Limited GPU and Memory clock speeds in nvidia-settings

I have a Linux only system with a ASUS GeForce GTX 670 with 4G ram which according to the manufacturer should have a 915Mhz base GPU clock speed and a 6008Mhz memory speed but the nvidia-settings (driver 304.14 or 310.32) only shows GPU 705Mhz max and Memory 3004Mhz max. I noticed during bench marking that the GPU never goes above 67C. So I think it really is running slower. But I can’t compare it to windows to see if it is just the Linux driver. Is NVIDIA acknowledging this issue? Or do they say they cannot replicate the problem? I’ve seen it reported several times for various cards.

I tried the following but get an error:
chris@linux-studio:~$ nvidia-settings -a GPU3DClockFreqs=915,6008

The valid values for 'GPU3DClockFreqs' are in the ranges 705 - 1411, 751 -
3604 (inclusive).
'GPU3DClockFreqs' can use the following target types: X Screen, GPU.

chris@linux-studio:~$ nvidia-settings -a GPU3DClockFreqs=915,3004

ERROR: Error assigning value 59968444 to attribute ‘GPU3DClockFreqs’
(linux-studio:0.0) as specified in assignment ‘GPU3DClockFreqs=915,3004’
(Unknown Error).

Can this only be done before the X server starts? Or does the X server need to be recompiled to allow setting the clock frequencies?

Saw this earlier today while looking at nvidia-settings on my GTX Titan:

I believe the above still applies… I’d go with what the deviceQuery CUDA SDK spits out. I’ve done some CUDA benchmarks and the card doesn’t seem to be underclocked as far as I can tell.