Siggraph Asia December 2011

Size: px

Start display at page:

Download "Siggraph Asia December 2011"

Martha Newton
5 years ago
Views:

1 Siggraph Asia December 2011

2 Advanced Graphics Always Core to NVIDIA Worldwide Leader in GPU Development & Professional Graphics

3 Advanced Rendering Commitment 2007 Worldwide Leader in GPU Development & Professional Graphics World Leader in Photorealistic Rendering for Companies wanting Integrated Solutions

4 Multiple A.R. Commitments World Class Middleware Application Acceleration Engines for Developers to Exploit the GPU Worldwide Leader in GPU Development & Professional Graphics DevTech Pro Graphics & HPC World Leader in Photorealistic Rendering for Companies World Class wanting Development Integrated Assistance Solutions for Companies to Exploit the GPU Many Priv. App s

5 Consolidated A.R. Commitment Today Worldwide Leader in GPU Development & Professional Graphics The Center of Gravity for Advanced Rendering and Professional Graphics Many Priv. App s

6 NVIDIA ARC: Integrating to Accelerate Disciplines of mental images are now combined with their NVIDIA counterparts to merge expertise for: Discipline NVIDIA mental images Ray Traced Rendering OptiX, NV Research mental ray, iray GPU computing OptiX, CUDA, HPC, etc iray, New Rendering Raster Rendering OpenGL, Cg, Tegra Realtime Rendering Scene Graphs SceniX Rendering Platforms Remoting Monterey Cloud Rendering Application Integration DevTech Rendering Integration Partner Engagements Strategic Alliances Business Development The mental images name has been retired, with agreements now being with: NVIDIA ARC GmbH

7 NVIDIA Advanced Rendering Importance Graphics is Core to NVIDIA s Foundation; Advanced Rendering is Core to its Graphics Future Inspire the industry with rapid advancements of what GPUs enable Commercial Products (ARC) complete a vital Feedback Loop: Advanced Rendering Languages & Tools Professional Products Professionals GPU Design GPUs Result: best of class solutions exploiting & influencing the full ecosystem

8 NVIDIA ARC: Middleware & Products Middleware free tools for exploiting the GPU Libraries and SDKs that are general, low level building blocks Good choice for developers with domain expertise needing custom solutions which they create and maintain Examples: OptiX, SceniX from ARC; Thrust, CUBLAS from NVIDIA Products licensed, complete rendering solutions Commercial components for professional rendering Good choice for companies wanting to add proven solutions to their products which are maintained and advanced for them Examples: mental ray, iray, etc.

9 NVIDIA ARC: Rendering Options mental ray platform focusing on Film Production needs iray integration platform focusing on Interactive Design needs, with rapid integration OptiX ray tracing framework focusing on general & custom GPU ray tracing development Strong teams on each technology

10 NVIDIA ARC: mental ray platform

11 NVIDIA ARC: mental ray platform Focusing on the needs of Film Production Integral within most major Animation and CAD products StandAlone (from Autodesk) for specialized studio pipelines Advancements Coming this year: Production ready lights and materials (including BSDFs) for hair, skin, volumetric effects, and more Fast and accurate motion blur with full ray tracing (vs. raster) Fast, efficient, and noiseless GI for massive scenes Improved rendering performance Soon gaining GPU acceleration in some areas

minutes 2 CPU ~18X 20 seconds 1 GPU Rendered with mental

12 NVIDIA ARC: mental ray AO mental ray code 3.11 & pipeline accelerated w/ NVIDIA OptiX ~21mil faces +3 minutes 2 CPU ~18X 20 seconds 1 GPU Rendered with mental ray 3.9 Subsequent frames can be *far* faster yet Model courtesy NVIDIA Creative

