IMO, best not "believe" any statement that starts with "if rumors are to be believed".
It doesn't look to me like he has any insider information on Haswell. Anandtech as a team of reporters typically teases early benchmarks (benchmark-happy!?) as soon as they get their hands on something, so... I'd take their statements with a healthy grain of salt. Only their graphs matter.
you know the worst part? he is mostly clueless
the 7730m is to compete with the 640m, the 7750m 650m.
Aside that he doesnt have anything to base that statement, aside the intel video, which doesnt have fps figures.
not to mention the performance of the HD4000 is around P700 in 3dmark11, to compete with the 650m it should need at least P2300 for the 5200, that is just 4x times faster, if we go by the math here of:
HD 4000 = 700
HD4600 = HD4000 * 20% = 840
HD 5100 = HD 4600 * 50% = 1260
For the HD 5200 to be P2300, it just need another 85% of performance boost based on 128mb of very fast and wide cache, and while I still think its possible to get to around the 640m, we are still facing P1800 in 3dmark 11, and for that we are so going to need a ton of that edram
Thus the HD 5200 for me is around P1600 = HD 5100 * 33%, and that is still a whole lot for just 128mb of a very wide and fast edram
I hope that gives some ideas of what we are facing on haswell vs ivy a healthy improvement on gpus, if its good enough its up to you to judge and me to take toms hardware word on the power of the 5100 and their estimates at 50% against the 4600 seriously
Just to make a comparison, the 680m gives around P6500-6800
