No WU finished yet, but I'm already very happy ...
The prior official version always took 100% idletime, resulting in the frightening heating of the CPU and rendering utterly useless a utility I used to "cool" the CPU during idletime. But now both processes share the idletime, meaning that einstein now will take longer in realtime to complete a WU and give the "Coolingtool" a chance (now 50% of idletime instead of 0%, as with einstein4.76) ... My temperatures with the old version were at about 70°C, now I'm crunching at halfspeed but have only got 55°C CPU. Now my machine will survive and therefore crunch longer...
Ok, so far I can confirm that there is indeed a slowdown, which may show up on some machines more than on others, and is has to do solely with the version of boinc we are using, not with our science code and not with the compiler. Seems that the fixes put in lately slow it down noticably. I will continue to investigate this.
Ok, so far I can confirm that there is indeed a slowdown, which may show up on some machines more than on others, and is has to do solely with the version of boinc we are using, not with our science code and not with the compiler. Seems that the fixes put in lately slow it down noticably. I will continue to investigate this.
BM
It would be wonderful if you could optimize the BOINC code and make it run faster. This would be great news for all projects that utilize the code.
Ok, so far I can confirm that there is indeed a slowdown, which may show up on some machines more than on others, and is has to do solely with the version of boinc we are using, not with our science code and not with the compiler. Seems that the fixes put in lately slow it down noticably. I will continue to investigate this.
Just a note to clarify, You're talking about the server side BOINC code, and not the core client versions that the end users are running, right? I don't want some users thinking that they can roll back their core client and see their speeds increase. ;)
As long as theres some value to the new changes in the library code, the slowdown isn't too big a deal but it is noticable. I wish us Windows crunchers were getting a speed increase like the Mac Altivec users are though!
Just a note to clarify, You're talking about the server side BOINC code, and not the core client versions that the end users are running, right? I don't want some users thinking that they can roll back their core client and see their speeds increase. ;)
Sorry, this needs some clarification indeed.
The boinc code can roughly be divided in "Server", "Client" and "Library". The Server code (which, in turn, has many parts) is what's running on our server, the Client code is what the "Boinc Core Client" you download is built from, and the Library code ist what is linked into the Applications (and the Core Client as well) that are downloaded to your machine by the Client. The slowdown happens in the latter part, the boinc library, which the user comes to see only as undistinguishable part of the Applications. The Application consist mainly of two parts: the "Worker" or science code which does the actual science work and number crunching, and the boinc library, which makes it fit into the rest of the boinc framework (e.g. handles communication with the core client, encapsulates system-dependent stuff like filename handling and some graphics stuff asf.).
So as a fix for the current slowdown we don't need to change something on our server and you don't need to change the BOINC Core Client on your machines, but we need to build new Apps when we found and fixed the problem.
Just a note to clarify, You're talking about the server side BOINC code, and not the core client versions that the end users are running, right? I don't want some users thinking that they can roll back their core client and see their speeds increase. ;)
Sorry, this needs some clarification indeed.
The boinc code can roughly be divided in "Server", "Client" and "Library". The Server code (which, in turn, has many parts) is what's running on our server, the Client code is what the "Boinc Core Client" you download is built from, and the Library code ist what is linked into the Applications (and the Core Client as well) that are downloaded to your machine by the Client. The slowdown happens in the latter part, the boinc library, which the user comes to see only as undistinguishable part of the Applications. The Application consist mainly of two parts: the "Worker" or science code which does the actual science work and number crunching, and the boinc library, which makes it fit into the rest of the boinc framework (e.g. handles communication with the core client, encapsulates system-dependent stuff like filename handling and some graphics stuff asf.).
So as a fix for the current slowdown we don't need to change something on our server and you don't need to change the BOINC Core Client on your machines, but we need to build new Apps when we found and fixed the problem.
Does this make it a bit clearer?
BM
A lot...but again, when a new app is released (newer than both 4.79 and 0.03) will our BOINC client automatically update it?
A lot...but again, when a new app is released (newer than both 4.79 and 0.03) will our BOINC client automatically update it?
Now I get it!
No, with participating in this beta test you actually switched the platform from "x86-windows" to "anonymous". The client will never replace an anonymous App by itself. However the machines that are still running the "official" 4.79 App will download and use a newer Windows App when it is available automatically.
Just got .3 started on my 3.2 HT box; Radeon 300 series; using BOINC 4.45. (Under BOINCview, this shouldn't make a difference?) Don't know if I have the latest drivers for the card.
Crunching proceeding normally, messages were as you said to expect; will report on times later.
