The drivers I had in September (before the update to opencl) didn't had the opencl feature so they won't work. That was the reason I updated the drivers from ATI.

At Seti forums I've read, that the current ATI drivers have some issues with OpenCL. Not sure if that applies here, but the latest version they recommended there is 12.8. Maybe you could try these.

Running an ATI Radeon HD 5850 (cat 12.1) I just checked my application details:
Consecutive valid tasks 15020
for
milkyway_separation 1.02 Windows x86 double OpenCL

Einstein and Milkyway on the ATI gives quite amount of invalids, Milkyway way more though.

Than something is wrong with your GPU (or eventually something else in that system), even the old CAL app I have to use for my HD3850 does not have such high error/invalid rates as you reported in your first post (and it has a known bug where sometimes the output is not written to the std_err). The new OpenCL apps are known to run more or less without any noteworthy issues.

So you think nVidia GPUs don't generate invalid results...


EDIT@Rantanplan: with an app_info.xml in the projects folder you don't need to update BOINC.

I am currently using my ATI/AMD cards at Moo right now but have crunched on and off using them here and have never had a problem. I am currently using my Nvidia cards and don't have a problem either. I have some questions if you don't mind: what kind of AMD card do you have? What driver version are you using? Are you leaving a cpu open while you crunch or are they all busy crunching their own units? What version of Boinc are you using? How much ran does the pc have in it? What OS are you using? Your pc's are hidden so I can't see any of this for myself.

I haven't change anything as from September. It all started when there was a change in the searches.
I don't have the latest drivers, I never use them. They resolve some bugs, but create new ones. I even try to use a BOINC version that does what I want, but sometimes a newer is necessary to run certain projects.

I use F-Secure as a AV and FW, and have nothing excluded. They solved all the issues with not uploading and automatically new work, and everything is running smooth and save.

Only Milyway produces a lot on invalids (can't validate, while wingman are validated).

Are you running MW, Gpugrid and Einstein on the same gpu's at the same time? If so set the swap time so no unit gets suspended mid task before it switches to the next task if you can. I never run more than one gpu project at a time on a single pc, a reason to start building towards a 'ranch', which is 15 or more pc's! I tend to focus on a project, giving myself a goal then switching my resources to the next project etc, etc.

Is anyone from the project actively working on this problem or are we going to waste cruncher's resources until it magically goes away? I moved to Einstein but would rather be back here. Oh well ...