260GTX nvidia-drivers-190.42-r3 - powermizer issue

[gio___]

Even here one time a sudden crash happened (maybe twice) with battery power but I had given it no importance. If I'll get more trouble I will post my dmesg.

I can say that even if ACPI notice the transition from battery power to AC, it is not the same for nvidia-settings that continue to recognize the power present at boot but may be is a problem not directly related to nvidia-drivers... I will investigate

gio

[queen6]

Hi Stephen,

Exactly the same story with 180.60. I downgraded to xorg-server 1.6.5 and kernel 2.6.30.9

By default card stays on lowest clocks settings (despite reporting maximum performance) - (powermizer2.png)

Running "nvidia-settings -q all" clocks jump to level 1 (powermizer3.png).

If there's anything else I can do to help - please let me know.

[Stephen Warren]

queen6, thanks for the information re: PowerMizer. I'll add it to our bug reports.

For the X crash, it's probably best to start a new thread. it might be useful if you could:
a) Switch to text console
b) As root, run "gdb X"
c) At the gdb prompt, type "r" to run
d) When X crashes, type "bt"
e) Paste the information to the new thread, and re-attach your bug report there too.

Thanks.

[queen6]

Hi Stephen,

When I get home, I'll recompile X with -ggdb and post a backtrace.

Thanks,
Rob

[gio___]

Not fixed in 195.22 beta.

gio

[iSpy]

GT 240 DDR5, Powermizer show "Maximum perfomance", but clocks not change.
Tested on 190.42 and 195.22
My 3d frequencies on this card 575/1800, but work always 405/324.
Please fix it.

[pootle]

yup, on my laptop with this version on battery it seems to drop to perf level 0, wind up correctly (for example glxgears bumps it up to level 3 correctly.

on mains power, once it has wound up to level 3, it just stays there (and my knees get toasted )

Oops, wrong, it does wind down, but after running Cuda apps, it seems to take a little longer...

[gio___]

Not fixed in 190.53

[mrn]

Signing up to this.

[ttsiodras]

Same thing happens to me - my card is a GT240/GDDR5 - I realized it after I found out that my GPL CUDA Mandelbrot zoomer run at double speed under Windows (I blogged about it here: http://users.softlab.ece.ntua.gr/~ttsiod/mandelSSE.html )

bash$ nvidia-settings -q all | grep -A 1 PerfModes
  Attribute 'GPUPerfModes' (home:0.0): perf=0, nvclock=135, memclock=135 ;
  perf=1, nvclock=405, memclock=324 ; perf=2, nvclock=550, memclock=1700 

bash$ nvidia-settings -q all | grep  GPUCurrent | grep Attr
  Attribute 'GPUCurrentClockFreqs' (home:0.0): 405,324.
  Attribute 'GPUCurrentPerfMode' (home:0.0): 1.
  Attribute 'GPUCurrentPerfLevel' (home:0.0): 1.

The level doesn't change no matter what I tried (including setting the registry flags in xorg.conf, to disable powermizer, or hardcode levels...

Option "RegistryDwords" "PowerMizerEnable=0x0; PerfLevelSrc=0x2222; PowerMizerDefault=0x1; PowerMizerDefaultAC=0x1"

The bug exists in both 190.53 and 195.22.

Please fix this in the next driver release...

Thanassis Tsiodras, Dr.-Ing.

[iSpy]

Not fixed in 195.30

[ttsiodras]

Here's a screenshot showing the problem:



Notice that my GT240/GDDR5 runs glxgears at around 4700 fps (which is very slow for this kind of chip), and that even though I run an OpenGL app, powermizer still doesn't use the 1.7GHz mode...

The same applies to CUDA code, like nbody:

Under Linux:
./nbody -benchmark -n=30000
Run "nbody -benchmark [-n=<numBodies>]" to measure perfomance.
30000 bodies, total time for 100 iterations: 19701.262 ms
= 4.568 billion interactions per second
= 91.365 GFLOP/s at 20 flops per interaction

Under Windows XP:
Run "nbody -benchmark [-n=<numBodies>]" to measure perfomance.
30000 bodies, total time for 100 iterations: 12137.919 ms
= 7.415 billion interactions per second
= 148.296 GFLOP/s at 20 flops per interaction

I verified (via GPU-Z) that under Windows XP, the card uses the 1.7GHz mode.

This is definitely a Linux driver bug, that affects cards with GT240/GDDR5 - maybe others, too...