Originally Posted by netllama
I don't believe that I've written anyone off. There's no doubt that you're experiencing a problem, however until it can be reproduced internally at NVIDIA, I'm afraid that it cannot be debugged.
That is a rather strange position. Having developed (and debugged) operating system software for 15 years and worked on network design and testing for 6 years, I can tell you with some certainty that the opportunity to reproduce a (non-application level) problem is rare.
We have platforms that can reproduce the problem reliably; all Nvidia needs do is collect the data. The code where the clock is determined has obviously changed. Put some debug statements in to dump the registers and memory locations you need at the appropriate locations in a 7676 and 8178 build of the driver, and see what's going wrong. It may take two or three iterations to narrow the problem, if done carefully. The corrected code can be tested by those impacted as confirmation.
There's no need for developers to personally experience the error. Odds are, your Linux user base has more skilled developers than work for Nvidia. Not utilizing that resource, when it's available at no cost to the company, is... strange.