ABP1 3.09 App for Windows available for Beta Test
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250515710
RAC: 34286
7 Aug 2009 13:49:16 UTC
Topic 194485
(moderation:
A new Einstein@home ABP1 App for Windows is available for Beta Test at Beta Test Page.
This is a big package to download, as it contains the 3.09 ABP1 application as well as the current 3.05 S5R5.
The new ABP1 App should mainly benefit from using SSE in the FFTw (and some other parts). I'd be curious if there are still machines out there without SSE and how they'd react on this App (the FFTW doku is somewhat ambiguous at that point).
app_info.xml does contain entries for ABP1 3.07, 3.08 and 3.09 and S5R5 3.05, tasks assigned to older apps will error out.
Please test and report.
BM
BM
ABP1 3.09 App for Windows available for Beta Test
)
Suspended everything, installed, and restarted. Taskid 135652335 (Arecibo) seemed to resume OK with the app_info. Guess I'll find out when it finishes and gets verified. Or not.
Mike
:( RE: 07/08/2009
)
:(
Why? Old graphics card?
RE: :( Why? Old graphics
)
This isn't a GPU / CUDA App. You should get work when your client requests CPU work.
BM
BM
Thank you! :)
)
Thank you! :)
Taskid 135652335 finished and
)
Taskid 135652335 finished and verified. Roughly the last third was run under the beta app. My earlier Arecibo wus ran for ~61000 secs. This one was 56000 secs. So at 1/3 old wus = 20000 with a reduction to 15000 with the beta, that looks like a useful speed gain. Only a one-off of course, but it looks promising.
XP home. SP3. Athlon 64 3000+. 2GHz.
Mike
First installed on the laptop
)
First installed on the laptop which has a Core2 Duo T7300 @2GHz (stock) with 32-bit Vista HP. Had no ABP1 tasks in cache and having downloaded a couple, suspended the other tasks. App started normally and at 60% complete the speed increase projects to be just above 10%. It's this task. With 3.05 crunch times were ~30,000 seconds, with 3.07 ~40,000 seconds.
I first thought that I'd have to abort my 3-day cache on the Quad to get stuck in to this Beta test right away. That's because it's a 64-bit Vista system and I have previous experience of an app_info introduction dumping the entire cache with 'No app_version for result' errors.
(Apparently, app_versions declared in an app_info default to the client platform, in my case windows_x86_64 rather than the server provided windows_intelx86 that the cached tasks have. Adding tags to the app_version sections didn't help but I did try just the once and may have made a mistake (it was a while back).)
Then I remembered that the tasks don't actually error out; they are just removed. And Einstein, unlike Seti where I had my previous bad experience, of course resends lost results. So I just allowed the tasks that were 3/4 of the way through run to completion and suspended the rest. And true enough, after the new app install all Einstein tasks were gone and a scheduler request resent them with the ABP1 tasks rebranded as 3.09.
I see it has just started crunching four ABP1 tasks which the scheduler nicely resent ahead of the S5R5 tasks. One example. Run times for v3.05 and v3.08 were ~22,500 seconds while v3.07 pushed out to 30,000 seconds. Looking forward to seeing improvements in the morning!
By the way Bernd, if it is possible, it would be really nice to get rid of this 'Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.' and the accompanying 4-hour delay! It happens when the client foolishly asks for CUDA work even with no CUDA app specified in the app_info.
RE: By the way Bernd, if
)
Your request will be difficult if not impossible to grant.
Cuda__is ‘All important ‘ whether you have it or not.
I’m afraid you’ll have to bypass Bernd, bow down and
send your appeal directly to:
The Grand Pooh-Bah
Der Führer
Boincifex maximus
Also known as ‘DA’ over at Seti .
Bill
RE: Taskid 135652335
)
Those 61,000 sec tasks were crunched with v3.07 which was ~30% slower than 3.05/3.08. In your task list there are still some 3.05 tasks and they took about 49,000 seconds. A third of 49,000 ~= 16333 so some improvement but not quite 25%.
The laptop's first task
)
The laptop's first task finished and validated at 11% speed improvement (from 30,000s to 26,500s). Only one core available to BOINC.
On the Quad the speedup wasn't as good: 22,500s to 21,000s = 6-7%. This was running first four and then three ABP1 tasks in parallel. Two tasks have validated. Will try running just one ABP1 task with two ABC and one Seti unit + Seti MB on GPU.
I finally completed 2 ABP1
)
I finally completed 2 ABP1 work units with the 3.09 app. averaging 5.79 hours each.
That is pretty good ( I think ) my computer is a core 2 at 2.93 ghz.
The last three I completed took 8.27 hours each with the ‘ slow ‘
3.07 app.
Bill