If BOINC shows up listing OCL3 in the log for the 950 you should continue to get 1.28 tasks.
Tom M
the project actually only delivers the app to Pascal and newer with OCL2+. His 950 is Maxwell and excluded.
but when Keith tested with his 970, he saw almost no change in speed. it's possible some maxwell cards would see a benefit while others wont.
. . Sorry, I didn't see these two messages until after I posted that last one. It seems the message function doesn't update until you give a kick in its dignity. But my question stands. Is it possible to deceive the project that I actually have a 1050?
... But my question stands. Is it possible to deceive the project that I actually have a 1050?
Why would you want to do that?
The 1.28 app now seems to be mainstream and if your 950 is successfully using that app with such a good improvement, that should continue indefinitely. I wouldn't be doing anything that might upset that situation :-).
... But my question stands. Is it possible to deceive the project that I actually have a 1050?
Why would you want to do that?
The 1.28 app now seems to be mainstream and if your 950 is successfully using that app with such a good improvement, that should continue indefinitely. I wouldn't be doing anything that might upset that situation :-).
he got the 1.28 tasks downloaded while a 1050Ti was installed. Then swapped in the 950 to see if it worked and/or ran faster. But with the 950, the scheduler won’t send the 1.28 app anymore, probably just the 1.20 app
but to answer the question. If you edit the coproc_info file, and lock it from editing, you should be able to pass the scheduler checks. I’m almost certain they are using the compute capability for checking generation. BOINC doesn’t record “Maxwell” or “Pascal” but does record CC as 5.2 on Maxwell and 6.1 on Pascal. The scheduler probably has a rule that CC needs to >6 or something like that. If you edit the coproc file to say 6.1 then it’ll likely work to get the new task
... But my question stands. Is it possible to deceive the project that I actually have a 1050?
Why would you want to do that?
The 1.28 app now seems to be mainstream and if your 950 is successfully using that app with such a good improvement, that should continue indefinitely. I wouldn't be doing anything that might upset that situation :-).
. . Because even though the card runs the app superbly, the servers will not send the new app tasks to a machine running Maxwell cards. :(
but to answer the question. If you edit the coproc_info file, and lock it from editing, you should be able to pass the scheduler checks. I’m almost certain they are using the compute capability for checking generation. BOINC doesn’t record “Maxwell” or “Pascal” but does record CC as 5.2 on Maxwell and 6.1 on Pascal. The scheduler probably has a rule that CC needs to >6 or something like that. If you edit the coproc file to say 6.1 then it’ll likely work to get the new task
. . Thanks muchly Ian I will give that a go.
:(
. . Either I got the edit wrong or it is not the solution. Still getting 1.20 tasks.
he got the 1.28 tasks downloaded while a 1050Ti was installed. Then swapped in the 950 to see if it worked and/or ran faster. But with the 950, the scheduler won’t send the 1.28 app anymore, probably just the 1.20 app
but to answer the question. If you edit the coproc_info file, and lock it from editing, you should be able to pass the scheduler checks. I’m almost certain they are using the compute capability for checking generation. BOINC doesn’t record “Maxwell” or “Pascal” but does record CC as 5.2 on Maxwell and 6.1 on Pascal. The scheduler probably has a rule that CC needs to >6 or something like that. If you edit the coproc file to say 6.1 then it’ll likely work to get the new task
I tested that with GTX 960 in Windows 10. It worked the way you said. I run the tasks with maximum power limit (max power 46%).
. . That is encouraging. May I ask where you changed the values? I changed what looked like them but it did nothing.
Stephen
what section did you change? You should see a section about compute capability, with 5 and 2 subsections below it. Change these to 6 and 1.
to “lock” the file under Linux you need to do two things. Set the file to read only (can be done in the GUI under properties/preferences, Permissions tab). And you need to set the file to be immutable, has to be done in the command line (in the directory where the coproc file is).
Ian&Steve C. wrote:Tom M
)
. . Sorry, I didn't see these two messages until after I posted that last one. It seems the message function doesn't update until you give a kick in its dignity. But my question stands. Is it possible to deceive the project that I actually have a 1050?
Stephen
Ian&Steve C. wrote:Tom M
)
Sorry again, this was an 'oopsy'
Stephen "Heretic wrote:...
)
Why would you want to do that?
The 1.28 app now seems to be mainstream and if your 950 is successfully using that app with such a good improvement, that should continue indefinitely. I wouldn't be doing anything that might upset that situation :-).
Cheers,
Gary.
New dev driver 510.10, works
)
New dev driver 510.10, works great
https://forums.guru3d.com/threads/nvidia-geforce-510-10-windows-11-dev-driver.439593/
Gary Roberts wrote:Stephen
)
he got the 1.28 tasks downloaded while a 1050Ti was installed. Then swapped in the 950 to see if it worked and/or ran faster. But with the 950, the scheduler won’t send the 1.28 app anymore, probably just the 1.20 app
but to answer the question. If you edit the coproc_info file, and lock it from editing, you should be able to pass the scheduler checks. I’m almost certain they are using the compute capability for checking generation. BOINC doesn’t record “Maxwell” or “Pascal” but does record CC as 5.2 on Maxwell and 6.1 on Pascal. The scheduler probably has a rule that CC needs to >6 or something like that. If you edit the coproc file to say 6.1 then it’ll likely work to get the new task
_________________________________________________________________________
Gary Roberts wrote: Stephen
)
. . Because even though the card runs the app superbly, the servers will not send the new app tasks to a machine running Maxwell cards. :(
. . A bit of a misjudgement perhaps ...
Stephen
Ian&Steve C. wrote:but to
)
. . Thanks muchly Ian I will give that a go.
:(
. . Either I got the edit wrong or it is not the solution. Still getting 1.20 tasks.
Stephen
Ian&Steve C. wrote:he got the
)
I tested that with GTX 960 in Windows 10. It worked the way you said. I run the tasks with maximum power limit (max power 46%).
https://einsteinathome.org/host/12650852
The normal app version that Maxwell's will run in Windows is 1.22 ...
https://einsteinathome.org/task/1165638582 (run time 1706 sec)
And coproc_info hacked with values "6" and"1" this GPU got app version 1.28 . It run without problems.
https://einsteinathome.org/task/1165685516 (run time 1183 sec)
. . That is encouraging. May
)
. . That is encouraging. May I ask where you changed the values? I changed what looked like them but it did nothing.
Stephen
Stephen "Heretic wrote: . .
)
what section did you change? You should see a section about compute capability, with 5 and 2 subsections below it. Change these to 6 and 1.
to “lock” the file under Linux you need to do two things. Set the file to read only (can be done in the GUI under properties/preferences, Permissions tab). And you need to set the file to be immutable, has to be done in the command line (in the directory where the coproc file is).
sudo chattr +i coproc_info.xmlthen BOINC can’t change the file at startup.
_________________________________________________________________________