Eracer:
You can't have it both ways. Either Nehalem is a quad core with an IGPU, which uses a lot of power and can do much for HPC type work or its a dual core which will be not really all that useful in places outside of games and such. Trouble is that for certain areas like de/encoding and rendering, a GPU core is far batter at it than any CPU core. So Fusion with a dual K10.5 core with a R700 class core (HD4200 size), which alone will do the encoding and rendering far faster than all four of Nehalem's high end desktop cores.
If that is defined as low end mobile, then Intel is in a heap of trouble. Now, my reasoning for Fusion to use a R800 class core (HD5200 size) is that they will have to relayout a GPU for 45nm SOI anyway. Since R800 is coming out at roughly the same time, it would make sense to do it then, so that 45nm SOI is available for really high end/low power discrete GPUs and bulk for volume GPUs. This is shown by TSMC's wanting 45nm SOI capability to go against Chartered/IBM for the business. With that perhaps bulk won't be necessary at all, which vastly increases the likelyhood of R8xx for Fusion. It gives a Puma style Fusion on one or two chips rather than three as it is now. I think it will be two, one CPU/GPU/NB package on a new socket and one SB/Video/Audio/LAN/etc in/output package.
Now that is low end mobile only because it isn't a discrete GPU with nearer to the full 800+ stream processors and 32+/16+ texture units and ROPs. 10% of a HD4850 is still far more performance than any Intel GPU, integrated or otherwise. 10% of a HD5850 will be at least twice that. Intel would need to use at least a HD4350 just to even the playing field against a HD4200 IGPU and that will cut the price of Intel's competing solutions by $20-30. Or give AMD or nVidia $20 for every mobile Intel's customers sell. Given the state of the current GPUs, that likely means that AMD would pick up the bulk of that business.
In essence, AMD will raise the bar of what is considered low end (value) mobile. That would be about what mainstream laptops need for Christmas this year and gaming (higher end) laptops do right now. Top end mobiles now would be satisfied with Fusion and a HD5350/70 discrete in Hybrid CrossfireX, a mainstream mobile by Q2/09.
Windows 7 likely will need good 3D performance to use the GUI. That is why a HD3200 would be a bare minimum of 3D in any system. And HD4200 performance GPU would be the smallest recommended size. Thus that becomes the specs for low end mobile since business looks 3 years out when specifying. And since Intel is behind by a year or two in IGPU performance, they are in a heap of trouble. I don't think Microsoft would allow Intel to do what they did for the 915 chipsets wrt Vista for Windows 7. Too much criticism from that episode.
As for DirectX 11, it will be a large marketing point. The same as 10.0 was. When you buy a mobile, you are stuck with it for 3 years. So its a large point in mobile and with most desktop users not upgrading much during its life (memory, HD space and optical mostly), they also look that far ahead. 3 years from now is Q3/11 and less than 30 months to 2011. By Q2/09, its within 18-21 months, well within the normal range. In fact I wouldn't put it past MS to make a popular series DX11 only game at the release of Windows 7. It would boost the early sales.
And if you think that Fusion will be too slow to play them (I don't agree), no way Nehalem IGP will and its buyers will say, "I'll never buy Intel again!"
Pete |
