There is a prevailing assumption that this problem is "caused" by bad software. This might be the case, but bad hardware could also be part of this! A lot of

There is a prevailing assumption that this problem is "caused" by bad software. This might be the case, but bad hardware could also be part of this! A lot of modern hardware has programmable parameters, "core voltages", PLL parameters that control running frequencies and the timings for DMA / memory access state machines. Who's fault is it if the hardware vendor says use X, Y and Z for these parameters. And they include the same parameters in the software they supply and.... What if some %'age of the hardware doesn't actually produce the timings the configuration parameters specify, or does "really bad things"? In reading the "source" article, there is a pointer to a discussions.apple.com thread. The "fix", if you go and read the thread, is on page 21 from Andrew Humphreys. What his solution does is change the timings and clock speeds for low power and idle modes of the GPU. Reading between the lines, the low power / clock speed settings don't work and will cause corruption. His changes are derrived from people having problems running Windows 7. The same changes seem to help people running OS X. Perhaps someone from the Linux community will be willing to force the ATI GPU cards in question into low power / performance modes and see if corruption occurs. I've no knowledge of what the parameters Andrew is overriding actually do, but for a portable platform "faster" almost always results in higher power usage, heat and therefore shorter battery life. This might not be such an issue for desktops, if you can tolerate the noise of increased cooling. Getting pragmatic about this, how could you design a Q/A test for this? Do you really afford to wait X number of minutes before / after each test to make sure the GPU is cold enough to go into low power mode? What if you decide to use hardware that can cool the GPU faster so you could get through all the tests -- doesn't that make the tests invalid as you are no longer using HW customers will get? As a final note, there might be evidence that some graphics drivers were updated ~1 week before 10.8.3 was shipped. The evidence is in /System/Library/Extensions/