BaronMatrix :
There are two things noticeable right away.
Some good points... and some bad points...
First, either they didn't try or AMD still hasn't fixed the RAM divider. 50MHz to 800DDR2 isn't earth-shattering but it would make up the 2-3% deficit.
Good point, but I'm afraid that RAM divider will continue to haunt AMD in the future. Since the die area is limited, AMD simply didn't have enough space to implement complex RAM logic in the memory controller. In addition to that, AMD would need to go back to the drawing board to come up with a new IMC. In Intel's approach though, since the memory controller is on the chipset, Intel can implement complex logic to accommodate more RAM speed.
I agree that RAM speed will indeed hinder Phenom's performance, but I'm not sure if AMD will do something to rectify that.
Second, HT3 is NOT 1600MHz, it's at least 2.4GHz. And the Agena is said to run at 2.6GHz. That's probably another 5-7%.
Bad point... bad bad point. Contrary to public, and especially your belief, HT3 will not cause a significant performance difference in desktop applications. Phenom is not data hungry, and having a wider (not faster) bus will not help at all; maybe on server, but definitely not on desktop.
Admittedly Penryn could get to 4GHz under 200W so we'll hope that the seeming ease they got Agena to 3GHz will mean the retail Agena - the last quad due - will get the stepping that has been said to hit 3.2 for quad.
Another bad point. So far we've not seen a 3.0Ghz Phenom demonstrated by a 3rd party (ie. Not AMD), and the faster Phenom tested to date is 2.5Ghz. According to VR-Zone, Phenom 2.5Ghz will likely to have a TDP of 125W, and a lot more by 2.7Ghz. Therefore, I'm not sure if 3.0Ghz Phenom will have lower TDP than Penryn at 4.0Ghz.
Thing could change though, as AMD will bin significantly lower leakage part for Phenom 3.0Ghz.
That was a quick and dirty test though so we don't even know if they have a retail mobo sample. Either way, it's pretty close. CrySis is TOTALLY GPU dependent though so that's a bad test to use.
I'm not entirely sure if Crysis is totally GPU bound. Its developer highly recommend the use of quad core over any other hardware component.
http://www.shacknews.com/featuredarticle.x?id=639
According to benches even a GTX can barely get above 50fps at even 1280x1024 WITH NO AA. Maybe they should use a DX9 game which will be CPU dependent at 1280.
Again, even though the demo seemed to have minimal effect on multicores, it appears that game is still slightly CPU IPC bound.
Although I agree it is still too early to hand down a verdict in terms of Phenom's performance, bear in mind that Phenom is slated to launch next month. Therefore, we probably won't see significant performance difference between the ES and the finalized version.