AMD: R8xx Speculation

[MfA]

Good looking displacement actually needs subpixel sized triangles... and there we enter a whole new world of aliasing and sampling issues. I guess there's a reason why Tim Sweeney's talking about micropolygon rendering and software based engines - they've probably did their own displacement tests with UE4 and Gears assets already. So they know very well that this is going to be a problem.

He wants to do texture AA with them too ... so basically each micropolygon rather than half-pixel sized becomes 1/(2*anisotropy) sized. It might be the future, but it ain't the now.

BTW, to maintain sharp edges can't you just add a per vertex parameter and do the appropriate work in the hull shader? Adding a row of triangles seems like such a nasty artist hack

 

[Laa-Yosh]

Edge weighting exists in subdiv related stuff for a while now but is generally frowned upon by artists. It's a kind of invisible parameter, making it a lot harder to judge on first glance how a tesseleted model would end up looking like.
There are some other, more technical issues as well, surface tension and such, but I'd rather not get into a lengthy discussion on this...

 

[nAo]

4-6 pixel triangles will usually cover four quads, so that's far from efficient. Even if you have favourable shapes (i.e. not skinny), you'll need more than 100 pixels per triangle to break 80% utilization in the shader/pixel pipes.

Don't modern GPUs shade more than one primitive at once? If that's the case higher efficiency could be achieved even with relatively small triangles. Even though the smaller they are and the more likely will be that shading on a 2x2 quad basis will fill the shader cores with work for fragments that don't really cover any pixel.

 

[PhilTaylor]

You only need one leaky board partner. but hey.. you're talking about a card that nvidia says is going to launch in six weeks.. they should have samples out by the bucket-loads already

just as an fyi, samples do seem to be out.

[Quote:]
Originally Posted by DennyA
TXNetCop, not sure where you're hearing GTX300 rumors, but the Tesla chips aren't out in the wild at all yet. As i understand it, the number of existing cards is still in the double digits -- it's a 2010 product, and they only got the first chips taped out a couple of months ago.

So yeah, the drivers would be pretty raw at this point.

Hey Denny, it's not rumors we are one of the test facilities
Ted
[/Quote]

http://www.sim-outhouse.com/sohforums/showthread.php?t=24622

surprised me too, but hey. FSX was a G80 launch partner (and I do have a greengoo iPod laser etched to commemorate the date ) so it looks like the FS community is still in the loop.

dont underestimate nV,for all their faults, they are hard-chargers when it comes to engineering.

 

[MfA]

Edge weighting exists in subdiv related stuff for a while now but is generally frowned upon by artists. It's a kind of invisible parameter, making it a lot harder to judge on first glance how a tesseleted model would end up looking like.

How would it be any less intuitive to see the hard edge/crease in say a different color than to see an extra line of degenerate triangles along the edge?

 

[Laa-Yosh]

Listen, this tech hasn't been invented today, it's been available in 3D modeling packages and offline rendering for about a decade. And yet noone's really using it.

There are many reasons and explanations but it's all 3d modeling specific and I really wouldn't like to spend an hour explaining it all...

 

[Lightman]

See http://forum.beyond3d.com/showthread.php?t=55330

I've mist such a great thread!

 

[Mintmaster]

Don't modern GPUs shade more than one primitive at once? If that's the case higher efficiency could be achieved even with relatively small triangles. Even though the smaller they are and the more likely will be that shading on a 2x2 quad basis will fill the shader cores with work for fragments that don't really cover any pixel.

They do quads from multiple primitives per batch, but I'd be shocked if they did more than one primitive per quad since all the sampling hardware is quad-based.

I suppose it's possible to have quad packing hardware with a way of storing multiple parameter sets per quad and taking variable cycles per interpolation, LOD calc, and load issue, but it seems too complicated to me and not worth the effort throughout the pipeline.

EDIT: And when you consider AA needing multiple sample masks, even pixel output gets complicated. Remember that for every triangle, RV870 will be able to output 8 quads and execute 200 scalar instructions on each, so it's not likely that rasterization efficiency for small triangles is going to be a big problem.

 

[Voxilla]

Two new DX11 demos on AMD's dev page

Really interesting demos, only they require Win7 to run, why would that be ???

With a little bit of effort I got them running on Vista, there seems no reason why they wouldn't.

Here how to get them running:

Just clicking on the msi installer file will not work as it will tell that the application requires Windows 7.
With a tool like Universal Extractor 1.6 the msi can be unpacked (use the msi2xml extraction option).
In a 'files' subdirectory you will find all required files. They don't have an extension, so insert a '.' before the three last letters of the file name. This enables the demos to run on Vista.
Before those DX11 demos can run, you need to install DX11 first though.

 

[neliz]

Really interesting demos, only they require Win7 to run, why would that be ???

Before those DX11 demos can run, you need to install DX11 first though.

Looks like you're giving yourself a good answer! With DX11 not being available "publicly" yet for Vista, it would be the current thought process on requiring 7 for it.
Though I admit, especially dev's couldn't care a rats ass about what's available publicly. A snafu I presume.

 

[Voxilla]

Looks like you're giving yourself a good answer! With DX11 not being available "publicly" yet for Vista, it would be the current thought process on requiring 7 for it.
Though I admit, especially dev's couldn't care a rats ass about what's available publicly. A snafu I presume.

If you are in marketing this must make a lot of sense.
The upgrade to Vista DX11 is publicly available for anyone to install as described above.

 

[Dave Baumann]

DX11 runtime is not available on Windows update yet. Additionally the demos have, so far, been validated on 5800 and Windows 7; likewise you can run them on a 5700 but we're going back and validating them at the moment.

 

[neliz]

The upgrade to Vista DX11 is publicly available for anyone to install as described above.

Imagine you need a guide like this to install something that should be a run-time update.

DX11 runtime is not available on Windows update yet. Additionally the demos have, so far, been validated on 5800 and Windows 7; likewise you can run them on a 5700 but we're going back and validating them at the moment.

I can imagine the questions and Flame AMD would get when the demo's don't run on Vista because of DX11.

 

[Squilliam]

Does anyone care to speculate when the chips speculated in the topic of this threads speculation will be available on Newegg, as a point of speculation?

 

[Voxilla]

DX11 runtime is not available on Windows update yet.

That's what Tom's Hardware Guide says:

"Microsoft has already uploaded the DirectX 11 update to the official Windows Update Server; it just remains hidden for now. The registry entries in our script activate a so-called beta download. Contrary to what the name may imply, this doesn’t mean that you’re about to download a beta version of DirectX 11. As you’ll be able to verify later by running DXDiag again, the version number is that of the final release of DirectX 11."

On the Microsoft site it is also explained how to get it, probably mostly intended for the development community.
And that is precisely the target audience of these AMD tech demos I would assume.

 

[Dave Baumann]

Somewhat splitting hairs there...

 

[Voxilla]

Why not just checking for the presence of DX11 instead of Win7 before refusing to install ?

 

[Jawed]

They do quads from multiple primitives per batch

Is there a test that shows this?

Jawed

 

[nAo]

Well, if they don't do this tessellation + 64 pixel batches (ATI) perf is going to suck.

 

[Humus]

Why not just checking for the presence of DX11 instead of Win7 before refusing to install ?

This is IMHO the right way to go. It bothers me when apps refuse to run because of a stupid check like this. I was able to run the demo now with your workaround.