the next evolutionary step in homogenous processing...
I should have phrased it better: The benefit of having a powerful gpu diminishes extremely rapidly as the gpu performance increases for normal applications. In other words, the onboard HD4200 and Sandy Bridge's integrated IGP is going to be enough, and Bulldozer's 400SP's will not provide a noticeable boost (noticeable as in you can see and feel that its faster, rather than faster only in benchmarks).
As a side note, I don't feel any difference between my Mobility 4570 and my sister's Mobility 5650 in web page performance.
Edit: Currently the main usage of processing power revolves around the x86 architecture, so as long as we are primarily using the x86, I cannot see how we will be able to migrate to GPGPU.
for your mobility reference, it's good that you have hardware that satisfies you for the present, and foreseeable future. no need to upgrade unless you perhaps need better battery life from more efficient processors.
regarding x86, that may change with ARM and Microsoft changing things with Google's and Nvidia's help (let alone the continuing investments from T.I., Qualcomm, Marvel, Samsung, etc), and it certainly will change with languages like OpenCL that mean homogenous acceleration on any hardware.
currently you need to program for either the GPU or the CPU, with our current baby steps to make languages like OpenCL programmable to automatically take advantage of any OpenCL compatible hardware. with the OpenCL the goal is to write code without thinking of the hardware. we're just starting, and we've already got traction and practical benefits today. this will only get better with time.
as for the hardware, the traditional CPU and GPU have been on a collision course for decades, with increasing features being shared such that it's getting more difficult to differentiate them by the day.
things like the GPU becoming programmable with DirectX's programmable shaders, than later unified shaders, then Direct Compute, and OpenGL's similar parallel efforts.
we now have GPU's like Nvidia's Fermi architecture with L1 and L2 caches, ECC and a bunch of other HPC and supercomputing features that rock the previously CPU-only-driven supercomputing world. even Intel with it's Sandy bridge CPU's has evolved from the cross-bar memory bus, to what GPU's have been using for years: a ring-bus.
the defining lines separating CPU's and GPU's are blurring and there will very soon be a single processor that computes everything, there will be no more GPU, there will only be an evolved new generation of processors that may still be called CPU's, but they will no longer be "general", they will be agnostic: neither specific, like a GPU used to be, or general, like the CPU used to be.
it only leaves what architecture these chips will follow: x86, ARM, or some other flavour.
these new generation processors will be no longer "jack of all trades, yet master of none" to evolve into "master of all"
signs of such evolutionary processor design have been seen already.
Sony evolved their thinking when designing their processors with Toshiba and IBM for their Playstation consoles along this way. their playstation 2 console had a PPC with custom V0 and V1 units. these were similar to DSP's that could blaze through vector maths and could be programmed more than traditional DSP's.
games developers used them in all sorts of ways, from improving the graphics functions to add to the GPU's hard-wired features set, to EA creating an accelerated software stack for Dolby 5.1 running solely on the V0 unit, freeing up the rest of the CPU.
with the PS3 Sony took the idea forward again, and evolved the idea of the V0 and V1 units, generalizing even more their functionality, further again from their DSP heritage, and came up with the "The synergistic processing unit (SPU)". sometimes called SPelements, confusingly.
these SPE's would not only be more powerful than their previous V0 and V1 units,, they would increase in number, such that whereas in the PS2 they were the minority of the processing in the main CPU, in the PS3's CELLbe, they would be the majority of the processing potential. these SPE's would amount to 8 units attached to a familiar PPC.
sony prototyped the idea of not having a separate CPU and GPU for the PS3, toying with the idea of two identical CELLbe chips, to be used by developers as they wish. the freedom was there, but he development tools to take advantage of massively parallel processors wasn't, as seen with Sega's twin SH2 processors from Hitachi generations ago.
we have the parallel hardware, we simply need to advance programming languages to take advantage. this is the idea behind OpenCL.
to find a better balance, Sony approached Nvidia late into development of the PS3 and finally decided on the 7800GT with it's fixed function vertex and pixel shader technology to go up against ATI's more modern unified shader architecture in the XBOX360.
it will be interesting to see Sony's plans for the PS4's architecture if they continue their commitment to massively parallel processing.
meanwhile PC architectures like Intel's "Larabee" and AMD's "Fusion" projects, show that the evolution of processing is heading towards homogenous computing, with no specialty chips, and all processing potential efficiently being used due to no idle custom functions.
AMD bought ATI and their Fusion project will eventually merge the GPU's SIMD units together with the CPU's traditional floating point unites to begin the mating of the CPU and GPU into what will eventually be homogenous processor.
just as the smart folks like PC game-engine designers and Sony have
predicted since 2006 and beyond
. So 60s commercial
I thought after that Larabee nonsense they should have came up with GPU supporting OpenCL already. Guess not 