Near the other extreme from Vista, I have a Coppermine CPU Windows98 SE host which recently completed its first 4.26 result.
On a model match basis, it seems to have used 92.3% of the CPU time I'd have predicted for 4.15.
The model I used is new. The odd waveshape of the cyclic CPU time dependency on sequence (task) number within a frequency which Richard Haselgrove noted many months ago has long suggested to me a sin wave with a fixed portion of each 2 Pi cycle omitted.
I finally acted on that notion yesterday. I've started a new "Cycle model" thread with my current version.
I'm running 4.26 under Windows XP Pro 64-bit edition, BOINC v5.10.13 (also 64 bit build) so far without issue.
However, I've only crunched one WU with it so far and there was absolutely no speedup in crunching time over the stock app.... :-( I'll see how things go.
This (validated) result is a v4.25/v4.26 hybrid - about 25%/75%. I couldn't make any direct speed comparisons; but, it did finish a little more than a hour quicker than BOINC originally estimated. That's pretty much in-line with the improvement others have been reporting.
First 4.26 WU completed, validated with a wingman running the stock 4.20 Linux application. Despite being downloaded for the stock 4.15 application, the WU was processed completely by the 4.26 application. Since the WU is from the same batch that this particular host has worked on recently, there appears to be ~20% speed improvement.
Something seriously broke on my vista laptop when I tried to switch over from 4.15. UNfortunately nothing dramatic like a crash with a helpful error code.
It's the only machine I have boinc running as a service, so that might've been related somehow.
I file-exited the GUI client. I opened services.msc and stopped hte boinc service. I then copied the entire boinc directory, pasted the 4.26 app, etc in. I then renamed a copy of 4.26.exe and .pdb to the 4.15 names to switch the current in progress WUs over. Once that was done I restarted the service and boinc manager GUI. Task manager showed the einstien processes averaging 10-40% CPU share each with most of the remaing 20-80% being wasted via system idle process. I rebooted and saw the same thing.
I then stopped everything again, and restored the original unmucked with boinc directory. And following a reboot everything resumed running normally.
Task manager showed the einstien processes averaging 10-40% CPU share each with most of the remaing 20-80% being wasted via system idle process. I rebooted and saw the same thing.
How long did you wait?
When new results start being processed on my 3 GHz Conroe machines, there is an initial state in which each Einstein process gets significantly less than its share of the available CPU, some more is consumed in a System process, and some more in idle. This condition ends abruptly after something like a minute. That time would presumably be much longer on a less capable CPU.
Just wondering--perhaps your abnormal condition would have resolved itself?
Something seriously broke on my vista laptop when I tried to switch over from 4.15. UNfortunately nothing dramatic like a crash with a helpful error code.
It's the only machine I have boinc running as a service, so that might've been related somehow.
I file-exited the GUI client. I opened services.msc and stopped hte boinc service. I then copied the entire boinc directory, pasted the 4.26 app, etc in. I then renamed a copy of 4.26.exe and .pdb to the 4.15 names to switch the current in progress WUs over. Once that was done I restarted the service and boinc manager GUI. Task manager showed the einstien processes averaging 10-40% CPU share each with most of the remaing 20-80% being wasted via system idle process. I rebooted and saw the same thing.
I then stopped everything again, and restored the original unmucked with boinc directory. And following a reboot everything resumed running normally.
Edit: Yonah core and vista32 premium
Why do what you did? All you have to do is shut everything down and bring in the new application EXEs, PDB, and app_info.xml. Just because on your side the tasks you already had are still identified with 4.15 is no problem. It is cosmetic to you only. When the task runs it properly identifies the application version in the result output.
Why do what you did? All you have to do is shut everything down and bring in the new application EXEs, PDB, and app_info.xml. Just because on your side the tasks you already had are still identified with 4.15 is no problem. It is cosmetic to you only. When the task runs it properly identifies the application version in the result output.
Is the thread on switching apps mid result out of date? It indicated that hte the EXE replacement was needed to switch over already DLed tasks.
I'm not sure how long I waited before doing the reboot, probably a minute or three since I was half doing something on my desktop and only semi paying attention to what to activity on the laptop.
Why do what you did? All you have to do is shut everything down and bring in the new application EXEs, PDB, and app_info.xml. Just because on your side the tasks you already had are still identified with 4.15 is no problem. It is cosmetic to you only. When the task runs it properly identifies the application version in the result output.
Is the thread on switching apps mid result out of date? It indicated that hte the EXE replacement was needed to switch over already DLed tasks.
I would guess so. I was the first person to reply in this thread and asked about mid-task switching. I decided to just do it the most logical way, which was to replace the app_info.xml and not muck with anything else... It worked fine. This isn't a service install, but I can't see it being any different if it were, as you'd only have to make sure that BOINC isn't running...
Have just added the 4.26 app on my Opteron 285 that was returning 40,000 seconds plus for the 4.15 app, it will be nice to see if there is any improvement.
A 10% improvement down to 36,000 seconds would be very nice, 20% a dream at 32,000 seconds.
It will then compare to my other Opteron 285 running Beta app 4.14 on Linux.
