Quote:Hi!
The video of Bruce Allen's talk and now also the slides (which were kind of difficult to read in the video) are available for download: Boinc 2008 workshop proceedings.
The slides contain a hint on GPU app plans.
CU
BikemanYes thats the one. Any idea how far along things are?
We have some code that doesn't compile with NVidias current SDK and last time it did it didn't give correct results. Might take a week of work to get this going at all, though the speedup is not that impressive. With some help from NVidia we also developed a new more promising approach optimizing memory access, but that's not gone into actual code yet. At the very moment nobody finds the time to drive this a bit further, due to a lot of other things with higher priority. It's definitely a pending item on Oliver's and my todo-list, but I don't know when we'll find the time to get back to that. Hopefully later this year.
BM

Is there a GPU version of the app in the works?
)
Actually the first version to come out will probably be for Linux. That's the platform the code is been developed on, both on our and on NVidias side. The Mac version should also be rather easy to derive from this; how to build an App for Windows out of that code is still a mystery to me, so that will probably come out last.
BM
BM
We are still working on it.
)
We are still working on it. There is some progress, but we are not done yet. The work on an ABP1 CUDA App is currently a bit ahead of a S5R5 one, so that might come out first, but there is still a lot of work ahead. Please bear with us.
BM
BM
RE: What performance of
)
I have no idea of the MilkyWay "performance".
We have some highly experimental (and non-optimal) ABP1 CUDA Apps that need more work to fit into the BOINC framework; they have shown a speedup of 1.5 (GeForce) to 3.5 (Tesla) compared to the standard ABP1 (Linux) App. The first CUDA release probably will show similar speedup, but we'll keep working on optimization once we have it working with BOINC at all.
BM
BM