There seems to be a lot of talk about nVidia chipsets being used not only for the MB and MBA but also for the MBP to enable Hybrid SLI and/or Hybrid Power. However, I think it's unlikely that Apple will move to an nVidia chipset for the MBP. Doing so will essentially lock them into using nVidia GPUs since I doubt HybridPower works with ATI GPUs. Intel's GM45 chipset however has an equivalent feature for switching between an integrated and dedicated GPU which would obviously be GPU brand agnostic since Intel doesn't yet make their own dedicated GPUs. This seems like the preferred option since I'm hoping Apple goes with the Mobility Radeon HD4670. The HD4670 has 320 SP which should overpower nVidia's mainstream GPUs regardless of their high shader clock since everything up to a 9700M GT only have 32 SP. More importantly even if ATI's drivers can't yet get the most out of their 5-way SP architecture (5x64=320 SP), presumably OpenCL software will be programmed with this parallelism and architecture in mind to better make use of the 320 SP. As well, the HD4670 supports 64-bit floats while nVidia's 9000 series only supports 32-bit.
One of the major justifications of going with an nVidia integrated GPU for the MB and MBA seems to be OpenCL support to fully take advantage of Snow Leopard. What I'd like to know though is whether Intel's GMA X3100 and GMA X4500 architecturally can't support GPGPU and OpenCL or whether Intel simply doesn't bother writing the drivers and interface for it because they want to focus on Larrabee. Seeing that the GMA X3100 and GMA X4500 also use a unified shader architecture and are DX10 compliant like nVidia and AMD GPUs, I bet it's the latter where Intel simply doesn't see the point in investing in GPGPU software support. Intel's unified shaders actually seem more flexible than nVidia's and ATI's given that Intel seems to use their unified shaders to assist in video decoding with a hardware scheduler between video and graphics mode rather than having dedicated hardware video units as nVidia and ATI have. Apple could probably develop the GPGPU interface for the GMA themselves if they want to, possibly make the hardware scheduler switch between video, graphics, and GPGPU modes and since they are defining OpenCL, they could easily define OpenCL in a flexible enough way to allow GMA support.
I'm also unclear whether nVidia's chipset will allow lower power consumption. nVidia chipsets have traditionally consumed more power than Intel's or ATI's solutions. And the Intel GM45 and PM45 have lower power consumption than their predecessor since they've moved from a 90nm process to a 65nm process so the bar has been set high for nVidia to justify themselves.
And the MCP7A-U that the article points to as a candidate nVidia chipset seems to be a desktop chipset since uATX is a form factor for small desktop motherboards. And mGPU doesn't stand for "mobile GPU" but "motherboard GPU", just another name for IGP. nVidia's upcoming mobile Intel chipset is supposed to be the MCP79 (
http://www.guru3d.com/newsitem.php?id=6859), however it's status is unclear since it was supposed to be released alongside Montevina 2 to compete against it.
I kind of feel sorry for Apple if they've actually had no intention of releasing nVidia IGPs since with all this hype if they just stick with Intel IGPs, people will be very disappointed. Still, they could go with Intel's yet to be released GM47, which is supposed to be a higher clocked GM45, with the IGP at 640MHz instead of 533MHz in the GM45, which Intel claims makes the GM47's IGP twice faster than the previous GMA X3100.
