Rich is spending the week at ETH Zürich attending the Platform for Advanced Scientific Computing (PASC) Conference. Many thanks to two of the co-organizers, Markus Püschel and Olaf Shank, for being great hosts!
If you are interested in getting a copy of Rich’s talk slides + related material, see: hpcgarage.org/pasc14
The view from Villa Hatt at ETH Zürich is not too shabby!
Piyush, Jee, and Rich are in Portland representing the lab at the SIAM Parallel Processing ’14 meeting. If you are also there, please drop by our various events! List below; resources (papers, slides) are here: [link].
- What Can the Roofline Model of Energy Tell Us About How to Build the Next Supercomputer? byJee Choi. [Poster, Tue Feb 18, 6-8pm in Salon E.]
- Self-stabilizing iterative solvers, by Piyush Sao. [Talk, Wed Feb 19, 11-11:20am in Salon F.]
- How much (execution) time, energy, and power will my algorithm need? by Rich Vuduc. [Talk (video), Wed Feb 19, 1:45-2:30pm in Salon F.]
- Frontiers of Performance Analysis and Tools, a minisymposium co-organized by Al Maloney (U Oregon) and Rich Vuduc. [Part 1 of 2, 2:40-4:20pm in Salon A; Part 2 of 2, 4:50-6:20 in Salon A]
Rich just returned from the Workshop on Visualization and Analysis of Performance of Large-scale Software (VAPLS), co-located with IEEE Vis. The workshop featured a number of very compelling use of visualization to aid performance analysis, which in the long run will help make performance engineering more productive, accessible, and best of all, fun!
Rich’s talk materials appear here: [www]
An example of abstractly visualizing communication volume on an AMR tree, taken from one of the other presenters (see VAPLS’13 website).
This week Rich has the great fortune of attending Dagstuhl Seminar 13401: Automatic Application Autotuning for HPC Architectures. He’ll be summarizing Jee’s work (with critical assists by Marat) on the energy archline model. For his slides and pointers to relevant materials, look here: [www]
“Classical” wing of Schloss Dagstuhl
We’re in Seattle this week attending the US DOE Workshop on Modeling and Simulation of Exascale Systems and Applications — “ModSim’13” [www]. A copy of Rich’s talk and pointers to supplemental materials is available here: hpcgarage.org/modsim13
Aparna and Jee are at SIAM CSE 2013 in Boston this week, giving talks the fast multipole method and the energy roofline, respectively. (UPDATE: Click on the title for Jee’s talk to access the PDF slides.)
- Aparna Chandramowlishwaran. “The FMM in CnC.” Talk at SIAM CSE, February 27, 2013, 4:30-4:55pm, Lewis Conference Room, as part of MS152: Applications and New Developments in Fast Multipole and Tree-based Methods.
- Jee Whan Choi. “A roofline model of energy (and what it implies for algorithm design).” Talk at SIAM CSE, March 1, 2013, 10:30-10:55am, Hancock Room, as part of MS231: Challenges of Energy-aware Scientific Computing – Part I of II. (See also the organizers’ website.)
Rooflines in time vs. “archlines” in energy
Check out our new technical report, which presents a thought-experiment on the question of whether engineering an algorithm to optimize time differs from doing so with respect to energy (e.g., Joules).
- Jee Whan Choi, Richard Vuduc. “A roofline model of energy.” Technical report no. GT-CSE-12-01, Georgia Institute of Technology, School of Computational Science and Engineering, Atlanta, GA, USA, December 2012. [PDF (452 KiB)]
We describe an energy-based analogue of the time-based roofline model of Williams, Waterman, and Patterson (Comm. ACM, 2009). Our goal is to explain—in simple, analytic terms accessible to algorithm designers and performance tuners—how the time, energy, and power to execute an algorithm relate. The model considers an algorithm in terms of operations, concurrency, and memory traffic; and a machine in terms of the time and energy costs per operation or per word of communica- tion. We confirm the basic form of the model experimentally. From this model, we suggest under what conditions we ought to expect an algorithmic time-energy trade-off, and show how algorithm properties may help inform power management.
Subject area: High-Performance Computing
Keywords: performance analysis; power and energy modeling; computational intensity; machine balance; roofline model
Note (January 14, 2013): The above link to the PDF now points to the official version on the Georgia Tech Library’s report repository. If you downloaded an earlier version (dated December 24, 2012), you may wish to re-download this newer version. (The only difference is the addition of Appendix B, made on December 31, 2012.)
Our Morgan & Claypool Synthesis Lecture on GPGPU performance analysis is now available! (Most libraries should have proxy access; the book is $30 otherwise, DRM-free.)
I (Rich) am giving a talk today [PDF slides] at the International Conference on Parallel Processing and Applied Mathematics (PPAM), which is being held in the beautiful and historic town of Toru?, Poland. This is a great and well-organized meeting, and I look forward to the next meeting!