Re: Incorrect video memory reading on nvidia video cards?
All GPUs from the GeForce 8 series and up support TurboCache. For a discrete card, TC memory is slower than dedicated video memory, but faster than memory that the GPU can't access at all, which is what you got on non-TurboCache chips when you ran out of vidmem. For mGPUs, I'd expect TC memory and the dedicated memory to be roughly the same speed. However, the dedicated "carveout" memory is reserved for the GPU at boot time, while TC memory is dynamically allocated and shared with the OS. The OS cannot reclaim it while it's in use, but if you stop using it as TC memory then sure, the OS can reclaim it for other uses.
I don't think there are any known stability problems with TurboCache right now.