Well then you're in luck, their is a free alternative. Direct x11 has it built in, compute shaders. Now you can support that.
Compute shaders, first of all it's not free or open and most importantly it is NOT a physics engine or middleware. It's just like CUDA or OpenCL, it's just a tool that let's you code for the GPUs in addition to the CPU. You still have to write the code and most game developers are just not willing to write their own physics engine. That's why most games use Havok, PhysX, Bullet and similar middleware physics engines.
Compute Shaders is not different to x86 in which you just have a bunch instructions you can use to make your program, but you have to code that program. Compute shaders does nothing on it's own.
Whenever I see a physics engine which uses CS or OpenCL that is not vaporware and that is as good as PhysX/Havok I will support it. I'm not going to ditch something that works for something that does not exist yet. Or may I ask you to sell/throw away your GPU and stop playing games now because in a few months something better will be released? At least in that case a better GPU is a certainty, a middleware API based on OpenCL/Compute is a panacea right now. Like I said I've been waiting for alternatives for 3 years and no one has even shown a damn demo.
In the case of proprietary technology, such as PhysX, I would argue that GPGPU has everything to do with how many cards are released and, more importantly, sold. Moreover, I do not believe that the pc conglomerate (intel, AMD, Nvidia, software developers) would ever collectively seek to bring about a situation where we are not encouraged to upgrade, less so where the situation benefitted one company (Nvidia via its PhysX) whilst adversely affecting everybody else. Indeed, the standards laid out in DirectX stand in direct opposition to such practices.
Whilst we're comparing CPU's and GPÙ's, how would off-loading work to the graphics card in any way benefit me? Wouldn't I eventually be required to go Sli or Crossfire simply because physics calculations had completely tied up one card? I only look at benches when I am in the market for new hardware, but as far as I'm aware games continue to tax the GPU much more than the CPU. I am unsure to what extent, if any, a Phenom II or a 775-socket quad would act as a bottleneck, but I have a feeling that much more work is expected of the graphics card and few games are able to take advantage of multiple threads. Doesn't that mean that even on quads that are a few years old, we have a couple of cores twiddling their thumbs in most games, whilst the graphics card is being increasingly taxed? Isn't it inefficient to off-load further tasks to the graphics card in these circumstances, assuming that my naive and rather simplistic appraisal of the situation is correct?
You are not understanding my point. I'm talking about future developments, posible future developments. The scalar performance of CPUs has not increased almost one bit since 2005. Everything they have done is increase parallel and vectorized execution and make parallel tasks faster. That is good when you can parallelize your program, but my uncle uses some old programs which do not parallelize and guess what? His P4 3.8 GHz is WAY faster than his Quad core at 2.6 GHz.
Intel and AMD should try to increase scalar performance, putting all their efforts and silicon into making it faster in those tasks instead of keeping adding and adding more FPUs and cores. A 6 core CPU does not things faster than a dual core CPU, except for those tasks that are parallel and those tasks run much faster on a GPU.
So make a fast dual CPU or make a fast 6-8 core asymetrical CPU, which has 2 complete cores and 4-6 small cores which lack complex FPUs and vector units and can accelerate the execution of simple tasks and pass parallel and vectoriced FP performance to GPUs which are orders of magnitude faster doing so. Also improve the platform so that the GPU is not bottlenecked.
Take any review and you will see where quads are faster than duals and 6-cores are faster than quads and it's always 3dsmax, maya, photoshop, video conversion etc. A 4/6 core CPU is completely useless for anything that is not that kind of tasks and a GPU can do those tasks 10 times faster so why not pass all those tasks to the GPU and streamline the CPU so that it's faster in a linear fashion and much smaller and cheaper?
Same with games, the games that require you to have a quad are those which have lots of physics and AI running off the CPU, both tasks can be done faster and more efficiently on the GPU.
You think it's a problem to need SLI/Crossfire, but you are completely wrong. I don't know how much you paid for your i7, but I do know that I have a Q6600 and I would have to pay at least 500 euros to buy i7, and it wouldn't be faster than my quad (at similar Mhz) except on those tasks that would benefit running in a better parallel execution processor like a GPU. What if I could buy an especialized $50 dual core CPU like I said and a GPU in the lines of HD6850/GTX560 and that was much faster than a future 12 core CPU in video transcoding, 3dsmax rendering, game physics, AI, etc, etc, etc? What's left? Where would a 12 core CPU be faster than the option that I'm presenting? Nowhere. All we need is fast dual cores in order to speed up linear performance and the ability to move parallel data better to GPUs, so they can act like math co-processors. In the past moving the co-processor, the FPU, inside the CPU was a big move, but that's because that allowed for decreased latencies and higher clocks. Well, now we don't need faster clocks (we can't actually, we hit a physicsal wall) and the FSB is much faster than it was. The clock limit was reached 5 years ago, and the biggest limitation is die size and heat nowadays, there's no reason not to move things to different dies and there's no need to duplicate efforts, adding useless vectorial FP performance to CPUs (adding cores that only kick in vectorial tasks or with AVX or whatever means) when there's already a processor that does vectorial FP tasks much much faster than CPUs and that it's 90% of the time idling: the GPU. Even on games, not a single one game pushes GPUs to 100% shader utilization, because other parts like ROPs and textures are a bottleneck, so use that 10% of shader power. 10% of the FP performance in any modern GPU is way more than what even the fastest i7 is capable off, in parallel tasks.