13 NVIDIA ARC: iray integration platform

14 NVIDIA iray: roadmap iray 2 Layered Material Model (car paint, subsurface scattering, decals, etc.) Increased Interactivity, Daylight Portals, Clipping Planes, more iray integration platform v1 iray 3 Formalized coming soon coming soon Better convergence for more difficult lighting conditions Increased production flexibility (Light Path Expressions, matte objects, etc.) Additional features (rounded corners, etc.) Increased interactivity

15 NVIDIA iray integration platform Focusing on the needs of Interactive Design Cross platform, quick to integrate, and inherently interactive Bunkspeed Shot & Move, Catia V6 Options for Cluster & Cloud Rendering and Multi-Rendering Comes with numerous SDK source code samples e.g., an interactive viewer application Additional renderers are straight forward for NVIDIA to add Additional renderers can communicate with one another Share data, simultaneously render, cross-fade, etc. Flexible licensing coming for private application usage

16 What we were making: application platforms RealityServer neuray API Application Application commands stream commands stream API Web Services API Web Services API neuray API neuray API Distribution API Distribution

17 What we re making now: the world s best rendering framework cluster & cloud rendering iray integration framework Cloud Rendering Client Application API Platform iray IRT OGL API Iray Platform API Distribution cloud cluster

18 NVIDIA iray integration platform (summer 2012) Cooperating renderers delivering a quality/speed continuum Real Time Raster Rendering Interactive Ray Tracing iray FPS Stereo Game Title Quality 15 FPS* Multi-Pass Effects Raster AO Soft Shadows, etc. 20 FPS 0.5 FPS* Accurate Reflections Soft Shadows Accurate Shadows Glossy Reflections Multi-Bounce Diffuse, etc. 10 FPS Degraded Simplified Minutes* Uncompromised Quality Increased Flexibility Strength: Weakness: Very High Resolutions Ray Tracing w/o Noise Physically Based Very Approximate Not Physically Based Noisy while Resolving End users will experience these as different modes that blend into one another

iray integration platform Details Blending Quality & Speed Smooth switching between

exposable as a dial State changes on user actions, mouse events, wait times,

Performace (FPS) iray Interactve Ray Tracing Realtime Raster Performance Solid on CPUs

19 iray integration platform Details Blending Quality & Speed Smooth switching between rendering modes Flexible APIs for controlling effects and modalities, that is exposable as a dial State changes on user actions, mouse events, wait times, resolutions, etc. Performace (FPS) iray Interactve Ray Tracing Realtime Raster Performance Solid on CPUs and highly optimal on NV GPUs Scenes can exceed GPU memory and still gain from GPU acceleration Highly scalable across cores, processors, and nodes Realism Size (Screen/Model)

support measured BRDF (in time) Procedural Operators (extensible) Procedural Textures Hierarchal Branching (mix,

20 iray integration platform Shared Materials Derived from physically based definition (but can diverge) BSDF based (with layering) iray result used as the gold standard, with each renderer adapting as it sees fit Designed to support measured BRDF (in time) Procedural Operators (extensible) Procedural Textures Hierarchal Branching (mix, blend, etc.) Compiles for both GPU compute and x86 for maximum performance Designed to be adoptable by third party renderers

21 NVIDIA ARC: OptiX ray tracing engine

NVIDIA ARC: OptiX ray tracing engine A cross-platform ray tracing framework for developers Similar to OpenGL in doing the heavy lifting of ray tracing, allowing developers

Proven to speed development as well as performance Single ray execution model for ease of programming Minimal CUDA knowledge required Very general and applicable to many

22 NVIDIA ARC: OptiX ray tracing engine A cross-platform ray tracing framework for developers Similar to OpenGL in doing the heavy lifting of ray tracing, allowing developers to concentrate on capabilities and techniques Provides acceleration structures, traversal, scaling, scheduling, etc. Proven to speed development as well as performance Single ray execution model for ease of programming Minimal CUDA knowledge required Very general and applicable to many markets Not tied to shading language Customizable ray pay loads, intersection, cameras, procedurals, etc. In use within hundreds of private applications Uses range from studio pipelines to acoustics, ballistics, and signal processing

