Macrumours is great for getting tid-bits of pertinent news, but the short length of articles makes it very clear that most users don't have enough information to draw accurate conclusions.
On naming, Clarksfield is a 45nm quad core mobile Nehalem based chip launching in Q3 2009. It has an integrated dual channel memory controller and an integrated PCIe x16 controller. Arrandale is a 32nm dual core mobile Nehalem based chip launching in Q4 2009. It has an integrated dual channel memory controller, an integrated PCIe x16 controller, and the controversial IGP. Essentially both Clarksfield and Arrandale integrate a CPU + northbridge.
This leads directly to TDP discussions. Current regular Intel Penryn mobile chips are available in 25W, 35W TDPs and 45W for Extreme Editions. This is just for the CPU. The northbridge part of the chipset is usually a bit above 10W TDP, more-so for northbridges with IGPs like the 9400M. Clarksfield will come in 55W Extreme Edition, 45W TDP standard edition while Arrandale will be in 45W and 35W TDP varieties. Since they both are CPU+Northbridge, overall power consumption is unchanged. Apple laptops can use up to 45W Clarksfield or Arrandale and iMacs can use 55W Clarksfields.
Now since Clarksfield and Arrandale integrate the northbridge, notably the PCIe x16 controller for discrete graphics cards, they do not have a high-speed QPI link like Bloomfield Core i7 on desktop. Clarksfield and Arrandale only have a low-speed DMI link to connect to a southbridge for things like SATA, USB, and Ethernet controllers. The limitation that most comes to mind for third-party IGPs like from nVidia is not Arrandale already having an IGP or Nehalem chipset licensing, but the fact that the DMI link does not have the bandwidth to support an IGP integrated into the southbridge trying to access shared system memory which is now on the CPU. An off chip IGP will be bandwidth starved.
Alternatives are for an IGP equipped (nVidia) southbridge to include it's own memory controller and include a small pool of dedicated graphics memory, 64-bit with 64MB or 128MB of DDR2, like AMD's Sideport or previous nVidia TurboCache designs. The problem with this approach is that you are adding cost, duplicating functionality, and using more motherboard area which all go against the point of using an IGP, especially for Apple.
Another way is to link an IGP equipped southbridge through both the DMI links and the PCIe x16 links that would otherwise go to a discrete GPU. This would of course prevent the usage of a discrete GPU which is problematic for the higher-end MacBook Pro and PC desktop chipsets that nVidia would be targetting. nVidia could then implement some type of switch, but that would again add cost, hardware and software complexity, and increase motherboard area.
It seems likely that if nVidia were to try to make an IGP for Arrandale or Clarksfield, it'll either be bandwidth linked through only DMI or will be some sort of less elegant compromise. This means it isn't unreasonable for Apple to seriously consider going back to Intel IGPs.
In terms of Arrandale's IGP, wild speculation on my part would put it's performance on par with the 9400M, perhaps a bit better. While the GMA X3100 was made on a 90nm process, the GMA X4500 was made on a 65nm process. Intel could have used the shrink to double stream processor count, but instead it only increased from 8 to 10 meaning they were already conservative by not using all the potential available to them. Arrandale's IGP will be further shrunk to 45nm, so Intel certainly has the ability to give it at least 16 stream processors or up to 32 stream processors without significantly increasing die area if they are really serious about performance. Thermals should controllable since the 45nm process has been remarkably successful and mature compared to the problematic 90nm process. If Intel were to promise 9400M level performance and OpenCL support in their IGP, I can see Apple deciding to cut their losses over nVidia's IGP uncertainty and just use what is already in Arrandale anyways. Apple could of course try to smooth any hurt nVidia feelings by continuing to use nVidia discrete GPUs, perhaps expanding it into the high-end 13.3" MacBook Pro if possible.
