Dragon Age II DX11 Technology Andreas Papathanasis BioWare
Sequel to the successful RPG Dragon Age: Origins Focus on combat, magic, cinematics Rebooted Art Style, focus on Graphics technology Out next week on PC / Xbox 360 / PS3
Our pitch for DX11 technology We discussed within our project tiered options for taking advantage of next generation PC GPUs: Tier 0: DX9 with very high quality shaders Tier 1: Basic DX11 integration No compute or tessellation shaders Takes advantage of shader models 4 and 5 Tier 2: Typical DX11 integration Compute and tessellation shaders fully supported Tier 3: Advanced DX11 integration Multithreading Requires significant refactoring of the higher level engine Each tier costs more than the previous one. But also brings more benefits to the project and the studio’s technology in the long run. We went to our PM with these options, presenting benefits and cost to implement each tier. We recommended Tier 2 as the best tradeoff between benefit and engineering/testing effort. I will present the same arguments it took for us to convince our project this is worthwhile. Picture shows an example screenshot from the pitch document that demonstrates potential terrain rendering improvements with DX11 + tessellation technology.
Why DirectX 11? Opportunities to integrate well documented, impactful visual features using SM4/SM5, new shader stages Guaranteed feature set (no cap bits) Great way for programming and art teams to prepare for the next generation Powerful, evolving debugging tools Guaranteed functionality means we no longer have to code for different implementation levels and provide alternatives based on caps checks. Another unexpected advantage from DX11 was advanced GPU debugging tools from both hardware vendors, that work on DX11 only. These tools helped us identify and fix issues with all versions of our game. We expect many of the techniques we investigated (and likely the API itself) to be very relevant next generation
GPU PerfStudio 2 – Shader Debugging on DX10/DX11
Why DirectX 11? Steam hardware survey Reliable backwards compatibility with older hardware Customers have the hardware and expect games that use it Steam hardware survey The main reason we still support a DX9 rendering path is support for Windows XP. We expect our next project to be DX11 only. Various data shows a wide adoption of DirectX 10 and higher GPUs. Customers who invented in them are expecting to get something more out of PC games.
DirectX 11 features overview Dynamic lighting Tessellation Diffusion Depth of Field Contact hardening soft shadows
Dynamic Lighting
Dynamic Lighting
Dynamic Lighting Light buffer Based on the popular light pre-pass method http://www.wolfgang-engel.info/blogs/?p=26 Extra pass to render scene normals Render all dynamic light spheres to a light buffer Allows for hundreds of lights Minimal engineering effort to integrate on DX11: Multiple render targets work with MSAA Guaranteed render target format support Small per-draw overhead on the CPU Makes a lot of sense for a spell-rich game like Dragon Age II Light buffer Our base DX9 forward rendering architecture only allows for 2 dynamic lights per object.
Tessellation Big research focus for the team Great potential for taking geometric detail to the next level Lends itself to LOD naturally, without affecting CPU or complicating the engine codebase
Tessellation challenges Piecewise level construction can cause gaps between adjacent objects when normals don’t match at the edges would require content to be redone for DX11 fell back to parallax mapping techniques for displaced look ongoing research topic for our next projects Some content (i.e. characters) is already tessellated at desired density for non DX11 platforms On Dragon Age II we used tessellation for terrain silhouette smoothing (using PN triangles algorithm)
Tessellation on terrain Without tessellation With tessellation PN triangles algorithm is ideal for surfaces that are supposed to be smooth (like terrain here).
Tessellation performance VERY easy to kill performance with excessive tessellation Performance improvements that made a difference in our engine (sorted in order of the performance win we got): Only tessellate objects close enough to the camera LOD in the hull shader View frustum culling (important for high polygon objects that span a large area, like terrain) Distance adaptive tessellation factor Only tessellate objects that will end up being visible (occlusion culling even more relevant if you’re rendering lots of tessellated objects)
Diffusion Depth of Field Custom multi-pass Diffusion DOF solver Technique will be described by Holger Gruen later today, and appear in Game Engine Gems 2 Much higher quality than our DX9 DOF solution, almost as efficient
Diffusion Depth of Field DX9 DOF Diffusion DOF
Diffusion Depth of Field DX9 DOF
Diffusion Depth of Field Diffusion DOF
Contact hardening shadows
Contact hardening shadows High quality shadow filter Accelerated by GatherRed/GatherCmpRed Shader Model 5 instructions Shader source code in the DirectX SDK Trivial to integrate, great visual quality improvement for directional shadows
Contact hardening shadows Old PCF filter Non uniform contact hardening filter
References Light pre-pass PN triangle implementation http://www.wolfgang-engel.info/blogs/?p=26 PN triangle implementation June 2010 DirectX SDK, PNTriangles11 sample Diffusion DOF solver Game Engine Gems 2 (http://www.gameenginegems.net/) Contact hardening soft shadows June 2010 DirectX SDK, ContactHardeningShadows11 sample
Conclusion API and market is mature Lots of algorithms available that can enhance your particular style of game Initial investment in DX11 integration should pay off quickly Hardware vendors can help you You should weight the cost of integrating DX11 technology against the tradeoffs. The sooner you do it . For us, we got significant benefits in the quality of our first DX11 game. Both hardware vendors are extremely helpful when it comes to providing integration and optimization assistance. They can also suggest features that can benefit your game.
Q & A DirectX 9 DirectX 11 andreas@bioware.com