A ramped up north bridge dose a lot for the Phenom II, but not nearly as much for the Athlon II. This was of course concluded through a barrage of benchmarks (purely numbers), but I wouldn't be surprised if the increase made for a snappier system.
I've never seen numbers on a Northbridge running at 290 myself so if you do happen to get some contrasting info from your buddy (stock vs 290) be sure to let me know, I'd be interested in seeing those.
I'm honestly interested in seeing his Rana numbers (it's Formulav8 btw). What was surprising to me is that increased NB speed had an . . . unexpected effect on system memory read, write, and latency. I was only expecting latency.
For what it's worth, at 3.7 ghz with 2850 mhz NB and DDR3-1520 6-7-5-15 1T:
Everest Read: 10597 Write: 5325 Latency: 37.0 ns
If you drop the NB to 1995 mhz:
Everest Read: 9093 Write: 4617 Latency: 41.2 ns
Ouch! That's quite a hit, especially on the read speeds. Take the synthetic for what it's worth, but still . . .
Now if only I could get DDR3-1800 CAS 6 stable on this chip (yeah, right), then we would have some good memory numbers. You can do that on a C3 Phenom II, but on a Propus? No way. Not the way they are right now. Maybe the 640/645 will fix that, but by then, everyone will be moving on to Thuban chips, and for good reason.
edit: Actually, since he's probably putting together charts and things, I'll just dump my raw testing data here for your perusal. And yes, it's not a massive increase in performance, but there was more there than I really expected, especially considering how low the memory speed is:
3705.3 mhz, 285 mhz HTT, 2850 HT Link
DDR3-1520 6-7-5-15 1T
NB: 2850.2 mhz
Everest Read: 10597 Write: 5325 Latency: 37.0 ns
superPi mod 1.5 XS 1M: 20.951s B43F7FA9 32M: 18m 25.449s 7BF80A1
Cinebench R10 1 CPU: 4014 x CPU: 14259 Multiprocessor Speedup: 3.55x
Fritz Chess Relative Speed: 18.70 Kilo nodes per second: 8976
Aquamark3 (64-bit patch) CPU: 16,661
ScienceMark 2.0 Overall: 2416.89
Cinebench R11.5 CPU: 4.24
7-zip 32MB Dictionary size, Passes 10 Total CPU Usage: 369% Total Rating/Usage 3658 MIPS Total Rating: 13521 MIPS
TrackMania Nations Forever 228 fps (1024x768, all settings none/lowest/fastest except anisotropic 2x, 8800GTX)
Blender: 8.57
NB: 2565.1 mhz
Everest Read: 10088 Write: 5029 Latency: 38.6 ns
SuperPi mod 1.5 XS 1M: 21.044s C823DABA 32M: 18m 34.232s 89C45805
Cinebench R10 1 CPU: 3948 x CPU: 14198 Multiprocessor Speedup: 3.60x
Fritz Chess Relative Speed: 18.46 Kilo nodes per second: 8859
Aquamark3 (64-bit patch) CPU: 16,337
Sciencemark 2.0 Overall: 2380.94
Cinebench R11.5 CPU: 4.18
7-zip 32MB Dictionary Size, Passes 10 Total CPU Usage: 368% Total Rating/Usage: 3583 MIPS Rating: 13218 MIPS
TrackMania Nations Forever 225 fps (1024x768, all settings none/lowest/fastest except anisotropic 2x, 8800GTX)
Blender: 8.64
NB: 2280.1 mhz
Everest Read: 9797 Write: 4791 Latency 39.1 ns
SuperPi mod 1. XS 1M: 21.216s 552086B3 32M: 18m 37.882s 62D890E1
Cinebench R10 1 CPU: 3959 x CPU: 14072 Multiprocessor Speedup: 3.55x
Fritz Chess Relative Speed: 18.40 Kilo nodes per second: 8833
Aquamark3 (64-bit patch) CPU: 16,065
Sciencemark 2.0 Overall: 2350.44
Cinebench R11.5 CPU: 4.16
7-zip 32MB Dictionary Size, Passes 10 Total CPU Usage: 371% Total Rating/Usage: 3576 MIPS Total Rating: 13296 MIPS
TrackMania Nations Forever 223 fps (1024x768, all settings none/lowest/fastest except anisotropic 2x, 8800GTX)
Blender: 8.66
NB: 1995.1 mhz
Everest Read: 9093 Write: 4617 Latency: 41.2 ns
SuperPi mod 1.5 XS 1M: 21.388s FEE1B54B 32M: 18m 52.921s F5A8B5E1
Cinebench R10 1 CPU: 3973 x CPU: 13993 Multiprocessor Speedup: 3.52x
Fritz Chess Relative Speed: 18:30 Kilo nodes per second: 8785
Aquamark3 (64-bit patch) CPU: 15,688
Sciencemark 2.0 Overall: 2320.70
Cinebench R11.5 CPU: 4.12
7-zip 32MB Dictionary Size, Passes 10 Total CPU Usage: 373% Total Rating/Usage: 3502 MIPS Total Rating: 13123 MIPS
TrackMania Nations Forever 217 fps (1024x768, all settings none/lowest/fastest except anisotropic 2x, 8800GTX)
Blender: 8.68
This is pretty much how it goes when you're trying to squeeze performance out of a system by increasing system memory speeds. The nice thing about it is that you get performance across the board, even when you're dealing with applications that have a large memory footprint and large working set. What made very little sense to me is that NB speed bumps would boost memory read or write speeds at all; there is certain to be a good reason for this, but I'd rather not make myself look stupid guessing wildly at what that might be.
Also, if you will notice, my single CPU scores in Cinebench R10 are a tad . . . anomalous? I shrugged those off since the x CPU numbers seem to make more sense.
In applications that can actually use four cores effectively (video encoding), the Athlon X4 is nearly 35% faster.
Ding ding ding, we have a winner. Sadly, there are a lot of multi-threaded apps out there that just don't hit 4 cores very hard. Yet. Obviously there are some, such as encoding, which is one of the few CPU-intensive things I actually do.