Fine for those workunits that meet ALL the following qualifications:

1. They include many different starting points instead of requiring each action to use part of what was calculcated during the previous action, and therefore being likely to make the GPU version SLOWER than the CPU version.

2. Each starting point requires so little graphics memory for the calculations that there is enough graphics memory for the speedup you want.

3. Each of the GPU processors can reach enough of the graphics memory that it can do its calculations.

4. The developers have enough time available that they can learn the necessary GPU-specific computer languages well enough to start writing all their program changes in those languages.

5. If the developers decided to do their programs in OpenCL (like POEM@HOME did), they'll find it very difficult to extend that to GPU workunits until newer versions of BOINC (past the 6.12.* series) add enough support for OpenCL GPU workunits. Milkyway@Home appears to be the only BOINC project to get past those difficulties and offer OpenCL GPU workunits so far.

Rather few of the current workunits currently available from various BOINC projects meet ALL of those qualifications.


If you want to speed up the changes to GPU workunits, why don't you start learning CUDA, OpenCL, and CAL, plus whatever computer languages were used for the older versions of the applications, so you can actually help in such a change, instead of constantly repeating essentially the same thing over and over? I've already started learning the first of the four computer languages needed to do that for one specific BOINC project.

Or, if you prefer, sign up as a BOINC developer or a BOINC alpha tester, and help get BOINC ready to handle OpenCL GPU workunits better.