Forum > Optimized Applications Release News
Windows installer v0.40
Pizzadude:
setiathome_enhanced 6.09 (cuda23) is the exact description that appears, hopefully the .zip file app_info will appear below.
Thanks again for your help.
--- Quote ---
--- End quote ---
Well, first the good news. That host is returning good work, which is validating properly, using the correct application - by that, I mean it will already be using the applications from this installer, even though it's still working through old tasks issued before you used the new installer.
I can't see anything that would explain the Cuda 6.09 - in fact, I was more surprised to see how many tasks were being assigned to the CPU, even for a 12-core i7. It takes a lot of searching to find the Cuda stuff in those 2695 tasks. When you see 'Cuda 6.09', what words do you see, exactly? (cuda), (cuda23), (cuda_fermi) - or something different?
It might be helpful if you could zip up the app_info.xml that the installer has made in your project directory and attach it here. Do it as a zip (or 7z), please, rather than pasting the text, so we can see the exact internal structure.
Richard Haselgrove:
--- Quote from: Pizzadude on 29 Mar 2012, 12:28:28 pm ---setiathome_enhanced 6.09 (cuda23) is the exact description that appears, hopefully the .zip file app_info will appear below.
Thanks again for your help.
--- End quote ---
Well, that turns out to be a fascinating little experiment you've carried out there.
Some background: there are two ways to run BOINC projects - using the stock applications sent out automatically by a project, or - what we do here - use an app_info.xml file to pick and choose our own applications.
What we've grown to expect over the years is that:
For stock apps, BOINC will try to pick the fastest
For app_info.xml, BOINC will use the first entry it comes across in the file.
But you've proved that we were wrong. The key lines in your app_info are:
<version_num>610</version_num>
<count>0.3</count>
<version_num>609</version_num>
<count>0.5</count>
Tasks processed two at a time (count=0.5) will, each individually, complete faster than tasks processed three at a time (count=0.3). The overall throughput for the machine may be better with three at a time, but BOINC is probably too dumb to work that out: I'm guessing that two at a time 'looks faster' to BOINC, and that is the reason it's choosing 609 as the 'best' application.
Something else for you to check, perhaps. If you have any old tasks labelled 610 (cuda_fermi) on the machine, are they running 3 per card (6 at a time for the dual-GPU machine as a whole? And when you get to the 609 (cuda23) tasks, do they run 2 per card (4 at a time overall)?
Once you've finished that, I guess you'll want to get back to normal with whatever <count> value you feel suits you best. If you put that value consistently in all six <app_version> sections, I think you'll find things go back to normal.
Pizzadude:
I only have 6.09 Cuda work at the moment and they are running at 2 per card.
I have adjusted all Cuda counts to 0.3 and I will let you know what develops in the next few days and thanks again for your assistance.
--- Quote ---
--- End quote ---
Well, that turns out to be a fascinating little experiment you've carried out there.
Some background: there are two ways to run BOINC projects - using the stock applications sent out automatically by a project, or - what we do here - use an app_info.xml file to pick and choose our own applications.
What we've grown to expect over the years is that:
For stock apps, BOINC will try to pick the fastest
For app_info.xml, BOINC will use the first entry it comes across in the file.
But you've proved that we were wrong. The key lines in your app_info are:
<version_num>610</version_num>
<count>0.3</count>
<version_num>609</version_num>
<count>0.5</count>
Tasks processed two at a time (count=0.5) will, each individually, complete faster than tasks processed three at a time (count=0.3). The overall throughput for the machine may be better with three at a time, but BOINC is probably too dumb to work that out: I'm guessing that two at a time 'looks faster' to BOINC, and that is the reason it's choosing 609 as the 'best' application.
Something else for you to check, perhaps. If you have any old tasks labelled 610 (cuda_fermi) on the machine, are they running 3 per card (6 at a time for the dual-GPU machine as a whole? And when you get to the 609 (cuda23) tasks, do they run 2 per card (4 at a time overall)?
Once you've finished that, I guess you'll want to get back to normal with whatever <count> value you feel suits you best. If you put that value consistently in all six <app_version> sections, I think you'll find things go back to normal.
--- Quote ---
--- End quote ---
Pizzadude:
Wow, did I say few days - I should have said few minutes. I already have eleven 6.10 Cuda's in my download que. Thankyou again
--- Quote from: Pizzadude on 29 Mar 2012, 03:40:07 pm ---I only have 6.09 Cuda work at the moment and they are running at 2 per card.
I have adjusted all Cuda counts to 0.3 and I will let you know what develops in the next few days and thanks again for your assistance.
--- End quote ---
Pizzadude:
Has anybody actually seen an Astropulse workunit since 0.40 went on general release ?
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version