Forum > Linux

Open BSD i386 Seti@HOME Applications

<< < (8/22) > >>

pp:
Yes, that was a little interesting actually. My library is called libfftw3f.so.5 so I had to make a symlink to libfftw3f.so.4 which was in the output from ldd. Still the version is 3.2.2 which is the current in Ports.
1st percent finished in 20 minutes and if it scales linearly, which it usually does, it should finish in about 30 hours. That's also a little funny. I run the Linux version on an sse3 capable Athlon64 which i have downclocked from 2,4GHz to 2,0Ghz to keep the fan silent. That CPU needs 52 hours to finish one AP. If the Celeron finish faster, despite it's only 1,6GHz, I'd be pretty surprised... but I have read that the sse3 implementation in the Athlon64s isn't very good but the tests I've seen at least put it on par with sse2. It will be interesting to see when this wu actually finish. Thanks for your excellent work Urs.

Urs Echternacht:

--- Quote from: pp on 30 Sep 2009, 06:50:07 pm ---Yes, that was a little interesting actually. My library is called libfftw3f.so.5 so I had to make a symlink to libfftw3f.so.4 which was in the output from ldd. Still the version is 3.2.2 which is the current in Ports.
1st percent finished in 20 minutes and if it scales linearly, which it usually does, it should finish in about 30 hours. That's also a little funny. I run the Linux version on an sse3 capable Athlon64 which i have downclocked from 2,4GHz to 2,0Ghz to keep the fan silent. That CPU needs 52 hours to finish one AP. If the Celeron finish faster, despite it's only 1,6GHz, I'd be pretty surprised... but I have read that the sse3 implementation in the Athlon64s isn't very good but the tests I've seen at least put it on par with sse2. It will be interesting to see when this wu actually finish. Thanks for your excellent work Urs.

--- End quote ---
Recalculate your time prediction after 5% or 10% into the wu. That should give you more realistic estimations (from experience with other OSs).

pp:
35% after 10 hours so 30 hours seems to be a good estimate. I felt brave enought to launch another Astropulse on the second core on the Celeron but was I surprised. 4% in 2,5 hours, estimates more than doubled despite the fact that the OS ran one wu on each core. This might explain why the stock AP app performed so slowly on this computer, 120 hours for one wu. Still don't understand the technical reasons for this though. Anyone else who has tried running only one instance on a multicore CPU and compared the times to running one instance on each core? Guess the small cache in the Celeron might be to blame... will continue using only one core for now.

Raistmer:
AstroPulse in some parts very hungry for L2 cache. AFAIK Celerons as like Semprons have reduced L2 cache. So they not very suitable for AP.
Aspecially being running 2 copies of it. Project pairing, damn lack of project/app pairing in BOINC....

pp:
The first AP with the FreeBSD version is still in validation but I notice some peculiarities:

- Claimed credit is only 473. I thought all the new APs were 800 credits?
- Run time is reported as 18 hours but unless something happened during the final phase (while I was sleeping) this wu ran for 30 hours.

stderr can be found here: http://setiathome.berkeley.edu/result.php?resultid=1376540593

It says something about "exiting" and "percent blanked".

Navigation

[0] Message Index

[#] Next page

[*] Previous page