Yeah but a 256-bit GDDR5 part definitely wouldn't have been a good idea considering the issues with GDDR5 availability now. Imagine if Nvidia cards also needed it?

What issues are they? I'm not aware of any.

I think it's going to be fascinating to see over the next couple years what happens on the software side here. You've got three players who have so many cross-currents of relative mutual and competing interests between them that it's incredibly complicated and unpredictable how it will come out.

1). The *market* would obviously much prefer an industry standard, like the market always does.
2). The IHVs to some degree like proprietary because they can charge a premium for it.
3). Otoh, no IHV would enjoy being left out of an industry standard if one *did* materialize, no matter how much they might try to sabotage one coalescing in advance.
4). An industry standard that you can be a part of while your enemy is avoiding it can be a beautiful thing (see NV overtaking 3dfx in part by leveraging DX while 3dfx clung to Glide).

Add to those general principles, the following observations about the players:

A) Would the two existing discrete graphics champs, nVidia and AMD, prefer to gang up against Intel? GPGPU clearly got Intel alarmed, and thus Larrabee was born. And a genuinely alarmed Intel is a bit scary, as it suggests they won't just take their ball and go home if v1 doesn't go as well as hoped.
B) Would Intel and nVidia rather gang up against AMD? On what terms? Given Intel's huge software mountain to climb in the first place, might they be grateful for a CUDA while at the same time sticking the shiv to AMD? Would NV be happy to go along in order to ensure CUDA as the defacto standard in the space?
C). Would the two CPU guys prefer to gang up and destroy cpu-independant GPGPU in its crib before it can really gain a foothold, then go back to battling each other like always? For Intel this might be attractive because the less expertise on the GPU part, the better for them, so kill nvidia's ambitions (Intel might even call them "pretensions" :) ) in this area first. Better the devil you've known (and generally beaten) for twenty years, than those ambitious upstarts. For AMD, they might also see (Fusion certainly suggests it) the utility of a closer relationship between GPU and CPU down the road. So why not gang up with Intel now to destroy Nvidia, then compete with Intel from a position of at least having a better GPU base to start from? But, uh, somebody better be able to write some pretty decent soup-to-nuts software in that potential (if temporary) partnership, and right now that's looking like the fly in the ointment.

Yep, strange days indeed. Most peculiar, momma.

So why are the 4870's trickling in and where's the 1GB version?

Production isn't gated by GDDR5 availability. There are three vendors who have as much supply as we need.

Currently Intel and AMD are ganging up against NVIDIA. Intel pushes Crossfire solutions for their platforms, opens up Havok to AMD.. They just prefer saying 'ATI' and using only the ATI logo for these occasions. :lol:

Intel and Nvidia are certainly very cold at the moment, cooperation there would be strange to see, very strange. Nvidia and AMD would be less strange but it doesn't look like that's going to happen either.

All three of them are cooperating in the Khronos group to develop 'OpenCL' (or whatever the final name will be).

Since this is a software issue you can't ignore Microsoft and Apple ... OpenCL basically is Apple. If Apple open sources their LLVM modifications (which I assume their compiler is based on) I think it will be the defacto winner.Will also open some more interesting GPGPU research rather than the map X to CUDA/G80 what passes for it now ...

Although it's probably possible to use OpenCL on Windows (and other OS), but it's unlikely to be able to bridge with Direct3D, and that could be a problem for some people. Currently, CUDA is able to bridge with both Direct3D and OpenGL, so technically it's still possible that OpenCL can bridge with Direct3D, but it's more of a political problem rather than a technical problem.

On NVIDIA hardware CUDA (or PTX) would almost certainly be used as an intermediary language, so if OpenCL is open source politics will stop no one from writing a language extension for working on Direct3D objects.

I don't know, but I'd guess that OpenCL is going to include some runtime functions. Of course, it may feature some extension mechanisms like OpenGL to provide some ways to bridge with Direct3D, but that would not be very "standard" anymore (unless, of course, both NVIDIA and ATI decided to support the same extension).