Your Wish List : Things that didn't make D3D10

[Andrew Lauritzen]

It's important to clear the stencil buffer every frame, not just once, even if you completely fill it again and again.

Again, I definitely did all of that; I went through the entire checklist with NVIDIA to no avail. It may be fixed in their latest drivers, but my point was that early stencil is a bit brittle, which I think you guys have all pretty well confirmed

 

[KindDragon]

You can easily do single pass stereo rendering in DX10 with the GS.

But GS slow for that and D3D10 doesn't support stereo backbuffer
In software I can only do anaglyph mode

D3D need also return global separation return for better stereo support

 

[KindDragon]

- Much more advanced texture compression based on wavelets ( like a JPG2000/WMP, but with block decoding ). This can be questionable, specially with GDDR prices coming down but.... well, some graphics cards can decode MPEG in HW so this could be possible.

and GPU Texture Compression for HDR textures

 

[Andrew Lauritzen]

and GPU Texture Compression for HDR textures

There are many papers on this subject that don't require any hardware support... see this year's I3D for example.

 

[KindDragon]

API for simple rendering cube with mapped volume texture or volume data generated by special shader.

 

[Megadrive1988]

I was going to make a thread to ask this question but I'll ask it here instead.

for DX11 and the hardware that supports it (next-next gen GPU architectures) is a Programmable Primitive Processor still something that's desirable? or does the Geometry Shader do the things a PPP would've done?

I recall a few years ago, here, on this very board and elsewhere that a built-in / on-chip (or off-chip) Programmable Primitive Processor was something that was very much wanted for upcoming GPUs / graphics cards, much like T&L was desired in the late 1990s before NV10 came out.


also, if the Geometry Shader does not fill the role of a PPP, does ATI's programmable tesselation unit in R600 (and Xenos) fill that role? IIRC, ATI's tesselation unit is not required by DX10 and Nvidia doesn't have one in G80. even if ATI's tesselation is good and even if it filled the role of a PPP, it won't get much support in games since it's not in the DX10 spec.

I haven't heard any talk on PPP lately so I thought I'd bring it up again.

 

[3dcgi]

The GS is a programmable primitive processor since it operates on primitives and is programmable. It can perform tessellation, though dedicated hardware can be faster.

 

[KimB]

Just fyi, it looks like nVidia's next iteration of workstation cards (Tesla) will support 64-bit floating point, due out late this year. The consumer versions of these cards will not.

 

[Mariner]

Just fyi, it looks like nVidia's next iteration of workstation cards (Tesla) will support 64-bit floating point, due out late this year. The consumer versions of these cards will not.

Tesla is the HPC derivative of NVidia's technology isn't it?

In the past there have always been those who tried to modify their GeForce cards whether through drivers or BIOS adjustment to become cheaper Quadro-equivalents. I wonder if this will be at all feasible with Tesla?

Personally, I'm guessing not, as the layout for the Tesla cards seem to be a fair bit different to their consumer graphics stuff.

No doubt there will be many who try some driver/BIOS hackery all the same!

I wonder, will the next gen of Quadro cards also support 64-bit FP or is this not really necessary?

 

[KimB]

I wonder, will the next gen of Quadro cards also support 64-bit FP or is this not really necessary?

I suppose it all depends upon whether or not Tesla and Quadro are one and the same.

 

[IgnorancePersonified]

stuff that isn't vendor specific

 

[Rys]

I suppose it all depends upon whether or not Tesla and Quadro are one and the same.

Certain Quadros will have FP64 exposed for CUDA :smile: