======================================================================= Competition Announcement: GPUs for Genetic and Evolutionary Computation ======================================================================= 2010 Genetic and Evolutionary Computation Conference Wednesday – Sunday July 7 –11, 2010 Portland, Oregon, USA We are pleased to announce the official start of the GPU competition of GECCO-2010 with the publication of the competition rules and the scoring system. The Goal ======== This competition focuses on the applications of genetic and evolutionary computation that can maximally exploit the parallelism provided by low-cost consumer graphical cards. The competition will award the best applications both in terms of degree of parallelism obtained, in terms of overall speed-up, and in terms of programming style. Rules and Regulations ===================== Entrants must submit (1) the application sources with the instructions to compile it and (2) a two page description of the application. Submissions will be reviewed by a committee of researchers from the evolutionary computation community and from industry. Each reviewer will score the submission according to 12 criteria concerning the submitted algorithm, the speed-up it achieves, and its impact on the evolutionary computation community. The total score will be obtained as the weighted sum of the 12 separate scores. Submissions should be mailed to gecco2010@gpgpgpu.com no later than June 4th, 2010. The final scores will be announced during GECCO. Important Dates =============== Submission deadline: June 23rd 2010 Conference: July 7th-11th 2010 Scoring ======== Submissions will be reviewed by a panel of researchers from the evolutionary computation community and from industry who will score each submission according to the following criteria. Algorithm (50% of the total score) --------- - Novelty 10% Does the algorithm exploit the GPU in a novel way? (e.g., not just for fitness evaluation?) - Efficiency 10% Does the algorithm efficiently use the GPU? - GPU-side 10% How much of the algorithm is implemented GPU side? - Elegance 5% Is the algorithm simple, easy to understand? - Portability 5% Is the code parameterized for different GPU architectures and/or across vendors? - Suitability 10% Does it use features of the GPU architecture logically and to the advantage of the program? Speed (20% of the total score) ----- - Speedup 10% How much is the speed up compared to a well coded CPU version? - Resources 5% What is the resource utilization? (Ideally a program should use the 100% of the GPU). - Scalability 5% Will it scale? E.g. to new hardware, multiple GPUs, GPUs with fewer/more processors? Evolutionary Computation (30% of the total score) ------------------------ - Utility 10% Do the results benefit the EC/GA/GP community? - Practicality 10% Were the results practically obtainable without GPU acceleration? - Science 10% Is the system used to generate better quality science? For example, increasing statistical significance, increasing coverage of test cases or demonstrating greater generalization. Organizers ========== Simon Harding, Memorial University of Newfoundland, Canada David Luebke, NVIDIA Pier Luca Lanzi, Politecnico di Milano Edmondo Orlotti, NVIDIA