News

Universidad Técnica Federico Santa María has officially been named a CUDA Teaching Center by NVIDIA, the world leader in visual and high-performance computing. The institution is the first in Chile to obtain this honor, and it is among the first CUDA Teaching Centers in Latin America.

CUDA is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of the graphics processing unit (GPU). CUDA Teaching Centers are recognized institutions that have integrated GPU computing techniques into their mainstream computer programming curriculum.

This distinction is an important milestone in the research and teaching activities on GPU computing at UTFSM, and it has been received at a very exciting time for the institution.

As part of the CUDA Teaching Center program, NVIDIA will provide the Informatics Department with donated GPUs to help upgrade a teaching laboratory. With this infrastructure, UTFSM will strengthen its position as the Chilean leading center in GPU computing.

On the teaching track, the Informatics Department has committed to including GPU programming in CUDA as a permanent part of its curriculum. UTFSM has already hosted other GPU-related training events, the most important having been the PASI school on massive parallelism held last summer. For this school, more than 80 students from all around the world attended a series of lectures given by world-class researchers on High-Performance Computing. In August 2011, CTI-HPC also hosted an intensive 4-week course on CUDA, attended by 30 participants from the university community who were enthusiastic about incorporating CUDA to their own work.

We are very pleased to announce that— thanks to new funding secured by the PI, Prof. Lorena Barba— 16 PASI fellows will have travel support to attend the top event of the HPC world: the Supercomputing Conference 2011, being held in Seattle, WA.

This funding is made available via NSF award OISE 1143988, to hold a workshop as a follow-up to tha Pan-American Advanced Studies Institute (PASI) held in Valparaiso, Chile in January 2011. The primary goal of this workshop is to reinforce the network created during the PASI, reconnecting young scientists with senior researchers (PASI lecturers) and their peers, and, ultimately, to stimulate new collaborations.

In addition, being co-located with the top event in the field of high-performance computing, this PASI follow-up will be invaluable for these participants to enhance their careers and feed their enthusiasm for continuing their work in this field.

Special social occasion

Prof Ridgway Scott, one of the PASI lecturers, will open the doors of his house in Seattle to receive the PASI fellows and attending lecturers for a lovely party of nice food and conversation. The party will be held on Friday, Nov. 18, 6pm. We are grateful to Prof Scott for his hospitality. We also thank the students who formed a committee to help with the logistics of this party: Anshu Arya, Ben Payne, Andrew Reiter, and Jeff Stuart.

The students who receive travel funding through this initiative are:

Other PASI fellows attending SC11 with own funding who will also join our PASI get-together are:

See the News announcement (in Spanish) of the course at the host university website.

Find lecture slides online and follow the links for recorded lectures on iTunes U.

The intense two weeks of the PASI in January 2011 covered a broad range of topics, from basics of CUDA programming, parallel linear algebra and fast N-body algorithms, to a selection of computationally intensive applications.

The institute generated such great interest locally, that the HPC group at Universidad Técnica Federico Santa María (UTFSM) was motivated to organize a follow-on training event. They invited PASI organizer, Prof. Lorena Barba, to lead such an effort.

The post-PASI training is taking place from August 8 to September 3rd, and will cover the basics of the GPU architecture and CUDA programming, use of libraries such as CUSP and Thrust, and applications.

Christopher Cooper, PhD candidate at Boston University under Prof. Barba's supervision.

The series of 10 lectures and 8 hands-on lab sessions is being instructed by Prof. Barba’s PhD student Christopher Cooper, who is a graduate of UTFSM in Mechanical Engineering. He first visited Prof. Barba at University of Bristol in 2008 as a fellow of the EU-funded project “Scientific Computing Advanced Training”, and then applied to Boston University for PhD studies, which he started in January 2010. Christopher is a recipient of Becas Chile, awarded by CONICYT.

