DirectX Next

[Ilfirin]

I meant the size of the window where you type text in, not the text size itself. The textbox for posting is only wide enough for about a dozen words per-line, and only tall enough for a little over a dozen lines. Leads to scattered thoughts, typos, and lots of edits. Or it does for me, at least.

 

[Simon F]

I meant the size of the window where you type text in, not the text size itself. The textbox for posting is only wide enough for about a dozen words per-line, and only tall enough for a little over a dozen lines. Leads to scattered thoughts, typos, and lots of edits. Or it does for me, at least.

Ditto.
[Edit fixed spelling]
[Edit and again ]

Seriously though, it is a bit crazy that the edit box doesn't rescale with window width (assuming such a thing is possible).

 

[vb]

good thing for keeping posts short

and great thing for making excuses about spelling

not that I have much to say anyway.....

 

[JohnH]

+Move CPU/GPU transition to point of very low bandwidth

What does that mean? Are they talking about software or hardware level?

It means a reduction in data BW between CPU and GPU e.g. allowing primitives to be generated on the GPU itself.

John.

 

[JohnH]

I think it's entirely possible that "DX Next" will be too flexible. I don't think it really is all that desirable to make a GPU into a fully-generalized processor. By doing so, you inhibit the GPU's ability to make use of massive parallelism, and you add the potential for far more pipeline stalls.

It hasn't been made into a fully generalised processor, its still "stream" freindly, more I cannot say!

John.

 

[JohnH]

I meant the size of the window where you type text in, not the text size itself. The textbox for posting is only wide enough for about a dozen words per-line, and only tall enough for a little over a dozen lines. Leads to scattered thoughts, typos, and lots of edits. Or it does for me, at least.

Ditto.
[Edit fixed spelling]
[Edit and again ]

Seriously though, it is a bit crazy that the edit box doesn't rescale with window width (assuming such a thing is possible).

If you're doing a big post just use anexternal ediror...

Edit: Obviously one including a "spol chocker"

 

[zsouthboy]

Opera will increase the size of edit boxes when you zoom, btw.

<- was using Firebird for the longest time, but I switched..

 

[baj2904]

Is this the version of DX that Microsoft will use for Longhorn and Xbox 2?

 

[nobie]

+Move CPU/GPU transition to point of very low bandwidth

What does that mean? Are they talking about software or hardware level?

It means a reduction in data BW between CPU and GPU e.g. allowing primitives to be generated on the GPU itself.

John.

I wonder is HOS specifically will ever be taken advantage of. On the one hand you save bandwidth over the AGP (or PCI-express) bus, but there will be an additional computational overhead as the GPU has to create polygons from the b-spline or whatever. So which will lower performance more, using more bus bandwidth or giving more work to the GPU? I'm inclined to think bus traffic won't be a problem, especially with PCI-express.

 

[DemoCoder]

Not every geometry generation scheme is a HOS scheme.

Take for instance, a scene graph that "instantiates" geometric objects from a library multiple times.

For example, let's take a crowd scene at a stadium like in (I must say, excellent game and graphics) X-Box TopSpin, where you have hundreds of individual human models sitting on chair models at a tennis court.


Rather than expand this entire scene out and send all the geoemetry over the bus, the GPU only needs to store the model for one particular chair and human character. It can then "instantiate" a few hundred of these objects at differing starting positions, with perhaps, slightly different textures, bones, and other shader constants bound.

 

[Xmas]

And there is the advantage of adaptive tessellation.

 

[Wunderchu]

Is this the version of DX that Microsoft will use for Longhorn and Xbox 2?

Sort of ....


- Windows Longhorn will come with DirectX Next/10 already installed by default

- However, Longhorn's much talked about new 3D GUI will not use DirectX 10 .. i.e.: Longhorn's new GUI will run at its full potential, as long as you have at least a DirectX 9 card

- Xbox Next / Nextbox / Xbox 2 will feature support for an API which is similar to DirectX 10 ... it may or may not not be exactly identical to DirectX 10 ........

 

[KimB]

I'm sure it won't be identical to DX10. For a console, you want to give developers access to a lower-level interface.

 

[K.I.L.E.R]

I thought primitives could already be generated on GPUs/VPUs?

 

[DemoCoder]

Nope. Vertex shaders can't create anything. You can use a pixel shader to create "vertices" in a render target, then treat that render target as a vertex buffer, but this isn't exactly what people mean when they think of procedural geometry.

 

[NeARAZ]

Nope. Vertex shaders can't create anything. You can use a pixel shader to create "vertices" in a render target, then treat that render target as a vertex buffer, but this isn't exactly what people mean when they think of procedural geometry.

But your example of "single geometry, many instances of it" is exactly what's going on already for several years (since T&L). You just have do issue separate drawing call for each instance (or, if drawing calls do limit you, you can use various tricks to draw multiple instances in one call).
This isn't exactly "geometry generation", it's just instancing. And it's nothing new, IMHO.

 

[Kristof]

But your example of "single geometry, many instances of it" is exactly what's going on already for several years (since T&L). You just have do issue separate drawing call for each instance (or, if drawing calls do limit you, you can use various tricks to draw multiple instances in one call).
This isn't exactly "geometry generation", it's just instancing. And it's nothing new, IMHO.

Actually there are other possibilities to create instances of geometry. Take for example the possibility of automatically looping an index buffer. So you store the indices (into a vertex buffer) for say a single sphere and if you loop over that list of indices multiple times you get multiple spheres. To get the spheres into different positions you can then use a second input stream which runs at a lower rate (instead of reading a value for each index you read one every x, say 100, indices) - this stream could indicate the transformation of the sphere or a lookup value in a table of constants. Lots of possibilities... and this stops the nasty driver call overhead etc... not to mention it would be much cooler if you could actually just procedurally generate your sphere from an input containing centre point and radius. Ah well the developer in us all will never be happy with the functionality given

K-

 

[no_way]

Looks like an excellent playground for stream processor architectures, like Stanford's Imagine

 

[Bjorn]

(I must say, excellent game and graphics) X-Box TopSpin

Off topic warning but:

I used to play tennis games on the amiga back in the good old days. Haven't played one tennis game (though i haven't tried Virtua Tennis) that i have been interested in since then until i tried Top Spin. Awesome game.

 

[Mintmaster]

Finally, float RT on NVidia hardware.

New Features in 2003
D3DFMT_A16B16G16Rfc -conditional
New texture format with conditions
Same limitations as NONPOW2CONDITIONAL
No mip mapping, no wrapping, no tiling, no cube maps

I heard something about that at ATI Mojo Reloaded. In another thread I couldn't remember the details.

Did anyone come across anything related to MSAA and rendering to textures? I also heard something about this at ATI Mojo Reloaded, but the presenter went a bit quickly.