Before installing .3, "Show Graphics" *did* work for me; although they definitely hadn't on earlier BOINC versions, for sure not on 4.25. Sorry I can't say exactly which version they started working for, since I hadn't tried until 4.45.
As to the screensaver: With 4.79, the screen would go black, the starsphere would show for about a second, go black again, and then return to the desktop. No lockups or GPF of any kind after the return.
With .3 it is exactly the same, "Show Graphics" works fine (and is beautiful), but the screensaver acts exactly as described for 4.79.
No WU finished yet, but I'm
)
No WU finished yet, but I'm already very happy ...
The prior official version always took 100% idletime, resulting in the frightening heating of the CPU and rendering utterly useless a utility I used to "cool" the CPU during idletime. But now both processes share the idletime, meaning that einstein now will take longer in realtime to complete a WU and give the "Coolingtool" a chance (now 50% of idletime instead of 0%, as with einstein4.76) ... My temperatures with the old version were at about 70°C, now I'm crunching at halfspeed but have only got 55°C CPU. Now my machine will survive and therefore crunch longer...
Ok, so far I can confirm that
)
Ok, so far I can confirm that there is indeed a slowdown, which may show up on some machines more than on others, and is has to do solely with the version of boinc we are using, not with our science code and not with the compiler. Seems that the fixes put in lately slow it down noticably. I will continue to investigate this.
BM
BM
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/account/tasks
most results took about 21,000 sec. even w/ Boinc 4.45
http://einsteinathome.org/task/5636660
http://einsteinathome.org/task/5645806
The top one took about 25,000 seconds with E@H 0.03 and the bottom one shows computation error
RE: Ok, so far I can
)
It would be wonderful if you could optimize the BOINC code and make it run faster. This would be great news for all projects that utilize the code.
Good luck!
RE: Ok, so far I can
)
Just a note to clarify, You're talking about the server side BOINC code, and not the core client versions that the end users are running, right? I don't want some users thinking that they can roll back their core client and see their speeds increase. ;)
As long as theres some value to the new changes in the library code, the slowdown isn't too big a deal but it is noticable. I wish us Windows crunchers were getting a speed increase like the Mac Altivec users are though!
RE: Just a note to clarify,
)
Sorry, this needs some clarification indeed.
The boinc code can roughly be divided in "Server", "Client" and "Library". The Server code (which, in turn, has many parts) is what's running on our server, the Client code is what the "Boinc Core Client" you download is built from, and the Library code ist what is linked into the Applications (and the Core Client as well) that are downloaded to your machine by the Client. The slowdown happens in the latter part, the boinc library, which the user comes to see only as undistinguishable part of the Applications. The Application consist mainly of two parts: the "Worker" or science code which does the actual science work and number crunching, and the boinc library, which makes it fit into the rest of the boinc framework (e.g. handles communication with the core client, encapsulates system-dependent stuff like filename handling and some graphics stuff asf.).
So as a fix for the current slowdown we don't need to change something on our server and you don't need to change the BOINC Core Client on your machines, but we need to build new Apps when we found and fixed the problem.
Does this make it a bit clearer?
BM
BM
RE: RE: Just a note to
)
A lot...but again, when a new app is released (newer than both 4.79 and 0.03) will our BOINC client automatically update it?
RE: A lot...but again, when
)
Now I get it!
No, with participating in this beta test you actually switched the platform from "x86-windows" to "anonymous". The client will never replace an anonymous App by itself. However the machines that are still running the "official" 4.79 App will download and use a newer Windows App when it is available automatically.
Does this answer your question?
BM
BM
RE: A lot...but again, when
)
I have the same question. Now it feels like the app_info.xml file inside project/einst../ folder stops it from doing that.. ???
Edit* Never mind.. saw your answer above ;)
so no need for more testing from me and I can switch back to 4.79 then. :)
Hi, Just got .3 started on
)
Hi,
Just got .3 started on my 3.2 HT box; Radeon 300 series; using BOINC 4.45. (Under BOINCview, this shouldn't make a difference?) Don't know if I have the latest drivers for the card.
Crunching proceeding normally, messages were as you said to expect; will report on times later.
Before installing .3, "Show Graphics" *did* work for me; although they definitely hadn't on earlier BOINC versions, for sure not on 4.25. Sorry I can't say exactly which version they started working for, since I hadn't tried until 4.45.
As to the screensaver: With 4.79, the screen would go black, the starsphere would show for about a second, go black again, and then return to the desktop. No lockups or GPF of any kind after the return.
With .3 it is exactly the same, "Show Graphics" works fine (and is beautiful), but the screensaver acts exactly as described for 4.79.
HTH
Ken