This new one-month CUDA course at UTFSM has garnered tremendous interest. The maximum capacity of 25 participants was quickly reached, and email messages continued pouring from interested students up until the first day of classes; unfortunately, there were not enough places to accommodate everyone. The limited capacity is dictated by the hands-on lab sessions, which are carried out with access to a small cluster locally with 8 GPUs, and with guest access to Boston University’s 32-GPU bungee cluster.

Participants in the course come from various departments at UTFSM, from nearby Universidad de Valparaíso and Universidad Católica de Valparaíso, and even as far as Universidad de Concepción and the ALMA observatory, in Atacama.

A detailed syllabus of this intense course is available for each week of activities (including lecture slides and references). Note that the course is being instructed in Spanish, but the lecture slides are in English.

Screencasts are being uploaded shortly after each lecture, using the PASI collection of media on iTunes U.

TADMAC stands for “Tecnologías de Alto Desempeño Aplicadas a la Modelación Matemática y Computacional”, meaning: High-performance technologies applied to mathematical and computational modeling. This is a new seminar series created to cover in greater depth the topics presented at the PASI in Valparaíso, and to apply them to the research topics of interest to the participants at Mexico’s national university, UNAM.

A group of Latin American participants at PASI, Valparaíso, with Dr Luis Miguel de la Cruz fourth from right.

Dr Luis Miguel de la Cruz, faculty member at UNAM’s Institute of Geophysics, is one of the organizers of this new seminar series, which runs weekly from March to June 2011. He explained to Prof Lorena Barba, PASI lead organizer, that before PASI the various participants from Mexico were working in isolation on topics involving parallel computing. Attending PASI created the opportunity for them to learn about the work each of them was carrying out, in addition to learning from leaders of the international scientific computing community about the cutting-edge tools being used for massively parallel computing in different areas of application.

During PASI, Luis Miguel and the other Mexican participants reflected on the need to disseminate back at UNAM the various topics covered during the institute. Thus, they set out to, each of them, in collaboration with other researchers at UNAM, study in more depth one of these topics and prepare for its dissemination via a seminar at TADMAC. In this first stage of the seminar series, each speaker is introducing the topic and developing simple examples. In a second stage starting August 2011, they will aim to show how the tools and techniques are being used in their own research projects.

They have started a TADMAC Google group, which currently counts 79 members from several educational institutions in Mexico. They also were inspired from PASI to share online all the educational materials developed for the seminars, including videos of seminars. All of the PASI participants from Mexico are currently attending the TADMAC seminars, which draw between 30 and 50 people each week.

Other participants of PASI contributing to the TADMAC seminars include:

• Emilio Zavala Sosa, MSc student in Earth Sciences—he is working on a problem involving Enhanced Oil Recovery (EOR) applying Newton’s method for linearization of the equations. After PASI, he decided to make use of Python tools, and also FEniCS to solve his problems in parallel with PETSc.
• Daniel Monsivais Velázquez, MSc student in Computer Science and Engineering—he is carrying out a comparative study with CUDA, OpenMP and PETSc for the solution of two-phase flow problems. He currently is refining his measurements using new knowledge acquired at PASI.
• Saúl Piedra González, MSc student in Engineering (Temixco, Morelos)—he is working with the front-tracking method for simulation of bubbles in fluids. He is now using CUSP and Thrust to solve his equations on GPU hardware.
• Michel Rivero Corona, PhD student in Engineering (Temixco, Morelos)—he works in magnetohydrodynamics, and aims to carry out numerical simulations to compare with experimental results. He is also using CUSP and Thrust in his simulation codes.
• Antonio Muñoz Flores, MSc student in Computer Science and Engineering (Cuaititlán)—he is working with the lattice Boltzmann method (LBM) in fluid dynamics simulation. He is currently using OpenCL to parallelize  his codes.
• Ursula Iturrarán Viveros, associate professor at the Faculty of Science (Mexico City)—she is working in computational fluid dynamics and earthquake simulation, and currently is developing some of her methods with CUSP and Thrust.

