MacOS PPC Beta Test App 4.56 available
3 May 2006 13:54:29 UTC
Topic 13322
(moderation:
I put a new MacOS App for testing on the Beta Test Page. It incorporates ideas from E@H usser "AkosF" and some others (including me) making it faster than the current official App.
On non-AltiVec CPUS (G3) which use a "generic" code this App should run about twice as fast as the current official one. The AltiVec code seems to be about 25% faster than the older one - have a try.
This Altivec code suffers from the same problem than the 4.55 Linux App, which show up as occasional invalid results. However the speedup may be worth it.
The App has been built with a modified build process on MacOS 10.4 (Tiger). It runs on the few MacOS 10.3 machines I could get access to, too, but I'm not sure about that in the global scale, which is the main reason for posting it here. Please report especially problems or success with running this on Panther.
BM
BM
MacOS PPC Beta Test App 4.56 available
)
Your recent results have (in stderr):
006-05-03 22:09:04 [Einstein@Home] execv(../../projects/einstein.phys.uwm.edu/albert_4.39_powerpc-apple-darwin) failed: error -1
execv: No such file or directory
Did you delete the 4.39 App? The instructions say that new work will be assigned to the 4.56 App. The old work that has already been assigned to the 4.39 App isn't touched and expected to be done with that. Seems it had been that old results that failed. Update the project and see if new work is assigned to the new App and how it works.
BM
BM
RE: by the way, does the
)
I'm not sure I understand the question.
The Workunits and Tasks haven't changed at all, assigning the tasks to the App versions is done by the BOINC client, not by the Apps. The client will make sure that a single task is done by the same App from beginning to end, it doesn't allow to change the App in the middle of a task.
On the App side the format for the intermediate result and checkpoint file hasn't changed, so in principle it should be possible to continue a task that has been started with 4.39 with 4.56. However, as I said, the BOINC client should normally prevent this. I'd rather consider this to be a good idea, as what applies to 4.39 and 4.56 here can by no means be assumed in general.
BM
BM
To the people observing a
)
To the people observing a slowdown: Are these G4 or G5 machines? Is there a difference between them?
BM
BM
That's actually pretty weird
)
That's actually pretty weird - my machines here are all G4 and show the 20% speedup of the AltiVec code. I really wonder why the very same code doesn't have any (positive) effect on G5 machines. The only thing I can think of is that the G5 further optimizes the old AltiVec code by out-of-order execution so much that it doesn't gain anything from the new code. I'll have to consult some people more knowledgeable than me...
BM
BM
RE: RE: To the people
)
Thanks, that was helpful, especially the Velocity Engine part. I'll have to shark the code again.
BM
BM
RE: Which times are to
)
The OS version shouldn't make much of a difference; I'd expect crunching times to be rather slightly slower on OSX.4, because of background processes (Spotlight).
This machine might be comparable to yours, it took 113004.48s CPU-time (maybe 32h wall-clock if you're lucky) for a result with the official App 4.39. The 4.56 should be about twice as fast on non-AltiVec machines.
BM
BM
RE: I'm not sure if I
)
This is intentional. The error message was a bug in the BOINC version we were using back then.
BM
BM
RE: Any updates regarding
)
Gaurav Khanna, who also contributed to the AltiVec code of the official App, claims that the very same (source) code I am using compiles into an App that runs about 30% faster on his G5 than the 4.56 App. We're both using the same Xcode version and the same compiler flags - and are both plexed at the difference. This definitely needs further investigation, but there are also some other urgent tasks to do.
BM
BM
Strangely enough the 4.56 App
)
Strangely enough the 4.56 App doesn't show the same speedup on G5s than it does on G4s. We are looking into this.
BM
BM
Ok, the issue with this App
)
Ok, the issue with this App on G5 (and early G4) has finally been resolved. Now having found it, I'm more surprised about the speedup on G4 than about the slowdown on G5 - pretty weird. Anyway, I made a new App, read about it here.
BM
BM