In terms of Intel, their business plan is to lower power and increase acceleration of very common features, hence gpu intergration into the die. They are still persueing high end but, the problem is you are thinking short term. There needs to be an IGP in EVERY Cpu sold in the world, every last one, including high end, for openCL and other standards to become used as standard in EVERY piece of software that can use it, and I mean every single last one.
Right now a few pieces of software use it, not nearly enough, quicksync is limited because companies that can utilise it don't have to, because only 1/50th of the people that use their software(if that) have access to it. its not cost effective, yet, to support it in every single piece of video encoding software. In 2 years when 1/2 their users have an IGP, they will put a lot more effort into it, in 3-4 years when 70-80% of their customers all video software will try to use it as best as possible and you'll find that a quicksync enabled quad core will spank a 16 core no IGP core in that software. But its not just video software, the wider the base of users the more software will be writen to utilise it where possible. Every generation the GPU on die will increase both in efficiency and in features to make it more suitable for use in accelerating more types of software.
CPU's alone are extremely fast at SOME types of data processing and extremely slow at others, an APU is extremely fast at BOTH, it will simply take time for the GPU part to become useful.
If Intel don't push it, and AMD, then we'll never get to the point where software makers use it, that is how the computer industry has always been.
They make more powerful cpu's with more cores when no one used them, and as they became standard more and more software used them, its no different, gpu's get more powerful THEN software starts to require that much power.
In 5 years there will be a x core CPU with X amount of gpu shaders that will be much more versatile and MUCH faster than a CPU with simply 2x cores. The transition happens before the software, that is the way its always been.
Also, the more ridiculous part is, there are already hexcore cpu's, and there WILL be octo core Haswells, just not desktop ones afaik.
It's actually likely they will introduce 6 or 8 core gpu having or not having Haswell's.
The problem is desktop now has dual/quad cores and dual channel mem, triple/quad channel mem increases price drastically as would more cores. They need a cheap "desktop" platform for the masses and making it quad channel and 6-8 cores isn't a viable option at all. There are times when products split up nicely and times they don't.
first i7's had not THAT expensive i7 triple channel memory and faster cpu's, then they moved them to more efficient cheaper midrange platform back on two channel. The quad channel hexcore was too expensive for an upper midrange part, the next platform could have.
Haswell could be like the first i7/i5 platform, dual channel quad core, and a potentially unannounced 6-8 core triple/quad channel platform that isn't as expensive as the current Sandy-E platform.
They've talked about one platform, that doesn't mean it will be the only non high end one so I wouldn't take anything for granted yet.
