Session Title

Speaker(s)

Description

Advanced Rendering Techniques with DirectX 11

Justin Hensley and Jason Yang, AMD

The advanced features of DirectX 11, such as DirectCompute 11, enable developers to implement unprecedented real-time effects in games. This presentation covers two such effects: order-independent transparency, and high-quality depth of field.  The techniques leverage features available only as part of Direct3D 11 and DirectCompute 11. In particular, OIT takes advantage of the AppendBuffers, as well as atomic memory operations. Using these advanced features, our algorithm creates lock-free data structures in memory to store the transparent fragments as they are rendered. Once all the transparent fragments have been enumerated, a second pass is used to put the fragments in the correct order and to resolve the final fragment color. Depth-of-Field also takes advantage of atomic memory operations, and additionally uses locally shared memory for optimized performance. The Depth-of-Field algorithm uses a novel image processing technique to implement a high-quality effect not seen in current games. Both techniques were recently demonstrated running in real-time on AMD's next generation hardware, which was native support for DirectX 11, including DirectCompute 11.

Best Practices for DirectX 11 Development

Andy Pomianowski, AMD

This talk takes a whistle-stop tour through the powerful new features provided in DirectX 11. We investigate the new capabilities such as DirectCompute and Tessellation, and discuss implementation guidelines that will steer you towards the discovery of high performance solutions and aid in the development of efficient DirectX 11 applications.
 

Block Compression Smorgasbord

David Cook

DXT style block compression has become a de facto standard for most textures in games.  We will cover several topics of recent interest in this area:

1. Fast block compression. With some optimization effort, real-time block compression is feasible, on either the CPU or the GPU. Real-time compression opens up a host of possibilities for disk space reduction and dynamic texture updates.  

2. Normal map compression. Block compression was designed for color data but adapted for use with normal maps. The results are not always pretty. What usage patterns should be favored, or avoided, in this context?

3. New block compression formats. DirectX 11 introduced two brand new formats:  BC6H for HDR textures and BC7 for high quality LDR textures. We discuss how and when these formats should be employed, and how to handle the enormous search space for compression.
 

Building an Uber-Fast Crowd Renderer for Your Next Xbox 360 Engine

Tomas Vykruta

Rendering thousands of highly detailed, unique soft-skinned characters with shadows at 30 Hz on an Xbox 360 console requires a highly tuned rendering pipeline. A new type of instancing renderer that uses the GPU’s memexport feature is introduced here and compared to some other common, but less optimal, forms of crowd rendering. The renderer uses very limited CPU resources and is memory-conscious. This talk showcases an interactive, real-time demo.  Some features of the crowd renderer are: Highly tuned VMX animation; Memexport, vfetch, and tfetch techniques; Entire crowd is skinned and transformed with a single draw call and exported to memory for rapid depth prepass, shadow generation and color generation passes.

The Dark Art of Shadow Mapping

David Tuft

Adding shadows depth maps to a title is straightforward and simple. Making the shadows look good while staying within artistic, performance, and memory constraints can be very difficult. This talk explains how common shadow mapping artifacts occur and gives some basic guidelines that should be followed at all times. Next, the talk explains advanced techniques that benefit shadow maps, such as cascaded shadow maps, filtering, pancaking, and deferred shadows. Finally, the intricacies and caveats of combining these techniques are described.  Attendees should come away understanding how to identify artifacts, which techniques will mitigate them, and at what cost.
 

Deferred Shading with DirectCompute

Andrei Tatarinov and Alexander Kharlamov, NVIDIA

In this talk, we show how to implement a deferred shading pipeline using DirectCompute. We first provide an overview of a traditional deferred shading engine. Then we identify the main bottlenecks and limitations, and discuss how we can use DirectCompute in order to avoid them. We show how, by using compute, it’s possible to significantly reduce bandwidth requirements when accumulating lights by keeping the g-buffer data in shared memory, how this technique provides an elegant and efficient way of supporting MSAA, and how to reuse the same data structures employed for deferred rendering in order to render transparent surfaces in a more traditional forward rendering fashion, but applying an arbitrary number of lights in a single pass.

The purpose of this talk is not only to showcase this particular application, but in general to encourage developers to prototype innovative rendering algorithms that leverage the power of rasterization with the flexibility of compute.  The talk covers many common data parallel programming practices that are shared by many other applications. We show how to compute bounding volumes for culling using fast data parallel reductions, how to build complex data structures on the GPU using data parallel primitives such as scan, sort and compact, how to take advantage of shared memory to reduce memory bandwidth, and how to share resources between graphics and compute. Next, we discuss best practices and optimizations for NVIDIA GPUs, and show how to put them in practice in this particular application. We end the talk discussing some future research directions.
 

DirectX 11 DirectCompute–A Teraflop for Everyone


 

Chas. Boyd, Graphics PM Architect, Windows

At Gamefest 2008, we introduced DirectCompute as a new way to access the computational capability of the GPU that delivers even more flexibility and freedom for developers. It opens the door to operations on more general data-structures, and to new classes of algorithms as well. This talk quickly reviews the key features of DirectCompute’s compute shader, and then shows how they can save you when you really need that teraflop. We will show actual applications and demos of DirectCompute.
 

DirectX 11 Technology Update

Chuck Walbourn

Direct3D 11 is out and ready for use by your game today to exploit the latest in video hardware features as well as current generation machines. This talk brings you up to speed with the API, offers tips on how to get your renderer up and running, presents key feature overviews, and shows how to deploy your application. Attending this talk is highly recommended if you are attending other DirectX 11 presentations.

