsd=11.5.18
- Boston Limited, 7.30pm Bob's Steak & Chop House, https://www.boston.co.uk/events/2018/sc18.aspx [pre-booking required via booth #3255]
Tweet me @hpcnotes using hashtag #SC18 to add your reception to this list!
Monday, 5 November 2018
SC18 Tutorials
At SC18, I will be leading two tutorials, along with my long-time co-presenter Owen Thomas and new co-presenter for SC18, Ingrid Barcena Roig.
Please join us to learn more about how to get investment in HPC, how to spend it wisely, and how to measure the impact.
- 8:30am - noon : "The Business of HPC: TCO, value, metrics, and more ..."
- 1.30pm - 5.00pm : "Procurement and Commissioning of HPC Systems"
Both tutorials are on Monday 12th November, in room C140 of the Dallas convention center.
Please join us to learn more about how to get investment in HPC, how to spend it wisely, and how to measure the impact.
Saturday, 23 June 2018
A useful reading list for travelling to ISC18
Travelling to Frankfurt for ISC? Need to feed your HPC thirst while on planes, trains, or in hotel rooms? Here is my pick of things to download and read so that you are fully informed when you start ISC:
- (Obviously!) The @hpcnotes ISC18 preview at HPC Wire
- The official welcome to ISC18 by the organizers, via InsideHPC
- An article on the race between USA and China (and others) to get to the top of the Top500 at WiredUK "Why the US and China's brutal supercomputer war matters" and my follow up thoughts: https://www.hpcnotes.com/2018/06/does-it-matter-whether-usa-china-eu-or.html
- Sandia to 2.3 deploy petaflops Cavium TX2 ARM supercomputer at Top500.org
- Details on Japan's post-K exascale supercomputer from The Next Platform
- [added Sat afternoon] Top500 interviews Thomas Sterling on Exascale, Chinese HPC, ML and non-von-Neumann
- [more to follow during the weekend ...
See you in Frankfurt!
Andrew / @hpcnotes
Wednesday, 20 June 2018
NAG-TACC HPC Leadership Institute 2018
Just taken over a HPC management or leadership role? Or hoping to soon? Or know someone who could grow into those roles? Or been a HPC director for years but value ongoing personal development?
The HPC Leadership Institute is a partnership between Numerical Algorithms Group (NAG) and Texas Advanced Computing Center (TACC) to deliver training on the business aspects of High Performance Computing. The training covers strategy, total cost of ownership (TCO), cloud vs on-site, supercomputer procurement, governance, user services, and much more.
The 2018 course will be held in Austin TX September 11-13. Learn more and register now at:
https://www.tacc.utexas.edu/education/institutes/hpc-leadership-institute
The HPC Leadership Institute is a partnership between Numerical Algorithms Group (NAG) and Texas Advanced Computing Center (TACC) to deliver training on the business aspects of High Performance Computing. The training covers strategy, total cost of ownership (TCO), cloud vs on-site, supercomputer procurement, governance, user services, and much more.
The 2018 course will be held in Austin TX September 11-13. Learn more and register now at:
https://www.tacc.utexas.edu/education/institutes/hpc-leadership-institute
Does it matter whether USA, China, EU, or someone else has the biggest supercomputer?
Much fuss will be made over the ORNL's new Summit supercomputer at the ISC18 event next week - in particular the fact that it means the USA replaces China as the home of world's fastest supercomputer according to www.top500.org. This brings the usual question as to whether it really matters which country has the biggest supercomputer.
Having a supercomputer 20%, or even 2x, faster than a competitor isn’t critical on its own, because it is possible to make up 20% or 2x actual competitive capability through better software, better people, or better service delivery practices.
However, a 10x faster supercomputer would be an issue, because that would typically reflect a political commitment to High Performance Computing (HPC) involving hardware and software and people - and so could mean potential capability dominance.
Of course, if you had the 2x slower supercomputer without investing in people/software/practices to make up the difference, then that would be a meaningful competitive gap and would matter.
Read more in this article at WiredUK: "Why the US and China's brutal supercomputer war matters"
Having a supercomputer 20%, or even 2x, faster than a competitor isn’t critical on its own, because it is possible to make up 20% or 2x actual competitive capability through better software, better people, or better service delivery practices.
However, a 10x faster supercomputer would be an issue, because that would typically reflect a political commitment to High Performance Computing (HPC) involving hardware and software and people - and so could mean potential capability dominance.
Of course, if you had the 2x slower supercomputer without investing in people/software/practices to make up the difference, then that would be a meaningful competitive gap and would matter.
Read more in this article at WiredUK: "Why the US and China's brutal supercomputer war matters"
Wednesday, 22 November 2017
Benchmarking HPC systems
At SC17, we celebrated the 50th edition of the Top500 list. With nearly 25,000 list positions published over 25 years, the Top500 is an incredibly rich database of consistently measured performance data with associated system configurations, sites, vendors, etc. Each SC and ISC, the Top500 feeds community gossip, serious debate, the HPC media, and ambitious imaginations of HPC marketing departments. Central to the Top500 list is the infamous HPL benchmark.
Benchmarks are used to answer questions such as (naively posed): “How fast is this supercomputer?”, “How fast is my code?”, “How does my code scale?”, “Which system/processor is faster?”.
In the context of HPC, benchmarking means the collection of quantifiable data on the speed, time, scalability, efficiency, or similar characteristics of a specific combination of hardware, software, configuration, and dataset. In practice, this means running well-understood test case(s) on various HPC platforms/configurations under specified conditions or rules (for consistency) and recording appropriate data (e.g., time to completion).
These test cases may be full application codes, or subsets of those codes with representative performance behaviour, or standard benchmarks. HPL falls into the latter category, although for some applications it could fall into the second category too. In fact, this is the heart of the debate over the continued relevance of the HPL benchmark for building the Top500 list: how many real-world applications does it provide a meaningful performance guide for? But, even moving away from HPL to “user codes”, selecting a set of benchmark codes is as much a political choice (e.g., reflecting stakeholders) as it is a technical choice.
Benchmarks are used to answer questions such as (naively posed): “How fast is this supercomputer?”, “How fast is my code?”, “How does my code scale?”, “Which system/processor is faster?”.
In the context of HPC, benchmarking means the collection of quantifiable data on the speed, time, scalability, efficiency, or similar characteristics of a specific combination of hardware, software, configuration, and dataset. In practice, this means running well-understood test case(s) on various HPC platforms/configurations under specified conditions or rules (for consistency) and recording appropriate data (e.g., time to completion).
These test cases may be full application codes, or subsets of those codes with representative performance behaviour, or standard benchmarks. HPL falls into the latter category, although for some applications it could fall into the second category too. In fact, this is the heart of the debate over the continued relevance of the HPL benchmark for building the Top500 list: how many real-world applications does it provide a meaningful performance guide for? But, even moving away from HPL to “user codes”, selecting a set of benchmark codes is as much a political choice (e.g., reflecting stakeholders) as it is a technical choice.
Subscribe to:
Posts (Atom)