Open beta for OpenCL AstroPulse for Intel GPUs started

Hi!

Don't seem to be to much interest in testing this app but I gave it a try.
I'm running an i7 3770K @ 4.2 GHz, OS is Win 7 x64 and Boinc 7.0.42 x64.
Link to host 6468133

First impression is that the app seems to work and with the command line parameters set to: -unroll 4 -ffa_block 2048 -ffa_block_fetch 512 the display responds without any noticeable lag and GPU utilization is @>90%, started with -hp also set but the display lags made it very annoying to use the computer. There is probably more room to fine tune the settings.

Run time was 13,238.33s on result 2833151723, 0% blanked and one core left free to feed the GPU, HT is on and all other cores and the GTX 660Ti crunching Einstein.
In comparison two 0% blanked AP-task took 26,894.46s and 19,281.86s on the CPU with ap_6.01r557_SSE2_331_AVX.exe, the last one found 9 single pulses and 30 repetetiv pulses.

The main problem I've encountered is that the app seems to stall from time to time but Boinc still thinks that it's running. Mostly I've observed this behavior when i close some other program that is running, ex. Firefox or VLC, or when I turn the screen of, screen saver is set to none and "Turn of screen after" is set to "never". GPU utilization drops to near zero according to GPU-Z and the CPU support part also drops it's CPU utilization to zero. A simple suspend/resume gets it going again. This happened to result 2832987778 where I got it started and then turned of the screen during the night, the next morning almost no progress had been made but the elapsed time was over 7 hours, watched it more closely and got it running more or less constant afterwards, quite a few suspend/resume cycles as it kept stalling when I was using the computer.

The other problem (or are they connected?) comes from stderr and the lines:

I've now completed one more AP task using default settings of the app as suggested by Mike, that is no command line options specified.
And as suggested by Raistmer I left 2 cores free to support the GPU and anything else not Boinc. Did not change the core affinity on anything.
I did not experience any screen lags and for most of the runtime I was watching a movie in VLC @ 1080p resolution without any problems. The monitor is actually a TV and it's connected to the iGPU.

Link to task 2837364590, run time was 10,423.58s and GPU usage more than 95%, when not actively using the computer the usage was >98%. Am I right to suspect that most of the speedup is because 30 repetitive pulses was found?

As can be seen in the stderr the problem with Boinc supplying the wrong platform is still present even though I've upgraded to version 7.0.52 (latest as of now) and have taken everything else out of the app_info except the entry for this app with one app_version section and the platform is set to windows_x86_64.

Also the problem with the app stalling when closing Firefox and VLC or turning of the monitor is still present, so every time I close one of the programs i have to suspend the task in Boinc and then resume it or it will not continue processing but the elapsed time in Boinc still increases. If I'm going to turn the monitor of I suspend the task before.
Don't know if it's relevant or not but when I turn the monitor of I think some sort of energy saving function kicks in that reduces the resolution from 1920x1080 to something much less than that, it resets to the right resolution when I turn the monitor back on, but any open and not maximized window is now a bit smaller. Haven't been able to find any settings to stop this behavior.

To follow up on the first 3 tasks run all have now validated but one started as inconclusive, it had 30 repetitive pulses. So all looks good on that front.
____________

Also I wonder why Xeon E3-1200 w/ P4000 GPU are not being supported. Perhaps to push their Phi coprocessors instead.

2) do you connect TV via HDMI cable ?

3) Regarding stall when other programm closes (like FireFox) I think it's worth to write on Intel OpenCL forum too. Looks like some issue with OpenCL SDK. Stall on app close and stall on TV/monitor off can be 2 different issues.

Describe please how you monitor GPU usage when monitor switched off? Or you just look GPU usage (via what app?) when it's ON again and it stays idle untill restart ?

4) this build known to have false positives already, at least on my GPU. So, some invalids can appear eventually. While I working on precision issues lets find other ones like those stalls.

EDIT: i created topic for this stall issue on Intel's forum: http://software.intel.com/en-us/forums/topic/369622
Please add any relevant info about your config there.

EDIT: i created topic for this stall issue on Intel's forum: http://software.intel.com/en-us/forums/topic/369622
Please add any relevant info about your config there.

Just posted more info on Intels board.

In the mean time I've run one more AP task and it behaved just like the ones before.
Run time was 13,794s and 2.54% blanking.

Will hold of for a few days and see if something develops on the Intel board before i try any more.
____________

I seem to have run into some trouble getting new task to continue testing this app. Every time Boinc (version 7.0.52) asks for work either by it self or if I issue a update the response is this:

Still something odd with my workfetch, copied Raistmer's app_info from the last post and have only the entry for this app in my app_info. Still get the message from the server that there is no work available.

Then I once again put an entry for Nvidia OpenCL in the app_info and on the first try I got 5 APs assigned!
Did some digging in the client_stat.xml and have managed to reschedule one of the tasks to the Intel GPU.

Will post to the Intel board after I've done some more testing of the stalls.
____________