I have installed app. 4.29 on my two PowerBook G4 running OSX.4.10 and OSX.3.9 respectively. In both cases, the installation was done halfway in the crunching of long WU's (over 400 credits each), and reinstallation of BOINC was required after installation of app. 4.29 to get things going again. It seems that no crunching time was lost and no obvious gain in speed achieved. When completed, both WU's received the expected credit.
My first WU with 4.29 is going to provide a ?good? test. My home had power outages two days in a row while the app was crunching. The outages lasted longer than my battery backup did so my computer crashed. The WU is still crunching, but slowly and the To Completion clock is running both forward and backward. I guess I will let the WU go on because the progress meter is going only one way (up) like it usually does. Wish me luck. ;)
I reported that app 4.29 ran at 75% the speed of the stock 4.20 app after graphics were displayed briefly. I have now tracked the crunching time since restarting from the checkpoint the following day. Without the side issue of graphics, 4.29 runs at 82% the speed of 4.20. This finding accords exactly with a report from Josep on the Linux thread.
Mac PPC (or at least, my G5) and Linux betas appear to be losing 20% somewhere, Bernd. Wonder how Mac Intel and Windows are making out?
I can confirm this: I completed 2 WUs on different G5 machines, both took appr. 21% longer than before. However, in the past we often saw that the first work unit crunched with a new science app takes considerably longer than all other WUs, therefore we should wait until the second and third work unit is completed on each computer until a final conclusion can be made.
I reported that app 4.29 ran at 75% the speed of the stock 4.20 app after graphics were displayed briefly. I have now tracked the crunching time since restarting from the checkpoint the following day. Without the side issue of graphics, 4.29 runs at 82% the speed of 4.20. This finding accords exactly with a report from Josep on the Linux thread.
Mac PPC (or at least, my G5) and Linux betas appear to be losing 20% somewhere, Bernd. Wonder how Mac Intel and Windows are making out?
I can confirm this: I completed 2 WUs on different G5 machines, both took appr. 21% longer than before. However, in the past we often saw that the first work unit crunched with a new science app takes considerably longer than all other WUs, therefore we should wait until the second and third work unit is completed on each computer until a final conclusion can be made.
Mac Intel App seems to run a bit faster (!) now. I haven't disassembled the app, but my guess is that the Intel MacOS apps are compiled to make use of SSE and SSE2 instructions e.g. for floating point to integer conversion, which is much faster using SSE2. Because MacOS runs only on very recent Intel CPUs, it's safe to assume that the CPU supports SSE2. On Intel/Linux, the compiler has touse/generate code that covers a much broader set of CPUs, including those that do not support SSE2, using either FPU or standard integer instructions. Without SSE(2), it's much harder to get decent performance for floating point to integer conversion (which seems to be the area where the Linux beta app suffers from performance problems now).
I reported that app 4.29 ran at 75% the speed of the stock 4.20 app after graphics were displayed briefly. I have now tracked the crunching time since restarting from the checkpoint the following day. Without the side issue of graphics, 4.29 runs at 82% the speed of 4.20. This finding accords exactly with a report from Josep on the Linux thread.
Mac PPC (or at least, my G5) and Linux betas appear to be losing 20% somewhere, Bernd. Wonder how Mac Intel and Windows are making out?
I can confirm this: I completed 2 WUs on different G5 machines, both took appr. 21% longer than before. However, in the past we often saw that the first work unit crunched with a new science app takes considerably longer than all other WUs, therefore we should wait until the second and third work unit is completed on each computer until a final conclusion can be made.
Mac Intel App seems to run a bit faster (!) now. I haven't disassembled the app, but my guess is that the Intel MacOS apps are compiled to make use of SSE and SSE2 instructions e.g. for floating point to integer conversion, which is much faster using SSE2. Because MacOS runs only on very recent Intel CPUs, it's safe to assume that the CPU supports SSE2. On Intel/Linux, the compiler has touse/generate code that covers a much broader set of CPUs, including those that do not support SSE2, using either FPU or standard integer instructions. Without SSE(2), it's much harder to get decent performance for floating point to integer conversion (which seems to be the area where the Linux beta app suffers from performance problems now).
