Insidious
Diamond Member
So we are all amazed at the PPD that comes from our NVidia cards and want to OC to get it maxed out.... Here's a couple tips that I have found useful when trying to decide 'how much is too much'
First of all, you have the option of OC'ing the core, the memory, and the shaders. To the best of my knowledge, only the shaders clock will have any significant affect on your PPD. So when I overclock, I uncheck the (RivaTuner) box for linking the shader clock to the core clock and just leave the core and memory at their defaults. This saves quite a bit in terms of heat (and of course, power consumption)
Secondly, the NVidia GPU client is actually pretty graceful when it does EUE. Except for extreme circumstances, you won't even know it happened unless you go scrolling through that huge fahlog.txt file... that's a genuine PITA.
Soooooooo, here's a shortcut.
The GPU client actually does a good job of cleaning out it's work folder when it is working properly (unlike the SMP client). If you are using fahmon to monitor your clients, just double-click on any client and it will open an explorer window for that client's work folder. Open the work folder. It is obvious which files are associated with a given 'queue slot'. If you see files for more than one slot (slots are numbered 0 thru 9) you have had EUEs!
Further, if you leave the remnants of these failed WUs in the folder, it can lead to mis-interpretation the next time that slot is used (not always) and your client will become more and more unstable as more remnants are left. So, look at the Fahlog.txt file to see what slot is currently being processed and then delete the files assiated with any other slot. This will help prevent some future EUEs.
Then........... lower your clocks for that client because that's where the EUE came from (almost) every time!
-Sid
YMMV :beer:
First of all, you have the option of OC'ing the core, the memory, and the shaders. To the best of my knowledge, only the shaders clock will have any significant affect on your PPD. So when I overclock, I uncheck the (RivaTuner) box for linking the shader clock to the core clock and just leave the core and memory at their defaults. This saves quite a bit in terms of heat (and of course, power consumption)
Secondly, the NVidia GPU client is actually pretty graceful when it does EUE. Except for extreme circumstances, you won't even know it happened unless you go scrolling through that huge fahlog.txt file... that's a genuine PITA.
Soooooooo, here's a shortcut.
The GPU client actually does a good job of cleaning out it's work folder when it is working properly (unlike the SMP client). If you are using fahmon to monitor your clients, just double-click on any client and it will open an explorer window for that client's work folder. Open the work folder. It is obvious which files are associated with a given 'queue slot'. If you see files for more than one slot (slots are numbered 0 thru 9) you have had EUEs!
Further, if you leave the remnants of these failed WUs in the folder, it can lead to mis-interpretation the next time that slot is used (not always) and your client will become more and more unstable as more remnants are left. So, look at the Fahlog.txt file to see what slot is currently being processed and then delete the files assiated with any other slot. This will help prevent some future EUEs.
Then........... lower your clocks for that client because that's where the EUE came from (almost) every time!
-Sid
YMMV :beer:
