RBE did allow for adjusting the voltages within defined limits on the 4850. The minimum was 0.892 volts maximum was 1.158. This did work for my card I guess they changed some regulators with the 4770 series or support was just not added to RBE.
The 5800 series has a much more configurable voltage regulator and support is apparantly being added to the new version of RBE now so undervolting won't be a problem i'm sure.
Your probably better off with a first generation ATI reference design like my 4850 was and all the current 5850's and 5870's are because these versions generally have better support from 3rd party software. Second generation cards often have changes made that can prevent voltage modding or lack support in software.
Edit - ATI tray tools also allows voltage changes for the 4800 series as well as creating profiles but support for the new series would depend on when the creator can gain access to a card. It is worth checking out his beta forum though he has just released a version with better support for the 4770 and you might want to test it on your current card.
http://cid-a50350ea7a969f0c.skydrive.live.com/self.aspx/Public/attsetup.exe.
Thanks alot!
W1zzard doesn't mention voltages in his reviews of the 5870 and 5850. But working through the numbers, I would assume that the reasons that the 5850 gets substantially lower power draw could be that they shut down 10% of their chip and downclocked it to 725 MHz. This would yield (10/9)*(850/725)=1.303 or 30% higher power draw for the 5870. W1zzards review show 144/108=1.33 or 122/92=1.326 in the two non-synthetic load scenarios.
Why do I care? Because this suggests that the GPU voltage could be untouched between the two, and if so there could be considerable space to undervolt the 5850 chip, and still maintain stability at stock clocks. And it doesn't take all
that much to drop the power draw drastically, as power typically scales with voltage^2 with these kinds of commercial chips. If you could go from, say 1.1 to 0.9 V that would yield (1.1/0.9)^2=1.494. And incidentally, if you again refer to the review, HD5850/HD4770 is 108/69=1.56 and 92/62=1.484
First conclusion is that there exist people nerdy enough to actually
use the data that W1z provides.
Second conclusion is that it may well be possible to for a 5850 to get within spitting distance of a 4770 at load, and at default clocks. And if so, there it would enjoy twice the caculational power and bandwidth, and a far better cooler that exhausts the hot air outside the cabinet. And lower idle power.
Said and done. I already have a buyer for the 4770. Now I just have to order the 5850.
Wish me luck guys.
, for me it's the best buy actually, also at high resolutions as 1920*1080. I'm only disappointed we can't see any test with dx11 game 
..
