May I remind you that it can be quite confusing to have two different apps with exactly the same version number - in this case, the CUDA Beta and the screensaver CPU?
Just curious... On my Mac, it seems ABP1 5.10 performed worse than 5.09, with the former averaging around 17,000 sec per WU compared to the latter of 16,000 to 16,500 sec per WU.
Just curious... On my Mac, it seems ABP1 5.10 performed worse than 5.09, with the former averaging around 17,000 sec per WU compared to the latter of 16,000 to 16,500 sec per WU.
Any ideas...?
The runtime difference is in the order of 2,5% - We expect 5% to be normal fluctuation due to e.g. other processes on the same machine. There are some parts in the code where the runtime depends on the data being analyzed (e.g. the number of candidates found that pass a certain threshold), so not all workunits run exactly the same time. The science application binary is identical between these two versions, there's not a single Bit that's different.
Just curious... On my Mac, it seems ABP1 5.10 performed worse than 5.09, with the former averaging around 17,000 sec per WU compared to the latter of 16,000 to 16,500 sec per WU.
Any ideas...?
The runtime difference is in the order of 2,5% - We expect 5% to be normal fluctuation due to e.g. other processes on the same machine. There are some parts in the code where the runtime depends on the data being analyzed (e.g. the number of candidates found that pass a certain threshold), so not all workunits run exactly the same time. The science application binary is identical between these two versions, there's not a single Bit that's different.
BM
At first I was to believe both app-runtime should be averaging the same - plus minus a few hundred seconds, but even since I switched to 5.10, I've seen a lot more work units breaching 17,000 seconds.... I use to have one or 2 WUs out of 20 breaching that mark with 5.09, it seems a lot more now with 5.10. I hardly see any 5.10 WU going below 16,500 seconds anymore than 5.09 did...
Once I've drained the 5.10 WUs, I'll switch back to 5.09 to see if the problem persists.
If there really is a difference in runtime, it's purely accidental coincidence with e.g. properties of the data, or maybe a background process on your machine.
official einsteinbinary ABP1 x.10 Apps
)
Bernd,
May I remind you that it can be quite confusing to have two different apps with exactly the same version number - in this case, the CUDA Beta and the screensaver CPU?
Just curious... On my Mac, it
)
Just curious... On my Mac, it seems ABP1 5.10 performed worse than 5.09, with the former averaging around 17,000 sec per WU compared to the latter of 16,000 to 16,500 sec per WU.
Any ideas...?
RE: Just curious... On my
)
The runtime difference is in the order of 2,5% - We expect 5% to be normal fluctuation due to e.g. other processes on the same machine. There are some parts in the code where the runtime depends on the data being analyzed (e.g. the number of candidates found that pass a certain threshold), so not all workunits run exactly the same time. The science application binary is identical between these two versions, there's not a single Bit that's different.
BM
BM
RE: RE: Just curious...
)
At first I was to believe both app-runtime should be averaging the same - plus minus a few hundred seconds, but even since I switched to 5.10, I've seen a lot more work units breaching 17,000 seconds.... I use to have one or 2 WUs out of 20 breaching that mark with 5.09, it seems a lot more now with 5.10. I hardly see any 5.10 WU going below 16,500 seconds anymore than 5.09 did...
Once I've drained the 5.10 WUs, I'll switch back to 5.09 to see if the problem persists.
[root@einstein download]# cmp
)
If there really is a difference in runtime, it's purely accidental coincidence with e.g. properties of the data, or maybe a background process on your machine.
BM
BM
RE: RE: The science
)
You don't believe the developers? ;-)
RE: RE: RE: The science
)
Good call... I don't believe anything that's in bits and bytes...