Is 6.10 only a bugfix or a speed-app? I'm still running 6.05 without any problems with checkpoints...
ah, so that's why suddenly the CPU time required increased so much.. so I'm right in thinking that 6.08 is the latest speed app?
No , 6.10 and 6.08 should have about the same performance. Note that WUs vary in runtime by ca. 20%.
CU
Bikeman
Well then something went wrong. I was running 6.07 (which should have the same performance as 6.08 and 6.10 right?), then I switched to 6.09 and the CPU times went back to the 6.04 times. Then when I switched to 6.10 they increased to about 12 hours on a 2.5 GHz C2D... that can't be right, right?
this is the host: http://einsteinathome.org/host/1615446/tasks
Is 6.10 only a bugfix or a speed-app? I'm still running 6.05 without any problems with checkpoints...
ah, so that's why suddenly the CPU time required increased so much.. so I'm right in thinking that 6.08 is the latest speed app?
No , 6.10 and 6.08 should have about the same performance. Note that WUs vary in runtime by ca. 20%.
CU
Bikeman
Well then something went wrong. I was running 6.07 (which should have the same performance as 6.08 and 6.10 right?), then I switched to 6.09 and the CPU times went back to the 6.04 times. Then when I switched to 6.10 they increased to about 12 hours on a 2.5 GHz C2D... that can't be right, right?
this is the host: http://einsteinathome.org/host/1615446/tasks
No, this seems to be in-line with expected runtime, because the WUs that took really long (e.g. this one) are very close to the maximum expected runtime.
This thread is about how to compare app performance taking the varying WU runtime into account.
A "pencil and paper" method to predict if the runtime of two WUs are expected to be comparable is given in this message of that thread. When you follow the instructions there, you will see that the "period" of the runtime variations for WUs in the frequency range of your test units is 178 : The WUs which are slowest will have sequence numbers that are close to the multiples of 178. The fastest WUs will lie exactly between two multiples of 178 (so 1.5 * 178, 2.5 * 178 ... etc.).
Your slowest WUs had sequence numbers 361 and 362, very close to 2 * 178 = 356.
Your fastest WU had sequence number 605 , which is ~3.4 * 178 , very close to the runtime minimum.
So nothing to worry about performance wise, it was just bad luck that you happened to get very "fast" WUs for one app and very "slow" WUs for another version.
(Copied from Arecibo Binary Pulsar Search ABP1 thread)
Quote:
Is there a way for us that is using app_info to run beta 6.10 for win to get the files to run both. I´m wondering because I saw the files listed in current version section (3.02 for win).
You could manually download the einsteinbinary_ABP1_3.02* files from http://einstein.phys.uwm.edu/download/ and add a section for app einsteinbinary_ABP1 to the app_info.xml. However I doubt that this would make much sense right now as we are not issuing any ABP1 work in the next weeks and furthermore you'd miss updates to the einsteinbinary application.
Better help us test the 6.10 Windows Beta App so we can make it official and then switch back to the "official" path.
RE: RE: Is 6.10 only a
)
No , 6.10 and 6.08 should have about the same performance. Note that WUs vary in runtime by ca. 20%.
CU
Bikeman
Very odd. 1096885599 finished
)
Very odd. 1096885599 finished but took an unusual amount of time to complete. Over 16 hours. Done on a Dual core E2180
112398573 took the normal just under 12 hours. Done on AMD 64 3800+
Should also mention of the AMD I ran the new CUDA app along side the Einstein.
On the dual core I ran the AP opti app along side Einstien, as I have most of the time I run tasks on this pc.
Must have put the wrong path
)
Must have put the wrong path to the 1st task lol, not sure why I linked to a seti task, but the rest of the info on the task is correct.
For me the new app works
)
For me the new app works fine.
automatically uses the sse2 application einstein_S5R4_6.10_windows_intelx86_2.exe
time is faster than the 6.04 Version
average time 6.04: 10,5 hours
time 6.10: 8,7 hours
any plans for a sse3 or ssse3 version
Graphic works fine for me
right client
resultid=112610303
edit wrong client but also new app
resultid=112778978
Matthias
RE: RE: RE: Is 6.10
)
Well then something went wrong. I was running 6.07 (which should have the same performance as 6.08 and 6.10 right?), then I switched to 6.09 and the CPU times went back to the 6.04 times. Then when I switched to 6.10 they increased to about 12 hours on a 2.5 GHz C2D... that can't be right, right?
this is the host:
http://einsteinathome.org/host/1615446/tasks
RE: RE: RE: RE: Is
)
No, this seems to be in-line with expected runtime, because the WUs that took really long (e.g. this one) are very close to the maximum expected runtime.
This thread is about how to compare app performance taking the varying WU runtime into account.
A "pencil and paper" method to predict if the runtime of two WUs are expected to be comparable is given in this message of that thread. When you follow the instructions there, you will see that the "period" of the runtime variations for WUs in the frequency range of your test units is 178 : The WUs which are slowest will have sequence numbers that are close to the multiples of 178. The fastest WUs will lie exactly between two multiples of 178 (so 1.5 * 178, 2.5 * 178 ... etc.).
Your slowest WUs had sequence numbers 361 and 362, very close to 2 * 178 = 356.
Your fastest WU had sequence number 605 , which is ~3.4 * 178 , very close to the runtime minimum.
So nothing to worry about performance wise, it was just bad luck that you happened to get very "fast" WUs for one app and very "slow" WUs for another version.
CU
Bikeman
Completed first task on ME
)
Completed first task on ME OS. Screen saver works, show graphics works, average running time.
Another one completed, no
)
Another one completed, no problems on my E2180 dual core 112653314.
Screen saver works, Show graphics works. Same run time as the last.
(Copied from Arecibo Binary
)
(Copied from Arecibo Binary Pulsar Search ABP1 thread)
You could manually download the einsteinbinary_ABP1_3.02* files from http://einstein.phys.uwm.edu/download/ and add a section for app einsteinbinary_ABP1 to the app_info.xml. However I doubt that this would make much sense right now as we are not issuing any ABP1 work in the next weeks and furthermore you'd miss updates to the einsteinbinary application.
Better help us test the 6.10 Windows Beta App so we can make it official and then switch back to the "official" path.
BM
BM
112676015 done on E2180 Vista
)
112676015 done on E2180 Vista Home Premium took over 16.5 hours.
112905237 done on AMD Athlon 64 Processor 3800+ XP Home took under 11