The big question is the fact that the future of GPU based crunching at WCG is not clear at all for me.
We have no information about future GPU based projects, and WGC stated very clearly that the only project able to profit from GPU crunching was HCC. They also stated that approx 90-100 days of work are left. That makes it a pretty short GPU project if we compare with some other CPU projects that last for years.

This was the first experiment which I suppose is successfull at least crunchers side if we look at all the reactions and emotions that it has created. We saw also that many crunchers having hardware available when having the possibility of using it at WCG they are ready to move to WCG instead of crunching bitcoins or other projects.
We also must be sure that the scientists will be also completely satisfied.

New facilities like those mentioned in the above posts may be implemented only if WCG sees GPU computing as something to stay and become at least a regular offering in terms of projects. My feeling at this point, is more of a learning experiment which long term continuity is not assured at all. As long as this long term perspective is not defined, there is no sense to invest into a better interface allowing multi-modal, multi-project crunching, selecting the graphic hardware type etc. etc.

I hope simply that GPU crunching will become more of a regularly applied technique as the electrical efficiency or Watts/WU or Watts/flops or whatever other sensible energy based ratio, seems to me unbeatable with CPU processing.

Regarding the Xeon Phi solution it is to be checked as this could be a CPU approach that could also be very high in terms of electrical efficiency too.

Does not change the fact that the motivation for using an app_info is to be able to have multi-WUs-per-GPU to crunch. It also affirms the suggestion that -- unlikely WCG will add the multi-task per GPU option. By the way, the need to include CPU-based-WUs (in an app_info.xml) is just a necessary inconvenience (if CPU-WUs are desired) in pursuit of the path to multi-WUs-per-GPU.

Exactly. Use of multi-WUs-per-GPU is not for your everyday cruncher as the CPU-based-WUs may not load correctly via incorrect use of an app_info.xml file and/or the settings thereof.
;

It's like Then and When, full page or page full being of incomparable different meaning... turds float, but also sink ;>)

Hello retro77,
BOINC is attempting to develop an app_config.xml file that will allow a simple way to run multiple work units on a GPU. This file will not need the anonymous platform nor allow the fine-grain control of the app_info.xml file. Caution - This is a forward-looking statement, as the SEC calls it. The future may be different. But if BOINC does develop this way, computers will not allow ATI and Nvidia GPUs to work together on a project. In such a future, I expect that WCG would disallow the anonymous platform and app_info.xml files will stop working for WCG projects. Of course, this is a distant future if it ever comes to exist.

Lawrence