Winner of NVIDIA Superhero Challenge announced

Remember the CUDA Superhero Challenge NVIDIA announced with TopCoder in August? We think contests are a great idea to bring more folks into technology in general, so we were happy to report on the idea and will follow the series of challenges as they unfold in the coming months.

The first of those challenges is now complete

nVidia logoThis first Challenge in the series of contests ran Sept. 14 through Sept. 25, and centered on image processing using GPU accelerated connected component labeling (CCL). One of the most common processing steps in applications such as real time object recognition, machine vision and many others, CCL is a simple but computationally intensive process. A series of high resolution images were provided and competitors were challenged to identify all objects and areas consisting of groups of adjacent pixels of the same color within a specified threshold, with submitted implementations scored on the basis of correctness and total overall performance.

Yesterday NVIDIA and TopCoder announced that Micha Riser of Switzerland won the top $2,500 prize after getting a very late start with CUDA

“I started exploring CUDA only in the second week of the contest. I had heard of it before but never got the chance to try it out myself,” said Riser, a software engineer at carrara engineering GmbH, Switzerland. “The contest was very fun, and I learned much about the CUDA framework and the computation power of the NVIDIA graphic cards as well as about the CCL problem and different ways to solve it.”

Which is the whole point of the challenge. Good for NVIDIA, and good for the parallel processing community as a whole.

Comments

  1. What is more interesting about this contest is details.

    Single threaded (contest limitation) CPU (2.33 GHz) only solution got about 10-20% lower score then CUDA Tesla solution. Using all quad cores of CPU rise score more than 3x times. CUDA suck.