inCrysis - Crysis Forums

Lecturers:

Carsten Wenzel
Crytek GmbH

" Carsten Wenzel
R&D Software Engineer, Crytek
Carsten is a software engineer and member of the R&D staff at Crytek. During the development of "Far Cry" he was responsible for performance optimizations on the CryEngine. Currently he's busy working on the next iteration of the engine to keep pushing future PC and next-gen console technology. Prior to joining Crytek he received his M.S. in Computer Science at Ilmenau, University of Technology, Germany in early 2003. His latest contributions include two articles in "ShaderX 2" and a presentation at GDCE 2004 entitled "Porting & Optimizing Far Cry / CryEngine for AMD‘s 64-bit Architecture"."

Source

I gotta friend named SysTran:

Crytek with the siggraph
Wednesday August 2, 2006 at 15. 47
A small article to summarize the presentation of Crytek to Siggraph 2006 in Boston.

Effects atmostpheric
The course concentrated on the atmospheric effects and at all on the remainder (physical, motion blur, depth off field, lighting, shades, etc, were not approximate at all).

Are included in the atmospheric effects, the model of the sky, the total fog, the local fog (volumetric), the clouds, the effects of smoke, the effects of volumetric light, and the rivers (rivers).

depth not
Almost all these effects rest on one prépasse which fills a buffer with the values of the plane distance to the far. It is not exactly the Z-buffer, first of all because the scale is not the same one, and then because the format of storage of the zbuffer is opaque and is not exploitable generally with more difficulty directly by the pixel shader. This prépasse includes only the opaque objects, the translucent objects are problematic and will be treated separately. The format of storage is either a format 32 bits floating (R32F), or a format ARGB with coding/decoding in the pixel shader.

Returned sky
Contrary to Far Cry where the sky is painted in a skybox, the sky in next the cry engine is calculated in real time. Rather than all to calculate with high degree of accuracy, the values are calculès in a mesh weak resolution and interpolèe (with format HDR), calculation takes place on the CPU to have a greater flexibility on the platforms which do not support advanced shaders (the platform minimum of next Cry engine has a graphics board with shader model 2.0). Version D3D10 will be probably calculated on the GPU (GPUs should not have any problem with the maths of the atmospheric model).

The model in question is a version very simplifièe of Nishita and Al simplifications consist in considering that the sky is ad infinitum and that the observer is wedged on the ground. Moreover the fog and the clouds are calculated separately, the original model of scattering being too much complex for a result in real time.

The sun is also not included in the model. One of the problems is the weak resolution of the mesh sky which would cause artifacts near the sun. It is much easier to add the solar disc and the effects of flares separately.

Total fog
Rather than to calculate the fog with the fog hardware (which in any event disappears with the shader model 3), the fog in next the cry engine is added in a master key to share with a quad stretched on all the screen. The source of the pixel shader is the depth calculated in prépasse. With co-ordinates X and Y on the screen and the depth found since intermediate texture it is thus possible to find the co-ordinates of the current point.
A relatively simple formula taking account depth and height thus makes it possible to have a fog close to what exists in nature.

Rivers as the rivers profit from an effect of attenuation which uses a technique similar to that of the total fog.

local volumetric fog
Later master keys make it possible to add an arbitrary number of volumes of fog. For reasons of simplicity of the calculation of collision, the only primitives supported in the pixel shader are limp and of the ellipses. That should be enough for a certain number of volumetric effects.

Clouds
The sunlight which crosses the cloud generates a little bit of subsurface scattering. Rather than to simulate that what would be prohibitory a visual method with a chart thickness is used (the quoted reference being clouds of Fligh Simulator 2004 by Niniane Wang).

Software particles
The effects of fume and the clouds are billboards as usual, but with a trick to avoid the problem of the “hard” intersection with the objects of the decoration. By re-using the source of prépasse of depth it is possible to evaluate the distance to surface and thus to use a “software clipping” by using an attenuation factor which approaches zero more one approaches surface.

Problems
Like one on our premises says, it there does not have a “free buffet”, one of the problems of the method of returned with quads and while using a texture of depth like intermediary and that that does not go very well with the antialiasing. That should be improved with D3D10, but while waiting for that is likely to cause artifacts at the borders of the objects (if somebody remembers the first screens of Crysis, there were comments as what the screens were photoshopés, which is obviously false but this impression was caused by the volumetric fog added to the end and which gave the impression not “to stick” to the elements of the decoration like powerlines etc)

One of the other problems is the translucent or semi objects transparent. As the master key of depth can store only one depth, any superposition of objects cannot be treated with this method. The objects with blending are thus traced with share with an estimate by valuable article of the fog to the position of the object. Of course that does not go well if the translucent object is too large and covers an interval of too important fog. That requires a good collaboration on behalf of the artists not to place such objects in the levels.

Returned HDR
There was also video of Crysis in HDR on a screen HDR from Brightside, I will post photographs (LDR.) soon with a little chance.

EDITED post with proper SYSTRAN translation

Last edited by Doc (07-08-2006 09:19:08)

Actually, I used the FF extension Translation Panel 1.4.13+ which I thought used Systran which is incorrect.  It uses http://www.freetranslation.com/ amongst others which I used for this page.

SYSTRAN does a much better job so I must apologize for them.  Just plug the URL in:  http://www.systransoft.com/index.html

Thanks for that SysTran link Doc. It actually does a better job than any of those other translators

PE2: If you seek, you will find

Crytek's SIGGRAPH 2006 paper has been released. New media included.

http://www.ati.com/developer/siggraph06 … _Games.pdf

Wenzel, C. (Crytek) Real-Time Atmospheric Effects in Games. Course 26: Advanced Real-Time Rendering in 3D Graphics and Games. Siggraph, Boston, MA. August 2006.

Source: http://www.ati.com/developer/techpapers.html

The last slide is particularly interesting:

Interested in CryEngine2 HDR footage?

Check out BrightSide’s expo booth. It shows a fly through of Crysis level (Crytek’s upcoming title) captured in HDR on their latest HDR HDTV displays.

BrightSide's site: http://www.brightsidetech.com/aboutus/media.php

Note to self: Get lots of moneys and buy yourself a HDR display

Yes it looks like quite good in the pictures. Ingame it looks either better or worse.
Hopefully better.

How much do those HDR displays cost? Looks insane

Saw an article that said they would build one just for you at the paultry sum of $50,000.00

lol

worm

Still, those water pictures are only still images. The water didn't look too bad - until it moved.

However, I'm confident they'll be able to make it look good.

Nice find again, Mocib.

Have you seen the waves in Instincts? They're awesome.

As I said, I'm confident.