Fast GPU-based Scene Analysis and Post-Processing

Andy Luedke

The latest games use increasingly intensive scene post-processing techniques to deliver amazing visuals. Many of these techniques have traditionally required scene analysis on the CPU to drive adjustments to post-processing settings in a realistic way. This presentation demonstrates the use of a GPU-based sort to enable additional control of such post-processing, as well as prevent costly CPU synchronization and analysis. A variety of new post-processing effects that can be achieved will be demonstrated as well. This technique works well on Xbox 360 and scales nicely on all modern shader hardware.
 

Fluid Simulation Driven Effects in Dark Void

Bryan Dudash, NVIDIA

Fluid-driven effects are an integral part of gameplay in Dark Void. These effects are generated using a combination of accurate multigrid Eulerian fluid simulation and particle simulation, accelerated on the GPU. In this session, we explain—with the aid of live demos—how we implemented these effects, specifically talking about the fluid and particle simulation, lighting and rendering, game engine integration, and artistic control. In addition, we hope to dispel a number of myths about using fluid simulation in games, including: Fluid simulation will lead to a "smoke in the box" look; Fluid simulation is too expensive for games; and Fluid simulation is not art directable. We hope to convince the audience that highly detailed, interactive fluid simulation is not something we can only hope to see in movies or in tech demos; it is a feasible technology that can be used to make fantastic effects in real games today. We go over the details of how to implement a seamless, scalable, and directable fluid simulation in a game engine and how to use it to create the next generation of fluid-driven effects.
 

Lighting Volumes

John O’Rorke, Monolith

Lighting systems have always struggled to balance dynamic objects with high visual fidelity. Lightmaps have historically been the de facto solution, in order to decouple the cost of lighting from the cost of run-time performance. However, this breaks down in dynamic environments, and has generally resulted in a number of inconsistent lighting solutions being employed across dynamic and static objects. This talk presents a new approach for using volume textures in order to store and render lightfields entirely on the GPU. This approach is unified for dynamic and static objects of arbitrary size and complexity, and includes generation of the volumes using both direct and global illumination. The talk also discusses various techniques for rendering the volumes—covering a range of quality versus performance, yet still preserving key surface details such as normal maps and specular lighting. Finally, advanced usage of the lighting fields is discussed, such as simulating atmospheric effects using these volumes, and being able to leverage them for gameplay.
 

Realistic Rendering with Spatially-Varying Reflectance

John Snyder

In the physical world, reflectance varies spatially. Traditional computer graphics captures this as texture maps of simple BRDF parameters such as colors and specular exponents. We present new techniques for rendering reflectance more realistically, using models acquired from real materials, in real time and in the presence of environmental lighting.

Rendering with Conviction: The Graphics of Splinter Cell

Stephen Hill, Ubisoft

This session covers various aspects of the Splinter Cell: Conviction renderer, from the high-level architecture down to the low-level Xbox 360 nitty gritty. In particular, the talk focuses on our dynamic visibility and ambient occlusion solutions, which—driven by technical targets and production constraints—go somewhat against the grain of standard practice. Taken as a whole, these techniques can be adapted to boost performance or visual quality in a broad range of scenarios.
 

Still in the Trenches: More Lessons Learned About Xbox 360 Graphics Programming

Mikey Wetzel

The Xbox 360 GPU used to be a mysterious beast, but by now it’s pretty well understood. Despite implementing all the various tips and tricks over the years, it gets harder every day to find new ways to improve performance. And now that Microsoft has relaxed the Xbox 360 TCRs on resolution and MSAA, your competitors are already taking advantage of the new rules to increase scene complexity while boosting their frame rate. We present our findings on various experiments with lower-resolution rendering, and introduce the concept of sandboxing the GPU using real scene data extracted from PIX captures. Add in the usual grab bag of miscellaneous tips and techniques learned over the years and this is a must-attend event for any Xbox 360 graphics developer.
 

Stripped Down Direct3D: Xbox 360 Command Buffer and Resource Management

Wade Brainerd, Activision

Understanding how Direct3D interacts with resources and the command buffer can simplify your engine, enable cross-platform abstractions, and improve performance. This presentation dispels myths about Direct3D on Xbox 360 and describes a lower-level, engine-friendly approach to managing resources and submitting the scene.

Think DirectX 11 Tessellation! – What Are Your Options?

Xin Huang

Tessellation is one of the most exciting new features in D3D11. It makes film quality rendering possible for real-time rendering scenarios like games. With the availability of D3D11 hardware, what are your options to benefit from this new feature? How and where can you incorporate this into your game engine?

Xbox 360 Shaders and Performance: How Not to Upset the GPU

Cody Pritchard

Games are constantly pushing the limits to achieve better graphics, but unfortunately the Xbox 360 GPU is not getting any faster. Because of this, it is important to understand the inner workings of the GPU to squeeze the best possible performance out of it. This talk dives into what makes the GPU tick, in order to provide insight into why things perform the way they do. Then, armed with this information, we focus on best practices for measuring and optimizing your shaders. The goal is to promote good HLSL optimization practices, correctly identify and understand problem areas, and demonstrate tips and techniques to help improve performance. The target audience is Xbox 360 graphics developers who need to squeeze one more millisecond out of the GPU.

Back to top