Luis Miguel himself is directing a group of experts to develop simulation for oil reservoirs. They are developing (i) a finite volume method, (ii) a streamline-based simulator, and (iii) a finite element method code using FEniCS. In all cases, they are using PETSc and CUDA for the solution of the systems of equations. Luis Miguel is also working on problems of fluid mixing, for which he develops software using generic programming techniques, and is creating CUDA kernels based on CUSP and Thrust.

• Poster of the TADMAC seminar series, listing the topics of the first phase, March-June 2011.

Jeff Stuart in a classic tourist photo spot at Viña del Mar: the flower clock.

David Ketcheson and Jefferey Stuart—both PASI 2011 participants who met while in Chile—have formed a team of Applied Mathematicians and Computer Scientists from KAUST, UC Davis, University of Washington, and the American University of Beirut, to work on the explicit solution of hyperbolic partial differential equations on GPUs.

Specifically, the team has started work on porting compute-intensive Riemann solvers to CUDA for use in the package PyCLAW and for standalone use. They are preparing to submit their work as a paper to InPar 2011 and as a poster to SC 2011.

David Ketcheson and family, posing in front of the Naval Museum in Viña del Mar.

Jefferey Stuart is a PhD student at University of California, Davis, under the supervision of Prof. John Owens. He has focused almost exclusively on using GPUs in his graduate work, notably creating a MapReduce implementation on a cluster of GPUs. Jeff explicitly wanted to attend the PASI in Chile to network with other researchers with the purpose of starting new collaborations. It is already paying off for him to have participated!

David Ketcheson is an Assistant Professor of Applied Mathematics at King Abdullah University of Science and Technology since July 2009. Previously, he was a graduate student at the University of Washington, and a US Department of Energy Computational Science Graduate Fellow. His doctoral supervisor was renowned Professor Randall J. LeVeque.

The new collaboration between David and Jeff was instigated by Aron Ahmadia, a computational scientist from the KAUST Supercomputing Lab, who met Jeff at the PASI and invited him to KAUST with the idea of getting Jeff and David to work together. This initiative was strengthened by the in-person meeting at KAUST, during a workshop that David says was in great part organized during PASI—being in Valparaiso together with Matthew Knepley (PASI lecturer) and Aron Ahmadia (participant) allowed them to plan the event.

This is the type of ripple effect that a successful scientific gathering aims to have. We are grateful to David, Jeff, Matt and Aron for coming to Chile for this PASI and helping make it a success.

Screencasts

We are proud to present full-length screencasts of the lectures held during the PASI in Valparaíso. These consist of video of the speaker’s computer screen (slides, animations, etc.), with the audio of the lecture room. Most of these videos captured the sound with the built-in microphone of the presenter’s laptop.

Thanks to Anush Krishnan (Prof. Barba’s PhD student) for his hard work for the production of these screencasts.

• At this time, all the screencasts are available via a course on iTunes U.
• We are also adding the videos to BU’s YouTube channel (for non-iTunes users).

Discussion Group

A Google group was created for informal exchanges among students and other participants to the PASI in Chile. We hope that this can continue to be active after the event, if only for occasional announcements of interest to the group that met in Valparaiso for the two intense PASI weeks.

Universidad Técnica Federico Santa María, campus overlooking the ocean.

The host institution, Universidad Técnica Federico Santa María (UTSFM), has announced local student support, to help fray accommodation costs in a local bed & breakfast, and lunches during the institute.  This will help students from other cities in Chile to be able to participate in this unique educational event.

The Center for Technological Innovation in High-Performance Computing, CTI-HPC, based at UTFSM, is receiving online registrations, with requests for support.

There are only 10 supported spaces, so act soon!

The support will consist of $48.000 CLP (chilean pesos) for lodging (12 nights at a local B&B or pensión), and $24.000 CLP for lunches.

