AstroPulse for ATI GPUs released

[Raistmer]

"Error in ap_remove_radar.cpp: generate_envelope: num_ffts_performed < 100.  Blanking too much RFI?"

First task completed just OK, no prob here, just too much noise (not app problem).
(and second your task just the same). They are valid, will get credit and so on. (if your wingman will find same % of blanking too)

[timiman]

OK.
The same here http://setiathome.berkeley.edu/result.php?resultid=1709396634

How possible is 3/3 of having too much RFI? Unlucky of me 

[Raistmer]

 If they have sequental numbers (i.e, came from almost same point in time and space) - very possible
[Let see:
ap_05ap10ab_B3_P1_00022_20100917_13109
ap_05ap10ab_B3_P1_00021_20100917_13109
ap_05ap10ab_B3_P1_00022_20100917_13109

]
I would say - yes, very probable, especially taking into account that third just one of first two

[Raistmer]

BTW, I see 2 more AP tasks assigned for your host. They are from another day.If they will have too much RFI too - it would be more strange...

[timiman]

BTW, I see 2 more AP tasks assigned for your host. They are from another day.If they will have too much RFI too - it would be more strange...

Hmmm.. I see them... Assigned but never downloaded to my PC.

[Josef W. Segur]

If they have sequental numbers (i.e, came from almost same point in time and space) - very possible
[Let see:
ap_05ap10ab_B3_P1_00022_20100917_13109
ap_05ap10ab_B3_P1_00021_20100917_13109
ap_05ap10ab_B3_P1_00022_20100917_13109

]
I would say - yes, very probable, especially taking into account that third just one of first two

There have also been at least two extended periods in the past where all _B3_P1_ tasks had bad data which causes that "...Blanking too much RFI?" exit, some kind of problem at the multibeam recorder I think. By extended I mean weeks or months, but as it is only one channel out of 14 there were still many good WUs to do.
                                                                                         Joe

[Raistmer]

the only question remains - if SO long periods involved why to split apparently broken data... tasks were completed in few seconds, each task takes 8MB to download. Downloading time will exceed crunching time even for typical ADSL connection....

[timiman]

I have some data to share with you guys about my first non-"too much RFI" workunit of AP.

ATi 4870x2 running on stock clocks using Catalysts 10.9, BOINC 6.10.56 x64 and OpenCL_r456 application.

Up to now I can see that the application:
a. is doing 14,1% per hour meaning a ~7 hours for a 100% (% per hour is rising every interval so it might hit 4-5 hours for a 100%),
b. is using 45%-98% of the GPU with an average usage of 83%, and
c. is using 3%-50% one core of the CPU with an average usage of 24% (same core is shared with a MB workunit)

I also see that the intervals of the BOINC Manager is every 0,9% of the workunit.
BOINC Manager continuously asks for a task for the 2nd GPU, but up to now it seems unable to get one, although there are some available.

Any other date that might be interesting to give you, please tell me to.

PS. Great work guys!!!

[Raistmer]

Performance depends from % of blanking for particular task. From your description one could infer you seeing task with high enough % of blanking. To check this, please, post link to result for this task when it finishes.

[timiman]

Good news, managed to download another task for the 2nd GPU.
Bad news, 3D clocks do not kick in, so 2nd GPU is running on 507MHz and not in 750MHz (possibly an error of 10.9 Catalysts as this appears on some games also).

I'll post the result link of the first task as soon as it finishes.

[Raistmer]

For Cat 10.7 and HD4870 I see clock increase when app works.
It's Cat 10.9, definitely.

[timiman]

Just a small update on the 2 AP tasks of mine.

Task 1 crunching on the 1st GPU @ 3D clocks, having an usage on an average of 85%, with lows and peaks of 55%-98%
GPU Time 4:14 @ 60.360% (14,3% per hour) , found 2 pulses up to now

Task 2 crunching on the 2st GPU @ 2D clocks, having an usage on an average of 82%, with lows and peaks of 31%-91%
GPU Time 3:15 @ 21.621% (6,7% per hour) , found 30 pulses up to now

Both appear to have taken 94-114MB of memory.
Also, many times Boinc Manager paused and started the tasks on high CPU usage -as I have told it to, without any problem like a "Computation Error" or "Restarting the task from the beginning".

[timiman]

First AP is ready http://setiathome.berkeley.edu/result.php?resultid=1710297393  ( percent blanked: 23.30 )

[timiman]

2 more tasks finished during the night:
http://setiathome.berkeley.edu/result.php?resultid=1710348645 (percent blanked: 51.39)
http://setiathome.berkeley.edu/result.php?resultid=1710348658 (percent blanked: 2.39)
Last one gave me 1,357.91 credits. Nice boost!!!

[Raistmer]

2 more tasks finished during the night:
http://setiathome.berkeley.edu/result.php?resultid=1710348645 (percent blanked: 51.39)
http://setiathome.berkeley.edu/result.php?resultid=1710348658 (percent blanked: 2.39)
Last one gave me 1,357.91 credits. Nice boost!!!