Wow, only about ~2 hours after kneed's announcement. I guess there wasn't that many sent out.

I got about 45 GPU Beta WU's and all appear valid except for one that's still pending from my wingman. I'm running Win7, Nvidia GTX590 (2), driver Beta 301.24 (fixed the sleep problem), BOINC 7.0.25. I didn't even know the Beta's had run until I noticed that my non-Beta WU's were all in panic mode, thinking CEP2 needed 300 hours more to run per WU and C4CW thought it needed 30+ hours per WU to finish. Never saw that happen before, but then, this is my first GPU Beta that I've been able to get. Wish I could have actually watched it fly through the tasks.

Warning. This is at your own peril!:

There is a PTF to this: Stop BOINC, find the client_state.xml in the data directory. Make a copy. Open the original and find the first line that has "world" in it which is

<master_url>http://www.worldcommunitygrid.org/</master_url>

Then scroll further down and find the first line that shows as

<duration_correction_factor>1.006634</duration_correction_factor>

The DCF number is in normal operation near 1.000000 (6 decimals). In your case it will be in the hundreds. Edit it down to near 1.000000 , then save the file, start BOINCMgr, whiich for the latest version starts the service (if installed so) or BOINC service through the Task Manager, hit "show all user processes" button, then in service tab, right click BOINC and start.

Hope the techs have a working solution by the time the GPU tasks go to production as the estimation of work duration come off the actual processed work for a science. Separate feeder, I'm sure the first step is to get them to estimate their real run times, so the DCF wont go berserk when CPU crunching at WCG. It's a tough nut with the way BOINC is designed... one DCF per attached grid (for our 12.1 active researches).

If you need more support on this, then you're not ready to engage in "advanced cruncher" hacking (which GPU crunching ATM still seems to require to a certain extend). Let the client crunch and in 15-25 completed results the TTC estimates will be less extravagant.

--//--

edit: the spelling devil booted.

<dont_use_dcf> has not been activated by WCG for various reasons [techs working with developers on this], and is a server controlled value. The new client is similarly, just fitted with 1 DCF per attached grid, and I'm not mistaken ;P

--//--

Consider that everything is in flux and by the time production release is reached some additional device profile options are likely added, so requests in vane wont be issued if you have not selected the present "Send GPU work for the sciences I've selected".

Meantime, at a client level, there are controls you can set through the cc_config.xml. The BOINCTasks multi-client manager by Efmer actually has a GUI to edit this file on the fly to set the <no_gpus> and <use_all_gpus> values, and more to control at card level.

There was a "We have no GPU work, back-off for 14 days" rule at WCG, but forgot if that was does away with. This deferral counter is reset when hitting the Update button in the client.

--//--