Karl,
I am really interested in your experiment too. I have the same config as you -- stock 2x4 2.8GHz w/ATI and then added 4 x 2GB OWC for 10GB total. Also added a 2nd 320GB HD.
I know from another thread you also have seen the GPU freeze and dump. That is my main problem -- fortunately only happens once every 1-3 days so far.
But I have also had two non-GPU kernel panics ('Preview' and 'update') since the machine came online Jan-18.
(I do not sleep the machine so do not know if that problem lurks also. Sorry -- not hijacking the thread -- there is a chance the sleep and GPU problems tie together.)
I have been treating the GPU freeze and kernel panics separately because of their dissimilar circumstances, but I think now that has been foolish.
While trying various fixes for the GPU problem as posted [here|http://web.mac.com/sailhome/SailhomeMacZ/MacPro.html|Freeze Saga], at one point I did a clean install.
Upon first login to this new install, System Profiler only recognized 8 GB of 10 GB RAM installed. I tried several reboots but the condition persisted. Only the OWC 4 x 2GB were recognized
A1/A2 -- OWC 2GB/2GB'
B1/B2 -- OWC 2GB/2GB'
A3/A4 -- Apple 1GB/1GB' not recognized
I booted into AHT, which froze during the device probe. I tried this twice and each time it was a total freeze including cursor -- I had to press power button to shutdown, then again to boot.
I shutdown, disconnected cables and opened up the case to access RAM. Of course, by default this effected another SMC reset (had done several SMC resets earlier).
I removed the OWC RAM and put the original Apple RAM back in
exactly as from the factory -- original sticks in original slots.
A1 -- Apple 1GB
B1 -- Apple 1GB'
Booted into AHT and ran 'quick' mode -- all hardware passed.
Rebooted and checked System Profiler -- 2 GB RAM as expected.
A1 -- Apple 1GB
B1 -- Apple 1GB'
Shutdown, disconnected cables and opened up case again (yet another SMC reset). Installed 4 x 2GB OWC RAM + 2 x 1GB Apple RAM as before, but this time I swapped slots within the pair. In other words the stick originally in A1 now in A2, stick in A2 now in A1, etc.
A1/A2 -- OWC 2GB'/2GB
B1/B2 -- OWC 2GB'/2GB
A3/A4 -- Apple 1GB'/1GB
Booted and checked System Profiler -- 10 GB RAM as expected
A1/A2 -- OWC 2GB'/2GB
B1/B2 -- OWC 2GB'/2GB
A3/A4 -- Apple 1GB'/1GB
Looking back, I should have been more thorough and isolated RAM pairs and positions like you are doing. In reviewing my records and photos, I can see that the "flavor" of GPU corruption changed. Before, the episodes were instant and complete screen corruption. After, the episodes were partial corruption and sometimes cleared without a shutdown.
But that's with hindsight. Back on the date when I swapped RAM, the next thing I did was run Rember (memtest with GUI) overnight. The modes I used were All, Quit Finder, and Verbose. It exercised 9.5GB RAM and made it through 4 loops. All passed, so I figured at the time that was sufficient.
Again, looking back I should have tested pairs, swapped slots, and tested again. And probably use memtest booted in single-user mode instead of Rember from full boot. Clues were there...dimwitted...
I have wondered about how system RAM, video RAM, and power may play into the various problems users are having -- including this thread's boot-on-wake problem. The users who avoid the GPU freeze by running dual monitors (or a single monitor with dual inputs) are splitting the ATI card's onboard RAM -- and maybe that also affects addressing with system RAM?
And perhaps some number of users are not seeing problems because they add 3rd party RAM right away and leave the Apple stock RAM out? If users are seeing problems with stock configs, maybe it's a RAM problem and not an ATI card problem (I know, could be either or both...) That would help explain why some people still have problems even with new video cards or even new systems. Could also help explain why some 8800 GT systems are also beginning to exhibit unwanted behavior with sleep and GPU.
I suppose the Apple stock 800 MHz sticks were manufactured earlier than the 3rd party sticks. And the earliest 2008 Mac Pro systems probably received the earliest 800 MHz sticks.
So, my next step is to remove the Apple stock RAM and see how many days I go without a problem. I'm choosing that config since originally it was the Apple RAM that didn't show up after clean install.
If 4 x 2GB OWC is successful (let's say up to 3 weeks with no incidents) then I'll try 2 x 1GB Apple and compare behavior.
If 4 x 2GB OWC is not successful then I'll take half out and start testing the full matrix of paired positions. That could take awhile at my current rate of failures.
I'll probably reserve running memtest (or looping AHT) on configs only after any freezes/panics because I don't want to interrupt the uptime. Seems like my machine has to run past a certain period of time before an incident occurs (again, may be coincidence).
Which, by the way, might be a reason why at least some users with wake-on-reboot aren't necessarily seeing GPU freezes. If they are rebooting daily (or more often) then they may not get far enough into whatever it is that reveals the GPU problem.
Yes, I know that could be bogus. I'm not an expert. Just trying to make observations that might help come up with solutions.
Karl, thanks for your effort -sailhome