A new MacOS X Intel App is available from our Beta Test Page.
It is the first "separate graphics App" ("APIv6") for the MaOS X Intel platform. Furthermore it features some more tweaking with the sin/cos code and some prefetching instructions to speed up memory access in the "Hough" part. Many thanks to Bikeman for these improvements!
Happy testing!
BM
I am getting an error message from the app_info.xml, when BOINC starts.
Wed Feb 20 19:15:50 2008||Starting BOINC client version 5.10.42 for i686-apple-darwin
Wed Feb 20 19:15:50 2008|Einstein@Home|Found app_info.xml; using anonymous platform
Wed Feb 20 19:15:50 2008|Einstein@Home|[error] State file error: missing application file einstein_S5R3_4.04_i686-apple-darwin
It appears to be working anyway.
Me too. Weird.
But I'll say this - this app kicks ass. The first WU it threw back was done in 5 hours flat.
It takes ~10.5 hours for my AMD x2, and ~40 hours [subject to variations, etc] on a P3 900.
I am getting an error message from the app_info.xml, when BOINC starts.
Wed Feb 20 19:15:50 2008||Starting BOINC client version 5.10.42 for i686-apple-darwin
Wed Feb 20 19:15:50 2008|Einstein@Home|Found app_info.xml; using anonymous platform
Wed Feb 20 19:15:50 2008|Einstein@Home|[error] State file error: missing application file einstein_S5R3_4.04_i686-apple-darwin
It appears to be working anyway.
You can safely ignore is. The app_info.xml has a line for people switching from 4.04 to 4.33 in order not to loose work of the 4.04 App. This leads to nothing more that this error message for those that previously used a newer App.
The prefetching speeds up the "Hough" part of the code, which is also responsible for the variation in runtime between different workunits of the same frequency. I guess the variation should be reduced significantly wit this App.
An effect of this is that probably people currently running "longer" workunits of a charge will see more speedup than those that are on the "shorter" end.
Interesting. After I copied the 4.33 files to my E@H folder, I tried to restart BOINC. I got an error code concerning permissions and was instructed to reinstall BOINC. Reinstalled BOINC 5.10.34 (same version as before) and everything worked fine.
Same thing happened after I copied the 4.22 files, but I thought it was a fluke. Any ideas on what's happening?
This is caused by the special "sandbox" feature that BOINC uses on MacOS.
Fastest way to fix the permissions I found to "Get Info" of the project folder (einstein.phys.uwm.edu in BOINC Data/projects), show "Details" in "Ownership & Permissions" and click "Apply to enclosed items..." (confirm). This will apply the correct ownership and permissions the folder still has to all items inside, including the files you just put there.
This new app is awesome! My 8-way Mac Pro (the first model at 3.0ghz) ran two WUs with the stock app at ~31,100 sec (8.6 hrs). With the new app, they run ~15,500 (4.3 hrs). That cuts the run time in HALF.
Any fellow proud V8 MacPro owners notice an increase in STDOUTDAE errors like these since switching to 4.33? They don't seem to adversely affect performance, I just have noticed an increase in the frequency of these messages with the fast app and I wonder if others see the same thing.
2008-02-24 14:02:01 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 14:02:01 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
2008-02-24 16:44:36 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 16:44:36 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
2008-02-24 17:15:07 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 17:15:07 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
Agreed - this app is crazy fast. If only PCs had this kind of speed.
They do, now. The GNU/Linux 4.35 app is as fast.
Peanuts 3GHz rig uses lets say 13,800 sec each WU, my Debian rig runs at 4.2GHz, so divide by 4.2 and multiply by 3 for a rough estimate. Then I should have times around 9,850 sec. And i do, and some faster also (probably coz of the Wolfdales bigger cache)
Well, this is quite amazing... Core Duo 2.0GHz Macbook, goes about 30h with the standard app. With the beta app, that decreases to 7h49m! That's an incredible speedup! Gonna definitely crunching more Einstein from here on. Now we'll see if I get a similar speedup with my P4 running Fedora...
RE: RE: A new MacOS X
)
Me too. Weird.
But I'll say this - this app kicks ass. The first WU it threw back was done in 5 hours flat.
It takes ~10.5 hours for my AMD x2, and ~40 hours [subject to variations, etc] on a P3 900.
RE: I am getting an error
)
You can safely ignore is. The app_info.xml has a line for people switching from 4.04 to 4.33 in order not to loose work of the 4.04 App. This leads to nothing more that this error message for those that previously used a newer App.
BM
BM
RE: The prefetching speeds
)
An effect of this is that probably people currently running "longer" workunits of a charge will see more speedup than those that are on the "shorter" end.
BM
BM
RE: RE: Interesting.
)
Fastest way to fix the permissions I found to "Get Info" of the project folder (einstein.phys.uwm.edu in BOINC Data/projects), show "Details" in "Ownership & Permissions" and click "Apply to enclosed items..." (confirm). This will apply the correct ownership and permissions the folder still has to all items inside, including the files you just put there.
BM
BM
This new app is awesome! My
)
This new app is awesome! My 8-way Mac Pro (the first model at 3.0ghz) ran two WUs with the stock app at ~31,100 sec (8.6 hrs). With the new app, they run ~15,500 (4.3 hrs). That cuts the run time in HALF.
You can see the results here:
http://einsteinathome.org/host/1117584/tasks
This will eventually be rolled out as the stock app, right? There are no legacy processors to support.
Reno, NV Team: SETI.USA
Agreed - this app is crazy
)
Agreed - this app is crazy fast. If only PCs had this kind of speed.
Any fellow proud V8 MacPro
)
Any fellow proud V8 MacPro owners notice an increase in STDOUTDAE errors like these since switching to 4.33? They don't seem to adversely affect performance, I just have noticed an increase in the frequency of these messages with the fast app and I wonder if others see the same thing.
2008-02-24 14:02:01 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 14:02:01 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
2008-02-24 16:44:36 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 16:44:36 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
2008-02-24 17:15:07 [---] Can't rename client_state_next.xml to client_state.xml; check file and directory permissions
2008-02-24 17:15:07 [---] rename client_state_next.xml to client_state.xml returned error 2: No such file or directory
RE: Agreed - this app is
)
They do, now. The GNU/Linux 4.35 app is as fast.
Peanuts 3GHz rig uses lets say 13,800 sec each WU, my Debian rig runs at 4.2GHz, so divide by 4.2 and multiply by 3 for a rough estimate. Then I should have times around 9,850 sec. And i do, and some faster also (probably coz of the Wolfdales bigger cache)
Team Philippines
Well, this is quite
)
Well, this is quite amazing... Core Duo 2.0GHz Macbook, goes about 30h with the standard app. With the beta app, that decreases to 7h49m! That's an incredible speedup! Gonna definitely crunching more Einstein from here on. Now we'll see if I get a similar speedup with my P4 running Fedora...
4.33 has been made
)
4.33 has been made "official".
For the moment I'd say leave the app_info.xml in place; removing it will terminate all task in progress with a Client Error.
BM
BM