CPU energy usage increases after suspend/resume

Rebellious · Post by **Rebellious** » Tue Mar 10, 2009 10:51 am

I plugged in the "kill-a-watt" P3 power display/monitor, here's what I get with 2 systems:

Athlon 64 X2 4600+ with onboard NVIDIA GeForce 6150
Idles at 52w on fresh boot
Idles at 64w after suspend/resume (running Windows 2K or XP)

Athlon 64 X2 6000+ NVIDIA GeForce 8600GT
Idles at 75w on fresh boot (CPU temp 24)
Idles at 90w after suspend/resume (CPU temp 31)
But it doesn't happen with Ubuntu, Linux still idles at 75w after suspend/resume

In both cases Windows has a 20% increase in watts and idle CPU temp.

I checked because somebody spotted a problem with CPU frequency in one of my recent posts here viewtopic.php?t=52626

So installed KB835730 hot-fix that forces the system to use pmtimer (HPC) in Windows XP or 2k instead of the Time Stamp Counter (TSC). http://support.microsoft.com/kb/835730

That fixes the timing problems with some apps but both systems still run hotter after suspend/resume. Any ideas why one S3 cycle increases energy usage?

lowpowercomputing · Post by **lowpowercomputing** » Tue Mar 10, 2009 12:18 pm

Maybe Cool-n-Quiet is disabled after resuming in Windows. Can you check (using CPU-Z or something along these lines) if the CPU's frequency and voltage still drops after the system has resumed from standby? That's the only thing I can think of right now.

Rebellious · Post by **Rebellious** » Tue Mar 10, 2009 4:56 pm

Cool'n'Quiet is working, frequency looks ok after resume, (if it wasn't, the 6000+ system would be using 150W). Must be something about the CPU power state, I bet all machines do this under Windows NT 5.xx, it's just that people don't notice, I mean who's gonna plug a power meter into the wall and check?

smilingcrow · Post by **smilingcrow** » Wed Mar 11, 2009 6:39 am

Rebellious wrote:Cool'n'Quiet is working, frequency looks ok after resume, (if it wasn't, the 6000+ system would be using 150W). Must be something about the CPU power state, I bet all machines do this under Windows NT 5.xx, it's just that people don't notice, I mean who's gonna plug a power meter into the wall and check?

Me and I've not come across this problem with Win2K or XP but this was with Intel systems.

Rebellious · Post by **Rebellious** » Wed Mar 11, 2009 8:59 am

smilingcrow wrote:
Rebellious wrote:Cool'n'Quiet is working, frequency looks ok after resume, (if it wasn't, the 6000+ system would be using 150W). Must be something about the CPU power state, I bet all machines do this under Windows NT 5.xx, it's just that people don't notice, I mean who's gonna plug a power meter into the wall and check?
Me and I've not come across this problem with Win2K or XP but this was with Intel systems.

haha you have a Watt-o-meter too huh?

It could be just an AMD problem on the CPU timer, it seems the 2 cores go out of synch when one is halted or put in power-saving mode (I found plenty of reading on that issue). It could also be corporate politics, MS used new drivers for Vista and stopped NT 5.xx development except for bug fixes and security, and AMD may have lost out to Intel on that deal, only insiders could tell us. Regardless, as of Vista I consider Windows to be spyware and will not use their OS anymore.

.

lowpowercomputing · Post by **lowpowercomputing** » Thu Mar 12, 2009 11:42 pm

As one of my boxes has an Athlon LE and runs Win2K I'm going to check if I also have this issue (have never used standby on that box before). It seems it's more likely to happen with multi-core systems, isn't it?

Rebellious · Post by **Rebellious** » Sat Mar 14, 2009 6:36 am

lowpowercomputing wrote:As one of my boxes has an Athlon LE and runs Win2K I'm going to check if I also have this issue (have never used standby on that box before). It seems it's more likely to happen with multi-core systems, isn't it?

Yes, from what I read it only affects multi-core systems. Individual cores go out-of-synch when suspended (on some systems but not all).

I think Microsoft and AMD couldn't agree on further development of CPU drivers for the NT 5.xx platforms (2k, xp, 2k3, etc., all non-6.xx I suppose) so they just dropped it. Initially NT 5.xx OSes could not officially detect logical vs. physical processors, though they support multi-threading (but not as efficiently as Linux).