Just before the action gets going later this month at the GPU Technology Conference (GTC) in sunny San Jose, Calif., a new two-day event will kick off focused on cutting-edge academic work being done on advancing GPU computing technologies.
The inaugural Innovative Parallel Computing Conference, known as InPar, will take place May 13-14, serving as a venue for the presentation of first-tier papers focused on advancements in GPU computing.
The purpose of the event is to highlight not only top work in the field but also establish and recognize collaborations between domain scientists and computer scientists.
The problem
Today researchers are working hard to push the frontiers of "heterogeneous" systems that rely on both latency-optimized CPUs and throughput-optimized GPUs. While programming CPUs has been widely studied over the past 50 years, GPUs and similar technologies are opening up exciting new research related to programming the GPUs and programming heterogeneous systems that use both CPUs and GPUs.
The research being done on these types of systems fall into two broad categories: on one side, computer scientists are exploring the foundations of innovative parallel computing (architecture, programming models, algorithms, etc.), on the other side, we see domain scientists exploring the application of innovative parallel computing to a particular scientific problem.
When GPU computing researchers work with chemists to accelerate molecular dynamics (MD) simulations 100x, the computer scientists will get little academic credit when the work gets publish in the Journal of Organic Chemistry. The readership often won't care whether the results were calculated on a GPU or an abacus as long as the chemistry is sound.
Likewise, the chemists involved in the work will get little notoriety from publishing a paper in the International Symposium on Computer Architecture. The conference reviewers care little for a hundredfold speedup on MD simulation unless they see fundamental advances in the actual computer science techniques used.
The solution
InPar aims to bring together researchers working to advance both the foundations and applications of GPU computing. Johns Hopkins University's Alexander Szalay will deliver the conference' opening keynote, setting the stage for a couple days worth of in-depth discussions about the foundations, applications and optimizations involved in GPU computing.
There's still time to register for InPar 2012. Grab your conference pass at the InPar registration page.
No comments:
Post a Comment