I think they're talking "only" a bug nVidia introduced in the 270 series drivers. Older ones are fine and your current setting is good. It's been said the BOINC developers are working with nVidia to fix it.
I think they're talking "only" a bug nVidia introduced in the 270 series drivers. Older ones are fine and your current setting is good. It's been said the BOINC developers are working with nVidia to fix it.
MrS
Well I'll hold off then. I'm using driver version 266.
32bit Windows XP Home
AMD Opteron 180
ASUS A8N-SLI Motherboard
Nvidia 450GTS GPU
4GB DDR 400 Memory
I got a problem. So long as E@H is on my system, whether I turn boinc off or not I still get chop. Its like its not offloading or something properly. I guess it makes sense, it downloads so much stuff just to run the project. I'd like to help but I cant have it hogging my system like this.
I am going to try suspending it and see if that helps.
32bit Windows XP Home
AMD Opteron 180
ASUS A8N-SLI Motherboard
Nvidia 450GTS GPU
4GB DDR 400 Memory
FWIW - Under Windows 7 on this machine and even prior to having two video cards in it, when I want to play some video intensive game ( Supreme Commander is my fave ) :
- I suspend BOINC
- play the game, and when finished ....
- reboot the machine
- un-suspend BOINC
When I didn't do it that way either the CUDA WU's stuff up ( typically reported as in error, but occasionally not and then proceeding through to not validating in the quorum ) and/or I get funny visual behaviours in the game. Richard had earlier pointed out to me an issue with driver swapping ( regarding remote sessions into CUDA running machines ) - perhaps that is relevant here.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I think they're talking "only" a bug nVidia introduced in the 270 series drivers. Older ones are fine and your current setting is good. It's been said the BOINC developers are working with nVidia to fix it.
nope - this problem is very old. i remember the same thing happening back then in the early days on GPUgrid. it's just that some driver versions are more stable than others. and the current 270.xx ones are worse than others..
When I didn't do it that way either the CUDA WU's stuff up ( typically reported as in error, but occasionally not and then proceeding through to not validating in the quorum ) and/or I get funny visual behaviours in the game. Richard had earlier pointed out to me an issue with driver swapping ( regarding remote sessions into CUDA running machines ) - perhaps that is relevant here.
you can not use windows remote desktop or switch users while running cuda-jobs!
When I didn't do it that way either the CUDA WU's stuff up ( typically reported as in error, but occasionally not and then proceeding through to not validating in the quorum ) and/or I get funny visual behaviours in the game. Richard had earlier pointed out to me an issue with driver swapping ( regarding remote sessions into CUDA running machines ) - perhaps that is relevant here.
you can not use windows remote desktop or switch users while running cuda-jobs!
Exactly. So maybe a swapping of driver sets - as occurs with remote sessions - is responsible for similiar issues when he games. Probably neither Microsoft nor game developers were expecting GPU's to be used for DC purposes, thus wouldn't have accounted for the BOINC expectation of a persistent state of the resource. I don't know, but perhaps there is a significant difference when games are enacted in windowed vs full screen mode .... does a user initiated BOINC suspension trigger the creation of a checkpoint?
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
When I didn't do it that way either the CUDA WU's stuff up ( typically reported as in error, but occasionally not and then proceeding through to not validating in the quorum ) and/or I get funny visual behaviours in the game. Richard had earlier pointed out to me an issue with driver swapping ( regarding remote sessions into CUDA running machines ) - perhaps that is relevant here.
you can not use windows remote desktop or switch users while running cuda-jobs!
SUPPOSEDLY there is a new Boinc version coming out that works around this problem, I am using the latest released version and it is not mentioned though.
I'm confused (normal
)
I'm confused (normal situation).
I've been installing e@h on clients machines, with their permission.
I've been setting "Suspend GPU work while computer is in use"
It sounds like this is a bad idea to go along with the recommendation to let the machine run all the time if possible.
Should I disable GPU usage altogether in Boinc? It cannot interfere with normal operation if I want them to agree.
Joe
I think they're talking
)
I think they're talking "only" a bug nVidia introduced in the 270 series drivers. Older ones are fine and your current setting is good. It's been said the BOINC developers are working with nVidia to fix it.
MrS
Scanning for our furry friends since Jan 2002
RE: I think they're talking
)
Well I'll hold off then. I'm using driver version 266.
32bit Windows XP Home
AMD Opteron 180
ASUS A8N-SLI Motherboard
Nvidia 450GTS GPU
4GB DDR 400 Memory
I got a problem. So long as
)
I got a problem. So long as E@H is on my system, whether I turn boinc off or not I still get chop. Its like its not offloading or something properly. I guess it makes sense, it downloads so much stuff just to run the project. I'd like to help but I cant have it hogging my system like this.
I am going to try suspending it and see if that helps.
32bit Windows XP Home
AMD Opteron 180
ASUS A8N-SLI Motherboard
Nvidia 450GTS GPU
4GB DDR 400 Memory
Suspending didnt work. Only
)
Suspending didnt work. Only detaching from project and rebooting helped. Ah well.
32bit Windows XP Home
AMD Opteron 180
ASUS A8N-SLI Motherboard
Nvidia 450GTS GPU
4GB DDR 400 Memory
FWIW - Under Windows 7 on
)
FWIW - Under Windows 7 on this machine and even prior to having two video cards in it, when I want to play some video intensive game ( Supreme Commander is my fave ) :
- I suspend BOINC
- play the game, and when finished ....
- reboot the machine
- un-suspend BOINC
When I didn't do it that way either the CUDA WU's stuff up ( typically reported as in error, but occasionally not and then proceeding through to not validating in the quorum ) and/or I get funny visual behaviours in the game. Richard had earlier pointed out to me an issue with driver swapping ( regarding remote sessions into CUDA running machines ) - perhaps that is relevant here.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: I think they're talking
)
nope - this problem is very old. i remember the same thing happening back then in the early days on GPUgrid. it's just that some driver versions are more stable than others. and the current 270.xx ones are worse than others..
RE: When I didn't do it
)
you can not use windows remote desktop or switch users while running cuda-jobs!
RE: RE: When I didn't do
)
Exactly. So maybe a swapping of driver sets - as occurs with remote sessions - is responsible for similiar issues when he games. Probably neither Microsoft nor game developers were expecting GPU's to be used for DC purposes, thus wouldn't have accounted for the BOINC expectation of a persistent state of the resource. I don't know, but perhaps there is a significant difference when games are enacted in windowed vs full screen mode .... does a user initiated BOINC suspension trigger the creation of a checkpoint?
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: RE: When I didn't do
)
SUPPOSEDLY there is a new Boinc version coming out that works around this problem, I am using the latest released version and it is not mentioned though.