For additional information, please contact the local organizers on contact@hpc.cl

Additional funding for this PASI

The Scientific and Technological Center of Valparaíso, CCTVal, has awarded supplementary funding of almost US$10,000 to offer hosting (labs, AV equipment and staff, coffee breaks, speaker recognition dinner) and local student support.

At this time, we are able to announce the applicants from Latin America that we are able to fund to attend this PASI, thanks to NSF award OISE-1036435.  We would like to extend to all the interested students who cannot be offered support at this time our sincerest expression of regret.  Hopefully, some of you may be able to find support from another source to attend. If this were the case, we will be able to offer a waiver of the registration fees for these students/postdocs.

The selected applicants for a travel grant are listed below. The travel grant consists of US$1000 towards air fare from Mexico and US$300 towards air fare from Argentina, plus a per-diem allowance of $120 capped at $1200.  Please note that these travel grants are capped, which means that if your travel cost were to be higher than the grant amount, you are expected to obtain supplementary funding from your supervisor or another source.

We received 68 eligible applications for travel grants from students and postdocs from the US and Canada.  This made the exercise of selecting the awardees for these travel grants epecially difficult, and unfortunately means that many highly qualified applicants will not receive a grant.  We would like to extend to all the interested students who cannot be offered support at this time our sincerest expression of regret.  Hopefully, some of you may be able to find support from another source to attend. If this were the case, we will be able to offer a waiver of the registration fees for these students/postdocs.

The selected applicants for a travel grant are listed below. The travel grant consists of US$1000 towards air fare, and a per-diem allowance of $120 capped at $1200 (in other words, the value of the grant is $2200).

Two graduate students who have recently been awarded an Nvidia fellowship will be co-sponsored by the company to attend PASI.  They are Henry Cook, from University of California Berkeley, and Albert Sidelnik, from University of Illinois at Urbana-Champaign.

Each of them will receive a 50% travel grant from Professor Barba’s NSF award, and an additional $1000 from Nvidia towards their airfare.  Both students will contribute to PASI by making a presentation about their research and the activities going on at their labs. Here is a brief introduction to the two first student participants that we can today announce:

Henry Cook

PhD student, Dept. Computer Science, University of California Berkeley (advisors: David Patterson and Krste Asanovic)

Henry has a BSc in computer science (2007) from University of Virginia, where he worked under Prof. Kevin Skadron, and an MSc from UC Berkely (2009). He lists as his main research interests: manycore chip architectures, software-managed memory hierarchies, composable abstractions of data locality, design space search and optimization.

Currently, Henry is focusing on creating “programmable” software-managed memory hierarchies, including mechanisms that allow for software control of data placement, movement, coherence and partition-able capacity.  He’s also studying auto-tuning libraries and frameworks that can take a high-level description of the computation and emit software control instructions (e.g. Sequuoia).  Such frameworks meed knowledge of the structure of computations from domain experts, thus Henri will use his PASI experience to familiarize himself with a wider array of HPC applications.

Henry did an internship at Nvidia Research this summer and is a 2010 recipient of the Nvidia Fellowship.  He is co-author of four CS conference papers, including the 2010 International Symposium on Computer Architecture, and the ACM/IEEE Conference on Design Automation.

Albert Sidelnik

PhD student, Dept. Computer Science, University of Illinois Urbana-Champaign (advisors: David Padua and Maria Garzaran)

Albert has a BSc in computer science (2001) from Rutgers University, and MA in the same field from Boston University (2005).  His main interests are developing programming models and optimization techniques for large-scale and many-core parallel systems. He is currently developing compiler optimizations and analytical models for high-level GPU programming.  His focus is in language extensions and compiler techniques for translation of programs from array languages (e.g. Chapel) to efficient GPU code.

Albert has done internships at Cray and IBM and has a few years of experience as a software engineer, working in IBM’s BlueGene. He has 5 published US patents on architecture, and is a 2010 recipient of the Nvidia Fellowship.