@Unwinder and/or others: I have an old card by modern standards (8600 GT), so it uses "driver level" control to work the fan profiles (rather than hardware-level).
 
I just installed the latest precision last night, and have had your RivaTuner installed for a while previous to that (and even tried MSI afterburner once upon a time).
 
In any case, I have my polling time set to a relatively short interval (1000ms = 1s), but in certain applications (read directX video games) it forgets to override the hardware.  My lowest fan speed set in Precision (or before that Riva Tuner) is 50%, but the fan drops to 30%.  The fan speed is being monitored, as RT graphs showed the fan at a high setting, then dropping to 30 and with the current version of Precision I get display of my fan speed and GPU temp on-screen.  But obviously the fan speed is not being set by the software.
 
In the past, I also noticed that if I had any CUDA tasks end (BOINC projects) my fans would reset to HW default (30%) and would sometimes not reactivate until a few seconds after I took my computer out of "idle" mode.  I assume that is still the case with Precision, but I haven't tested it.
 
Let me know if there's a bug submission form I should be using instead of these forums.
 
Is it something as simple as setting an application or thread to "high" priority so the app has enough clock cycles to trigger a fan update?  Or is there some lower level hardware setting or driver file that resets it to hardware standard if it hasn't "heard" from software otherwise for a while?
 
My CPU is usually 100% utilized, although I believe I have this problem even when not.

Hi!
 I see this release takes up 5MB private memory in process explorer, very nice! :)
But it places almost 900 files in its directory after install. I see most of the files are Localization files in the root and the Bundle\OSDServer. How about an option during install to choose language files?
 
Thank you for such an efficient program in this managed .net world :)

pagelm

@Unwinder and/or others: I have an old card by modern standards (8600 GT), so it uses "driver level" control to work the fan profiles (rather than hardware-level).

I just installed the latest precision last night, and have had your RivaTuner installed for a while previous to that (and even tried MSI afterburner once upon a time).

In any case, I have my polling time set to a relatively short interval (1000ms = 1s), but in certain applications (read directX video games) it forgets to override the hardware.  My lowest fan speed set in Precision (or before that Riva Tuner) is 50%, but the fan drops to 30%.  The fan speed is being monitored, as RT graphs showed the fan at a high setting, then dropping to 30 and with the current version of Precision I get display of my fan speed and GPU temp on-screen.  But obviously the fan speed is not being set by the software.

In the past, I also noticed that if I had any CUDA tasks end (BOINC projects) my fans would reset to HW default (30%) and would sometimes not reactivate until a few seconds after I took my computer out of "idle" mode.  I assume that is still the case with Precision, but I haven't tested it.

Let me know if there's a bug submission form I should be using instead of these forums.

Is it something as simple as setting an application or thread to "high" priority so the app has enough clock cycles to trigger a fan update?  Or is there some lower level hardware setting or driver file that resets it to hardware standard if it hasn't "heard" from software otherwise for a while?

My CPU is usually 100% utilized, although I believe I have this problem even when not.

 
Take a look at your power profiles on the machine in question and ensure that it's not throttling down the video when it goes into a 'idle but processing' type state. I know with Vista/Windows 7 there is an option to power down the video card itself when the display is only relatively used even though the CPU is being utilized quite a bit. Since the threads in question to control the fan profile might be idle and yet the CPU itself is being used it may go into that 'power down' state to allocate additional cycles for the processing of your task. I have my power profile set to never power down any hardware even going into a screen saver state so that I can ensure I won't need or have any phantom error message being tossed when it tries to restore the system back to an active state. Might be something to look into.
 
The 8600GT's, from what I understand, had a BIOS on the card that initially controlled the fans and you could modify it for different active states. I understand that somewhere along the line the software profile states could and did control the automatic fan profile somewhere down the line just before the 9000 series cards were put into the channels but sometimes the 8000 series cards sometimes ignored these software command states. Might be something to look into also. I certainly don't suggest modifying the BIOS of the card as doing so can upset the card itself and cause more issues than it can benefit from but depending on the driver you are using it may pay to revert back to one of the 19x.xx series drivers rather than use something recent as anything after that really didn't have any operational benefit for the 8000 series cards. At least nothing that I noticed would help change anything significant.

pagelm

yes, I have yet to rollback my driver so maybe I'll eventually give that a shot.  As I recall it was the 8800s that introduced the hardware-level control.  At least I don't see an option to go into the 8600's vid card BIOS to change any settings, and I don't exactly need to fry my card by flashing the BIOS - eVGA already told me NOT to do exactly that.
 

Yeah I try to never flash a video card if it's not recommended to do so. There are utilities to do that with the 8600 series cards but I certainly won't recommend doing that either at all. Ever.
 

I'll take a look at my power profiles, but the card goes full tilt through CUDA tasks when I've stepped away from the computer long enough for my monitor to power down (and before that even:P), so it's definitely not permanently powering down the card.  It seems to most frequently lose the fan profile when a CUDA task completes, before the next one begins or when changing a level on TF2.  Maybe when it resets itself from 3D to 2D mode, it goes back to the hardware profile?  That still doesn't explain why, even when it's monitoring the fan speed and displaying it on my screen, the fan doesn't automatically ramp to 100% after a second or two at 30%.  Still, like you said, it's something worth looking into.

With the older versions of RivaTuner and The ATITool you could specify the card to not downclock from the 3D profiles being set but that is not necessary anymore since they don't control the fan speeds, just the GPU clock cycles to keep the games/applications from having a performance stutter due to shifting clock cycles. That would more than likely be exhibited by CUDA interrupts more than the fan having an issue. Since you are using an application which keeps the threads alive for the card and then re-initiates another CUDA thread application this would not be a necessary consideration.
 

Any chance a thread designation or PID change when a CUDA task is killed off or it leaves 3D mode?  Maybe the polling frequency thing is looking at a particular location in memory (or port or IRQ?) that is no longer allocated to anything to do with the card/driver?  In that case, can I poll and set the fan speed using the parent process somehow?  Or do I, for some reason, have to explicitly set Precision to run in an administrative mode (this is on XP, so I don't normally have to authorize any actions).

The PID of the application would be the same for the application and would definitely change if another is initiated. The fan profile would be for the card alone and not the application being used to drive the information through the card driver. The PID of the driver won't change at all since you aren't launching an additional instance of the driver. That kind of thing would be very detrimental to the OS itself since multiple instances of a video driver would also require multiple interrupts for the driver to handle each separate application instances. This is the way drivers for video cards used to work prior to the full GUI interface of Windows. They haven't done this since the introduction of Windows 95. Essentially it was very prevalent in DOS based systems not so much since the graphic drivers were integrated into the OS itself.
 
Along the lines of the PID for the applications they would follow a similar rule like the Linux PID's would follow since they are a POSIX emulation which has been carried over for ease of identification. It shouldn't be necessary to explicitly set it for administrative use as with XP there really is no need to do that as the normal user would have full control of the OS environment implicitly. The only time this is no longer true is if you had migrated to Vista/Windows 7 where the user controls were implemented to keep elevated applications from running without authorization. Yes kind of a failing point with XP but it was a bit more flexible to allow applications elevated control where necessary.
 
From what I understand the polling frequency specifically polls the driver reporting for the card information so that might not be something necessarily needing to be looked into since Precision has consistent graph information you can review which will indicate it's state. I think it's more along the lines of something else causing the software to hibernate or go into a non-polling state while the CUDA tasks are running and you aren't constantly moving the mouse or indicating that a user is present.

All good now...switched to MSI Afterburner which works the way Precision used to on my GTX 275 FTW.