Interesting thread. It indeed answers a nagging question i've had for a while.

My company builds systems based on the ultra-d something we've been doing since about the 2nd month of 2005. They're primarly used for multi-display systems (DXG is a cheap way to get high power, rather than very expensive quad display workstation video boards) Anyhow, moral of the story is, i get ALOT of boads/cpus/components coming across my workbench.

About 3 weeks ago I was assembling a fresh batch of ultra-d's. As a stability test, i usually push these systems up to around 225HTT or so and prime them overnight. I chose that # because thats about where most run of the mill a64's start requiring voltage. My general rule of thumb is, if they can do 225 and hold prime overnight i'm usually good. Well this latest batch curiously went up to 235 before needing volts. not so much different, but when you've built as many of these as I have you notice when something changes. The improvment was repeatable across the 5 systems i assembled that day. I didn't even think to look to see if there was a new revision. I chalked it up to a good batch of 3700's.

Nevertheless I took one home and swapped it out for my ultra-d (AA0). Copied out the settings from the bios line for line and entered them into cmos reloaded on the new one. After a few days of adjusting and tweaking to this board, there is a vast difference in the necessary voltage versus the older revisions.
This is what I ended up with for a daily runner.
http://home.comcast.net/~butterfry/o148.3000.png
Unfortunatly i don't have a good screenshot from the AA0 that matches. If enough people are interested i can throw it back in when i get time.

what i've noticed
1. lower cpu volts (1.55 AD0 opteron 148 CAB2E @ 300x10 vs. 1.58-160 AA0 depending on the mood of the thing)
2. fewer "MOODS" that many of you may have noticed. It seems less prone to random swings in performance.
3. BH-5 Memory @ 250 AD0 = 3.30v. AA0 = 3.40 (YMMV of course)
4. Higher memory stability. not by much but with the same settings i'd say 2-3HTT

again i'm glad someone noticed the revision, as i'm too dense to actually observe things like this. I could have reported on this weeks ago but i thought it was a fluke, or i had gotten a golden batch of boards.

-butter