I reported that app 4.29 ran at 75% the speed of the stock 4.20 app after graphics were displayed briefly. I have now tracked the crunching time since restarting from the checkpoint the following day. Without the side issue of graphics, 4.29 runs at 82% the speed of 4.20. This finding accords exactly with a report from Josep on the Linux thread.
Mac PPC (or at least, my G5) and Linux betas appear to be losing 20% somewhere, Bernd. Wonder how Mac Intel and Windows are making out?
I can confirm this: I completed 2 WUs on different G5 machines, both took appr. 21% longer than before. However, in the past we often saw that the first work unit crunched with a new science app takes considerably longer than all other WUs, therefore we should wait until the second and third work unit is completed on each computer until a final conclusion can be made.
That is nice to know. My 1st WU just completed with 4.29. Took 305101 CPU sec to complete 448 credits. 4.20 was taking only ~241600. I'm working on my second WU now. I'll see if it does better in around 80 hours.
My dual-core G5 with Mac OS 10.4.10 has completed its first couple of results with v4.29; it appears that its crunching time has increased about 23% (from a little under 23 h to 28 h, on 381-CS tasks). I didn’t have any problems with permissions or the like, probably because this host is still running BOINC v5.4.9.
I have installed app. 4.29 on
)
I have installed app. 4.29 on my two PowerBook G4 running OSX.4.10 and OSX.3.9 respectively. In both cases, the installation was done halfway in the crunching of long WU's (over 400 credits each), and reinstallation of BOINC was required after installation of app. 4.29 to get things going again. It seems that no crunching time was lost and no obvious gain in speed achieved. When completed, both WU's received the expected credit.
My first WU with 4.29 is
)
My first WU with 4.29 is going to provide a ?good? test. My home had power outages two days in a row while the app was crunching. The outages lasted longer than my battery backup did so my computer crashed. The WU is still crunching, but slowly and the To Completion clock is running both forward and backward. I guess I will let the WU go on because the progress meter is going only one way (up) like it usually does. Wish me luck. ;)
RE: I reported that app
)
I can confirm this: I completed 2 WUs on different G5 machines, both took appr. 21% longer than before. However, in the past we often saw that the first work unit crunched with a new science app takes considerably longer than all other WUs, therefore we should wait until the second and third work unit is completed on each computer until a final conclusion can be made.
RE: RE: I reported that
)
Mac Intel App seems to run a bit faster (!) now. I haven't disassembled the app, but my guess is that the Intel MacOS apps are compiled to make use of SSE and SSE2 instructions e.g. for floating point to integer conversion, which is much faster using SSE2. Because MacOS runs only on very recent Intel CPUs, it's safe to assume that the CPU supports SSE2. On Intel/Linux, the compiler has touse/generate code that covers a much broader set of CPUs, including those that do not support SSE2, using either FPU or standard integer instructions. Without SSE(2), it's much harder to get decent performance for floating point to integer conversion (which seems to be the area where the Linux beta app suffers from performance problems now).
CU
BRM
RE: RE: RE: I reported
)
Bikeman,
good answer to the wrong question. This is the MacOS PowerPC thread, NOT the Intel-thread. Here is the Intel-thread: MacOS Intel S5R2 App 4.26 available for Beta Test
RE: RE: I reported that
)
That is nice to know. My 1st WU just completed with 4.29. Took 305101 CPU sec to complete 448 credits. 4.20 was taking only ~241600. I'm working on my second WU now. I'll see if it does better in around 80 hours.
My dual-core G5 with Mac OS
)
My dual-core G5 with Mac OS 10.4.10 has completed its first couple of results with v4.29; it appears that its crunching time has increased about 23% (from a little under 23 h to 28 h, on 381-CS tasks). I didn’t have any problems with permissions or the like, probably because this host is still running BOINC v5.4.9.
Plese test the new App. BM
)
Plese test the new App.
BM
BM