crytek crysis dx10 vs. dx10.1

[Shtal]

I have 2 questions?

If Crytek were to release to Crysis 2 with DX10.1 support.

Question 1: ??
Will crysis game engine benefit of faster FPS with DX10.1

Question 2: ??
Will IQ quality improve with DX10.1 vs. DX10 V-High detail level option currently in Crysis

 

[almighty]

I know one thing, it'll boost performance on ATI DX10.1 enabled cards, looks at the performance increase Asassins Creed got from running in DX10.1 vs DX10

 

[Arnold Beckenbauer]

I know one thing, it'll boost performance on ATI DX10.1 enabled cards, looks at the performance increase Asassins Creed got from running in DX10.1 vs DX10

I'm afraid, it would cost too much performance.
http://delivery.acm.org/10.1145/129...coll=ACM&dl=ACM&CFID=15151515&CFTOKEN=6184618

 

[Dave Baumann]

What exactly are you pointing to in that doc?

 

[Arnold Beckenbauer]

What exactly are you pointing to in that doc?

I'm not sure, but here and there are statements, that it would be like Supersampling.
Crysis creates its Depth Buffer first. And this Depth Buffer is used later for almost everything. With a "MSAA Depth Buffer" you will do shading not only per-pixel but per-sample.

 

[homerdog]

Would a "MSAA depth buffer" allow AA to work on the foliage?

Per-sample shading, although a performance killer, would probably help the IQ in Crysis tremendously. I'm thinking of those POMed surfaces in particular.

 

[Dave Baumann]

I'm not sure, but here and there are statements, that it would be like Supersampling.
Crysis creates its Depth Buffer first. And this Depth Buffer is used later for almost everything. With a "MSAA Depth Buffer" you will do shading not only per-pixel but per-sample.

IIRC thats more or less what happened with Assassins Creed.

However, you are considering DX10.1 to only affect MSAA, which is inaccurate. There are many other areas where you can get quality or performance benefits from - STALKER, for instance, gets its DX10.1 performance differences by changing the shadowing mechanism.

 

[Humus]

With a "MSAA Depth Buffer" you will do shading not only per-pixel but per-sample.

It's an option, but certainly not required. And I doubt many developers would actually do that. The two most common choices would be to either grab one sample, or resolve the depth buffer to a non-MSAA buffer and use that.

Would a "MSAA depth buffer" allow AA to work on the foliage?

Just to clarify, MSAA depth buffers are as old as MSAA itself. You need one to use MSAA at all since the depth buffer and color buffer must have matching MSAA mode. What's new in DX10.1 is a sampleable MSAA depth buffer. This doesn't affect foliage in any particular way. However, DX10.1 also offers per-sample fragment shader evaluation, which can indeed be useful for foliage and allow proper antialiasing on alpha edges. It would be better looking that alpha-to-coverage, but at a performance cost. DX10.1 also has a programmable sample mask, which you could use to create your own alpha-to-coverage equivalent implementation, where you could use a better algorithm than the built-in hardware implementation, and should not be much slower than regular alpha-to-coverage. Also, that way you can return a useful non-coverage alpha value from the shader, which is often a very desirable feature.

 

[Davros]

wasnt the assasins creed thing either (depending on who you beleive)

faster, but it made ati cards look good and being a twimtbp game nv made them retract it
or
faster because it was broken and missed some stages out ?

 

[willardjuice]

Probably the former.

The whole point of DX 10.1 in AC (iirc) was it saved an extra rendering pass (or something along those lines).

 

[Arnold Beckenbauer]

It's an option, but certainly not required. And I doubt many developers would actually do that. The two most common choices would be to either grab one sample, or resolve the depth buffer to a non-MSAA buffer and use that.
.

Crysis' atmospheric effects like the global fog, fog volumes and soft z buffered particles and SSAO are using the same Z values from the Depth Buffer. Wouldn't your options break the precision of these effects?

Dave, MSAA is Crysis' biggest weak point and not shadows.
€dit: But it's a really interesting point.
Here some mouse over pics:
http://www.pcgameshardware.de/aid,665475/Test/Benchmark/Stalker-_Clear_Sky_im_D3D101-Test/?page=2
http://www.pcgameshardware.de/aid,665475/Test/Benchmark/Stalker-_Clear_Sky_im_D3D101-Test/?page=3

 

[homerdog]

Just to clarify, MSAA depth buffers are as old as MSAA itself. You need one to use MSAA at all since the depth buffer and color buffer must have matching MSAA mode. What's new in DX10.1 is a sampleable MSAA depth buffer. This doesn't affect foliage in any particular way. However, DX10.1 also offers per-sample fragment shader evaluation, which can indeed be useful for foliage and allow proper antialiasing on alpha edges. It would be better looking that alpha-to-coverage, but at a performance cost. DX10.1 also has a programmable sample mask, which you could use to create your own alpha-to-coverage equivalent implementation, where you could use a better algorithm than the built-in hardware implementation, and should not be much slower than regular alpha-to-coverage. Also, that way you can return a useful non-coverage alpha value from the shader, which is often a very desirable feature.

Oh my, thank you :smile:

 

[Humus]

Crysis' atmospheric effects like the global fog, fog volumes and soft z buffered particles and SSAO are using the same Z values from the Depth Buffer. Wouldn't your options break the precision of these effects?

None of these effects are particularly sensitive to depth buffer precision, certainly not at the sub-pixel level. There are some cases where you might want that kind of precision, like screen space shadows, so in those cases DX10.1 offers you the ability to do so, where in DX10.0 you can't, but generally speaking pixel level depth is good enough for most stuff and for games that target DX10 I don't think a lot of games are going to opt for sample level evaluation in DX10.1 for stuff that's done on per-pixel level in DX10.0.

 

[sebbbi]

Crysis' atmospheric effects like the global fog, fog volumes and soft z buffered particles and SSAO are using the same Z values from the Depth Buffer. Wouldn't your options break the precision of these effects?

You don't even need full pixel resolution for SSAO. For example Starcraft 2 uses a 2x2 downsampled depth buffer for SSAO. Volumetric particles and fog do not need pixel precision either. Many games are rendering smoke and other low frequency particles at much lower than screen resolution (2x2, 4x4 or 8x8) to save fill rate (8x8 downsampled smoke/fog is 64 times faster to render).