I think I now have running
Windows 4.15 on AMD X2 4800 @2.4GHz taking between 41,100 to 50,000 seconds;
Windows 4.15 on AMD Opteron 285 @2.6GHZ was taking 40,000 seconds plus;
Windows 4.26 on AMD Opteron 285 @2.6GHZ yet to process a WU;
Linux 4.14 on AMD Opteron 275 @2.2GHZ taking between 37,600 and 50,000 seconds;
Linux 4.14 on AMD Opteron 285 @2.6GHZ taking between 32,000 and 46,000 seconds;
Linux 4.27 on AMD Opteron 285 @2.6GHZ yet to process a WU.
Near the other extreme from
)
Near the other extreme from Vista, I have a Coppermine CPU Windows98 SE host which recently completed its first 4.26 result.
On a model match basis, it seems to have used 92.3% of the CPU time I'd have predicted for 4.15.
The model I used is new. The odd waveshape of the cyclic CPU time dependency on sequence (task) number within a frequency which Richard Haselgrove noted many months ago has long suggested to me a sin wave with a fixed portion of each 2 Pi cycle omitted.
I finally acted on that notion yesterday. I've started a new "Cycle model" thread with my current version.
I'm running 4.26 under
)
I'm running 4.26 under Windows XP Pro 64-bit edition, BOINC v5.10.13 (also 64 bit build) so far without issue.
However, I've only crunched one WU with it so far and there was absolutely no speedup in crunching time over the stock app.... :-( I'll see how things go.
TTFN - Pete.
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/task/91183370
This (validated) result is a v4.25/v4.26 hybrid - about 25%/75%. I couldn't make any direct speed comparisons; but, it did finish a little more than a hour quicker than BOINC originally estimated. That's pretty much in-line with the improvement others have been reporting.
First 4.26 WU completed,
)
First 4.26 WU completed, validated with a wingman running the stock 4.20 Linux application. Despite being downloaded for the stock 4.15 application, the WU was processed completely by the 4.26 application. Since the WU is from the same batch that this particular host has worked on recently, there appears to be ~20% speed improvement.
Soli Deo Gloria
Something seriously broke on
)
Something seriously broke on my vista laptop when I tried to switch over from 4.15. UNfortunately nothing dramatic like a crash with a helpful error code.
It's the only machine I have boinc running as a service, so that might've been related somehow.
I file-exited the GUI client. I opened services.msc and stopped hte boinc service. I then copied the entire boinc directory, pasted the 4.26 app, etc in. I then renamed a copy of 4.26.exe and .pdb to the 4.15 names to switch the current in progress WUs over. Once that was done I restarted the service and boinc manager GUI. Task manager showed the einstien processes averaging 10-40% CPU share each with most of the remaing 20-80% being wasted via system idle process. I rebooted and saw the same thing.
I then stopped everything again, and restored the original unmucked with boinc directory. And following a reboot everything resumed running normally.
Edit: Yonah core and vista32 premium
RE: Task manager showed
)
How long did you wait?
When new results start being processed on my 3 GHz Conroe machines, there is an initial state in which each Einstein process gets significantly less than its share of the available CPU, some more is consumed in a System process, and some more in idle. This condition ends abruptly after something like a minute. That time would presumably be much longer on a less capable CPU.
Just wondering--perhaps your abnormal condition would have resolved itself?
RE: Something seriously
)
Why do what you did? All you have to do is shut everything down and bring in the new application EXEs, PDB, and app_info.xml. Just because on your side the tasks you already had are still identified with 4.15 is no problem. It is cosmetic to you only. When the task runs it properly identifies the application version in the result output.
RE: Why do what you did?
)
Is the thread on switching apps mid result out of date? It indicated that hte the EXE replacement was needed to switch over already DLed tasks.
I'm not sure how long I waited before doing the reboot, probably a minute or three since I was half doing something on my desktop and only semi paying attention to what to activity on the laptop.
RE: RE: Why do what you
)
I would guess so. I was the first person to reply in this thread and asked about mid-task switching. I decided to just do it the most logical way, which was to replace the app_info.xml and not muck with anything else... It worked fine. This isn't a service install, but I can't see it being any different if it were, as you'd only have to make sure that BOINC isn't running...
Have just added the 4.26 app
)
Have just added the 4.26 app on my Opteron 285 that was returning 40,000 seconds plus for the 4.15 app, it will be nice to see if there is any improvement.
A 10% improvement down to 36,000 seconds would be very nice, 20% a dream at 32,000 seconds.
It will then compare to my other Opteron 285 running Beta app 4.14 on Linux.
I think I now have running
Windows 4.15 on AMD X2 4800 @2.4GHz taking between 41,100 to 50,000 seconds;
Windows 4.15 on AMD Opteron 285 @2.6GHZ was taking 40,000 seconds plus;
Windows 4.26 on AMD Opteron 285 @2.6GHZ yet to process a WU;
Linux 4.14 on AMD Opteron 275 @2.2GHZ taking between 37,600 and 50,000 seconds;
Linux 4.14 on AMD Opteron 285 @2.6GHZ taking between 32,000 and 46,000 seconds;
Linux 4.27 on AMD Opteron 285 @2.6GHZ yet to process a WU.
So I will be able to do some comparing.