There's not much difference between the capabilities of both APIs; D3D9 includes some features (like hardware geometry instancing) that aren't in core GL 2.1, OpenGL is a little easier in that it lets you mix and match old-style (immediate mode/fixed pipeline) code with newer-style (VBOs/shaders) code more naturally, but overall they're mostly the same and - especially if you're using VBOs and shaders - the learning curve is more or less identical. Driver support for D3D tends to be more robust, but that's irrelevant for you since you're on NVIDIA (which has a very good OpenGL driver). OpenGL has a lot of really old and/or esoteric capabilities that are either poorly supported or not supported at all in hardware. D3D tends to expose more of the ugliness of dealing directly with hardware, OpenGL shields you from it a little better, but the tradeoff is that D3D gives you slightly better hands-on control of some things.

But on balance you can treat them as being identical capabilities-wise; there really is very little difference.

The general recommendation that you'll get is that longer-term you should learn both, so the only real question is: which do you learn first? That's down to your own personal preference really, so why not try a basic "first triangle" program in both and see how you get on, then make a decision?

Thank You very much, I was waiting for this kind of anwser. I will render triangle in both and i will see what I like more.

I don't think rendering a single triangle means much. Rendering a single triangle in OpenGL is extremely simple compared to DirectX (as far as I know), but that really doesn't mean anything in the long run, so don't judge by it.

The point of rendering a single triangle is to just determine how comfortable the OP is with the general structure and usage of the API. In actual fact, using VBOs and shaders, the D3D code needed is something like 4 lines whereas the GL code is nearer 20 or so. Without VBOs and shaders the D3D code comes in at maybe 3 lines (init the data, set the FVF, DrawPrimitiveUP). The old perspectives on this are not really valid anymore.

The point of rendering a single triangle is to just determine how comfortable the OP is with the general structure and usage of the API. In actual fact, using VBOs and shaders, the D3D code needed is something like 4 lines whereas the GL code is nearer 20 or so. Without VBOs and shaders the D3D code comes in at maybe 3 lines (init the data, set the FVF, DrawPrimitiveUP). The old perspectives on this are not really valid anymore.

What about initialization, matrix setup, etc? And I'd guess the the OP would use immediate mode stuff, but maybe I misunderstood. Anyway, I still don't think a single triangle would be a good way to judge the comfortableness of the API:s. But I may be wrong.

Well it's more about coding style which seems to be the OP's primary concern. I'm thinking that a relatively short program should be sufficient to determine if the D3D style is completely objectionable or not.

Init in D3D9 is roughly the same amount of code as in GL, by the way. Matrixes can be slightly more; D3D9 does have a matrix stack interface which matches GL's more or less exactly, but you do also need an explicit SetTransform call.