I know how to check the temperature of a Nvidia GPU through nvidia-settings or nvidia control panel programs, but not how to establish an upper limit.
Nvidia-settings allows setting a temperature target when 'coolbits' has been enabled. Here is a link that describes how to set coolbits . Scroll down near the bottom for an explanation of each of the bit settings.
Nvidia-settings allows setting a temperature target when 'coolbits' has been enabled.
This is for Linux. Under Win it's far easier (well, for me) - just use the usual OC utilities. On nVidia Kepler GPUs and newer you can set a temperature target and the card will adjust its boost state (clock & voltage) to not exceed that. For other GPUs it's not as direct, but besides controlling the power draw (by setting clock speed and voltage) you can modifiy the fan curve, as archae86 said.
Or at least I think it is.. As I understand it BRP6 Parkes 1.50 tasks are allocated to diff GPU's by E@H's servers. ie:- ! copy of the WU goes to either and ATI or Intel GPU and the next 1 to another brand like NVidia..
I understood this was so 2 same brand GPU's couldn't produce a valid result due to a bug in code, drivers or whatever.
However in the case of this WU:- PM0016_01941_340_2 its been completed by NV GPU's, my GTX980 and another crunchers GTX 970.
Are E@H servers supposed to ensure diff brand GPU's are picked after the 1st set is dispatched? or is that only for the first 2 task dispatches, after which its a free for all?
Anyone know what the score is?
[edit]
just checked, its an inconclusive...
BRP6 Parkes 1.50 tasks are allocated to diff GPU's by E@H's servers. ie:- ! copy of the WU goes to either and ATI or Intel GPU and the next 1 to another brand like NVidia..
Where did you get that impression?
I have both seen and observed Einstein to impose a restriction during beta testing that only one of the quorum partners would be running a particular beta test code, but I've not heard of a restriction by "brand".
If they did that, we Nvidia users would get considerably restricted, as Nvidia GPUs account for considerably more than half the total GPU productivity here at the moment.
I had never noticed this before, but I think Cliff is right. I just looked through all my Parkes BRP6 valid results and they all were crunched by an Nvidia and an ATI. The BRP4G tasks however seem to go to anyone. Is Cliff correct that the project managers don't want similar graphic platforms working on the same task to prevent errors in processing ending up in the database?
Hi Yacob, RE: I know
)
Hi Yacob,
Nvidia-settings allows setting a temperature target when 'coolbits' has been enabled. Here is a link that describes how to set coolbits . Scroll down near the bottom for an explanation of each of the bit settings.
Gord
RE: Nvidia-settings allows
)
This is for Linux. Under Win it's far easier (well, for me) - just use the usual OC utilities. On nVidia Kepler GPUs and newer you can set a temperature target and the card will adjust its boost state (clock & voltage) to not exceed that. For other GPUs it's not as direct, but besides controlling the power draw (by setting clock speed and voltage) you can modifiy the fan curve, as archae86 said.
MrS
Scanning for our furry friends since Jan 2002
Hi Folks, Just come across
)
Hi Folks,
Just come across something a little odd...
Or at least I think it is.. As I understand it BRP6 Parkes 1.50 tasks are allocated to diff GPU's by E@H's servers. ie:- ! copy of the WU goes to either and ATI or Intel GPU and the next 1 to another brand like NVidia..
I understood this was so 2 same brand GPU's couldn't produce a valid result due to a bug in code, drivers or whatever.
However in the case of this WU:- PM0016_01941_340_2 its been completed by NV GPU's, my GTX980 and another crunchers GTX 970.
Are E@H servers supposed to ensure diff brand GPU's are picked after the 1st set is dispatched? or is that only for the first 2 task dispatches, after which its a free for all?
Anyone know what the score is?
[edit]
just checked, its an inconclusive...
Cliff,
Been there, Done that, Still no damm T Shirt.
RE: BRP6 Parkes 1.50 tasks
)
Where did you get that impression?
I have both seen and observed Einstein to impose a restriction during beta testing that only one of the quorum partners would be running a particular beta test code, but I've not heard of a restriction by "brand".
If they did that, we Nvidia users would get considerably restricted, as Nvidia GPUs account for considerably more than half the total GPU productivity here at the moment.
Hi Archae86, semantics,
)
Hi Archae86,
semantics, try by type, or by kind or whatever..
Cliff,
Been there, Done that, Still no damm T Shirt.
I had never noticed this
)
I had never noticed this before, but I think Cliff is right. I just looked through all my Parkes BRP6 valid results and they all were crunched by an Nvidia and an ATI. The BRP4G tasks however seem to go to anyone. Is Cliff correct that the project managers don't want similar graphic platforms working on the same task to prevent errors in processing ending up in the database?
Cheers, Keith
RE: Is Cliff correct
)
No, there is no such restriction.
Here is an example of two AMD results validating each other. Here's another of two NVIDIA results.
Just pick the tasks list of any productive host with the particular brand GPU on the website. Examples like the above are easy to find.
EDIT: In fact, here is an example from one of your hosts.
Cheers,
Gary.
I tried a quick look by
)
I tried a quick look by picking one valid Parkes result from each of my three hosts.
One of those three shows as purely nvidia
I tried one more batch of three, and again one of the three was purely nvidia.
If there is some preference implemented in work dispatch of this particular type, it appears not currently to be very strictly observed.
RE: EDIT: In fact, here is
)
Guess I missed that one when I went and reviewed my BRP6 tasks completed.
Cheers, Keith