Hi Christopher,<div><br></div><div>I'm a software developer working on mass spectrometer analysis software and proteomics research software.</div><div><br></div><div>Currently I'm implementing 3 dimensional peak detection software using CUDA. The data is 300,000, by a 2000 by 200 and I'm implementing a highly optimized 3d convolution. The filters are unique for each point in the matrix. </div>
<div>The data set is time by mass by mobility.</div><div><br></div><div>Processing of this data on the CPU takes in excess of 5 hours. At these rates this commercialization of this kind of processing is not feasible. I estimate it will process in about 30 minutes on the GPU.</div>
<div><br></div><div>The trouble with CUDA is that in order to get good optimization I have to take into account all the details of the hardware and hand optimize each part. I have to partition the problem into carefully sized blocks and then measure different ways of partitioning the problem. It turns a relatively simple problem into a huge bookkeeping project tracking each little block of data.</div>
<div><br></div><div>I would be very interested in helping with any project that would improve the compiler/language support for processing on the gpu. A language that is more declarative (describing what we want to do rather than how) with a gpu-aware optimizing compiler and a processing engine to drive the processing would be great.</div>
<div><br></div><div>My next GPU project is large scale searching of ion datasets against huge protein databases. I want to implement as much of the searching on the gpu as possible to speed this up. I'm looking at thrust now but I probably want more of the implementation in the kernel so that I can load databases of masses (on the order of billions) and do several 100,000 searches against it.</div>
<div><br></div><div><br></div><div>A fermi card would be of great use to me in this processing and I would be happy to contribute to your project by writing drivers or libraries for you.</div><div><br></div><div>My background: I've been a software consultant for 16 years and was an individual contributer for 10 years before that. I have a lot of experience in embedded systems, algorithm design as well as application development.</div>
<div><br></div><div>You can see more about me at my linked in profile:</div><div><a href="http://www.linkedin.com/in/dangolick">http://www.linkedin.com/in/dangolick</a></div><div><br></div><div>Regards,</div><div> Dan</div>