23 OptiX in use at Adobe Shown at SIGGRAPH 2011 After Effects with a full 3D pipeline Ray Tracing for final frame, 100% on OptiX Using CPU fallback for Mac & Windows

24 FPS FPS OptiX 2.5 Preliminary Tests 2.5GB 6GB 2.5GB projected quad core projected quad core # of 4k Images Millions of Textured & Smoothed Faces Quadro 6000 = 6GB on board memory Quadro 5000 = 2.5GB on board memory One time speed cliff drop; results with 1 GPU remain faster than multiple CPUs Example: entire 777 data set (15GB+) running on a single Quadro 6000 at 5 FPS

GPU 2mil tries in 10ms, 10mi tri s in 50ms Version 3 coming soon Optimized for

25 NVIDIA ARC: OptiX Roadmap Version 2.5 coming soon Out-of-core support, paging to system RAM Very fast BVH building on GPU 2mil tries in 10ms, 10mi tri s in 50ms Version 3 coming soon Optimized for NVIDIA Kepler GPU Architecture Cross platform CPU fallback (for interactive rendering) Being leveraged by iray integration platform

26 NVIDIA ARC: Seismic Data Visualization (SDV) Seismic volumetric data Horizon data Real-time visualization Interactive color map changes Supports today s huge and future massive dataset sizes cluster-aware scalable solution for CPU/GPU

27 SDV: Advanced Visual Quality and Accuracy Horizon data embedded in seismic volume data Depth-correct transparency Coming from NVIDIA ARC as a licensable technology for makers of Oil & Gas tools

28 SDV: Remote Visualization Client side Viewer Inter- or Intranet commands Video stream Web Services Server side Subsurface Data Visualization Framework

29 Many Programming Approaches in Use iray CUDA C, C Runtime finalrender CUDA C, C Runtime Furry Ball CUDA C, C Runtime Arion CUDA C, driver API Octane CUDA C, driver API V-Ray RT OpenCL Optis OpenCL CentiLeo CUDA C, driver API Massive Out of Core Panta Ray CUDA C, driver API Massive Out of Core OptiX (2.5) CUDA C, driver API & PTX (Out of Core) Adobe Research (in our booth) CUDA C, OptiX API Works Zebra, etc. (at our booth) CUDA C, OptiX API mental ray CUDA C, OptiX API

30 NVIDIA Goals for Advanced Rendering Make the GPU an essential part of ALL advanced rendering Create engines and libraries to make it easier for everyone to exploit the GPU (e.g., OptiX); learning what s needed for future GPU languages and architectures Create compelling commercial offerings to spur GPU adoption (e.g., iray); learning what s needed for success, so to help other developers and improve the GPU platform for it If you re developing solutions for Advanced Rendering, NVIDIA wants to help

Solutions Vary in their GPU Exploitation

typically ray trace 6 to 15 times faster than

challenge is to keep the GPU busy Gains on

ones Particularly evident with multiple GPUs,

Can mean the technique needs to be rethought

31 Solutions Vary in their GPU Exploitation Speed-ups vary, but a top end Fermi GPU will typically ray trace 6 to 15 times faster than on a quad-core CPU A GPGPU programming challenge is to keep the GPU busy Gains on complex tasks often greater than for simple ones Particularly evident with multiple GPUs, where data transfers impact simple tasks more Can mean the technique needs to be rethought in how it s scheduling work for the GPU CPU OptiX 2.1 example first tuned for simple, now tuned for complex, with a 30-80% speed increase 2010 GPU

32 Questions?

GPU Ray Tracing at the Desktop and in the Cloud. Phillip Miller, NVIDIA Ludwig von Reiche, mental images

GPU Ray Tracing at the Desktop and in the Cloud Phillip Miller, NVIDIA Ludwig von Reiche, mental images Ray Tracing has always had an appeal Ray Tracing Prediction The future of interactive graphics is