• Program

• Instructors

• Lectures

• Resources

• Participants

• Photo Gallery

• Evaluation

• Announcement

• FAQ

• Laptop Reqs

• Home

GPU Programming for Molecular Modeling Workshop - Urbana, Aug. 6-8, 2010

Lecture Evaluation of the GPU Programming for Molecular Modeling Workshop held at the University of Illinois, Urbana

August 6-8, 2010

Questionnaire: Gila Budescu, TCB group, UIUC, and modified by David Brandon, TCB Group, UIUC
Analysis and report: David Brandon, TCB group, UIUC

The Theoretical and Computational Biophysics Group (TCBG), an NIH Resource for Macromolecular Modeling and Bioinformatics is headed by Klaus Schulten, and CO-PIs Z. Luthey-Schulten,  L. Kale, A. Aksimentiev, and E. Tajkhorshid.  As part of its outreach, the Resource offers workshops to introduce and transfer its programs and technological solutions to the biomedical community.  The Resource organized a 2.25 day (August 6-8, 2010) workshop held at the Beckman Institute for Advanced Science and Technology on the campus of the University of Illinois at Urbana-Champaign.

Workshop lectures were provided by TCBG development staff members John Stone and James Phillips, and postdoctoral associates David Hardy and Andrew Magis. Development staff member Kirby Vandivort assisted with providing consultations to workshop participants on their programming issues. The program of the workshop consisted of lectures, participant presentations, and a programming project laboratory. Participants worked on their own laptops during the workshop. At the end of each day, participants were asked to complete evaluation forms rating the relevance of lectures, and to provide evaluative comments about the lectures. The lecture evaluation form can be found here.

Lecture evaluation are comprised of two elements, 1) the proportion rating the relevance of the lecture or tutorial as highly relevant (i.e. 'very good' + 'excellent' ratings; see Table 1: Summary of Relevance Statistics below), and 2) select comments considered illustrative of respondent opinion.  As is frequently the case with surveys, not all respondents answered all questions; the number of responses for the relevance ratings (r=) and comments (c=) are listed next to the name of each lecture and tutorial summary, e.g. (N: r=12, c=12).

Some issues to consider when reading the comments:

• Written comments, particularly when comments are extreme in one direction or the other, tend to stick in one's head more so than statistics that may present a more accurate summary of opinion.
• Attendees appear to have been somewhat heterogeneous in computational background, training, interests, and experience; so, for any lecture or tutorial there was likely always someone new to the topic who needed more time, help and explanation, and at the same time someone very experienced who wanted more breadth and/or depth on the topic.

Day 1 Lecture: GPU Particle-Grid Algorithms: Electrostatics (N: r=12, c=12)

All participants (100%) rated the lecture topics as ver good or excellent.

• Great lecture. Sample code really helped to see programming ideas in action, and the presentation of source code from improvement to improvement of the previous was a nice way of comparing different approaches and see the strengths/ weaknesses of each.
• It was a good mix of detail and science.

Day 1 Lecture: GPU Particle-Grid Algorithms: Non-bonded Force Calculation (N: r=12, c=12)

A majority of participants (67%), found the lecture content relevant.

• Very good also! Lots of nice tricks (struct, union, bank conflicts). Talk should follow this pattern of: tricks we used to solve this problem and why. Naïve implementation is easy, but the goods come from doing a bit more to squeeze out performance.
• Also, very clear. I think the talk could benefit from a side-by-side comparison of the NAMD-GPU implementation with the NAMD – CPU/MPI implementation. Not just in terms of results but in algorithm design.

Day 1 Lecture: GPU Histogramming: Radial Distribution Functions (N: r=11, c=12)

A majority of participants (67%), found the lecture content relevant.

• Very good! Especially liked the trick of offsetting the bits + 5. And the explanation about atomic updates in older us newer models. Oh, the volatile function was great!
• John is definitely a seasoned lecturer. He is clear, articulate and quite persuasive. I enjoyed his lecture, although I would have enjoyed a little more intro. And less code.

Day 1 Lecture: CUDA Algorithms for Stochastic Simulation of Biochemical Reactions (N: r=11, c=11)

Nearly all participants (91%) perceived the relevance of the lecture content as very good or excellent.

• Very Good. A very macro-level applied project brought the previous lectures back to base.
• Very interesting application GPU programming to systems modeling.

Day 2 Lecture: Single-Node Multi-GPU Algorithms: Molecular Orbitals (N: r=12, c=12)

All participants (100%) rated the lecture topics as ver good or excellent.

• Great. Nice to see the limitation of the hardware intercommunications. Good to know limitations of the hardware on laptops.
• This lecture gave me some idea about QM. Very interesting.

Day 2 Lecture: Molecular Dynamics on GPU Clusters (N: r=12, c=12)

All participants (100%) rated the lecture topics as ver good or excellent.

• Good talk; appreciate hearing some history about CUDA and NAMD.
• Great talk info on bus speed relative to IB, interconnect was great!

The complete set of comments is available by e-mailing brandon@ks.uiuc.edu.