DOCSLIB.ORG

Transformations a Quarterly Publication From

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Transformations a Quarterly Publication From issue 03 // fall-winter 07 Transformations A quarterly publication from Q&A with Marv Burkett The Future of 3D Graphics FEATURES Solving the World's Hardest Problems Welcome to Bollywood NVIDIA will celebrate its 15th year anniversary in February. As I reflect on the accomplishments and A breakthroughs that we’ve made during this time, a characteristic of NVIDIA that stands out is our ability to WORD innovate at an absolutely relentless pace, quarter after quarter, year after year. This passion for pushing the boundaries of technology and imagination drives all of our businesses—from consumer products to FROM professional solutions—and positions us firmly as a leader in the visual computing revolution. Developers, scientists, engineers, and researchers around the world are finding new and exciting ways to leverage MIKE the GPU across a wide range of areas, from film making to medical diagnosis to oil & gas exploration. From our vantage point, we see consumers with an insatiable appetite for as much processing power as we can offer, making NVIDIA a true “friend of Moore’s Law.” As I contemplate the potential of the GPU to transform industries and solve the world’s most difficult problems, I am excited to think about what the next 15 years will hold not only for NVIDIA but for our entire worldwide ecosystem of partners, developers, and consumers. Michael W. Hara VICE PRESIDENT INVESTOR RELATIONS AND COMMUNICATIONS ISSUE THREE // FALL-WINTER 07 // NVIDIA CORPORATION // WWW.NVIDIA.COM TRANSFORMATIONS // Q&A with Marv Burkett GPUs are one of the few areas that can Q&A efficiently use the doubling of transistors. with Marv Burkett CFO, NVIDIA CORPORATion Talk about the relationship between Does NVIDIA view Moore’s Law as NVIDIA and the investment community. a friend or foe? Please explain. The second change is the disappearance of the IDMs Describe some of the changes MB: I believe investors have a much better understanding (Integrated Device Manufacturers). In the 70's and MB: Moore's Law has lived longer than most expected. the semiconductor industry has about NVIDIA than they did a few years ago. Previously 80's most semiconductor companies were IDMs, It's over 40 years old and is still alive and well. At this there was the perception that graphics suppliers traded undergone over the past decade. meaning they had their own fabs. In the 80's we point, each doubling of the # of possible transistors position every cycle. This wasn't true, but it was a saw the start of the fabless business model and the creates significant challenges and opportunities. We common perception with investors. Now they understand MB: I see two significant changes in the industry emergence of pure foundries. The prohibitive cost of are at the point where each successive generation of that with few exceptions, NVIDIA has been and continues that are evident today. The first is the entrance of fabs, coupled with the skyrocketing cost of process technology adds not thousands or even millions of to be the technology leader. private equity into the semiconductor industry. The development, has forced many IDMs to abandon the possible transistors, but now a new generation could add acquisition of On Semiconductor, Freescale, and Phillips fab strategy. The announcement by TI that they were a billion transistors or more. The industries or companies The other perception that has changed is the inherent Semiconductor, all by private equity groups, signals a no longer going to develop process technology was a that can use a billion more transistors are dwindling. gross margins in our business. For a while, when we change in the perception of the business. Private equity sea change in the industry. In the U.S. that leaves only Memories can always use more density, so Moore's were improving gross margins, there was the perception has been around for a long time, but until recently they Intel and IBM as leading-edge process developers Law would be their friend. CPUs may be able to use that soon gross margins would fall and return to the have been unwilling to invest in the semiconductor and will eventually leave only Intel as an IDM. The the additional transistors, but can they do it efficiently? old levels. It is only with our continued progress and business. The primary reason for that is that in the past, emergence of companies like TSMC with significant That is, is a dual core processor twice as powerful as a an understanding of inherent causes of the change, semiconductor businesses were consumers of cash, leading edge fab capacity has not only allowed this single core? Is a quad core four times as powerful and that they have changed their minds. Now they believe not cash generators. If private equity uses significant change, but forced it, with very cost effective capacity. useful as a single core? Probably not. Therefore I think we have fundamentally changed the business model debt to finance their purchases, then they are unwilling There is no longer an advantage to owning your fab. it can be argued that Moore's Law is not the friend of to invest in cash consumers because they need to be for revenue per wafer, which changes their view of CPU companies. GPUs are one of the few areas that can able to service the debt. That has changed in the last our inherent gross margins. Investors are interested in efficiently use the doubling of transistors. This means ten years. Now most semiconductor companies are earnings growth. What excites them about NVIDIA is that with a doubling of transistors, GPU designers can generating cash because they are outsourcing the fab that we have the potential for both a.) revenue growth, double the performance. How long this will go on, we and/or sharing R&D costs. This attracts private equity which can lead to earnings growth, and b.) expansion of don't know, but for the foreseeable future, GPUs have and impacts the valuations of semiconductor companies. gross margins, which will also lead to earnings growth. ways of doubling the performance with successive generations. So certainly Moore's Law is NVIDIA’s friend. TRANSFORMATIONS // NVIDIA CORPORATION // WWW.NVIDIA.COM TRANSFORMATIONS // NVIDIA CORPORATION // WWW.NVIDIA.COM PROFESSIONAL TRANSFORMATIONS // The Future of 3D Graphics SOLUTIONS GPUs have evolved far beyond simply implementing a fixed function graphics The Future of 3D Graphics pipeline to becoming flexible, program- Spotlight on GPU Computing and Ray Tracing THE FIRST IN A SERIES OF ARTicLES AboUT 3D GRAPHicS BY Nvidia'S CHIEF SciENTIST, david KIRK mable, massively parallel computers. In recent years we’ve seen mapping units. We never looked tremendous interest in back—the current GeForce 8800 chip has 128 processor cores. And, utilizing the immense parallel not only does it have 128 processor processing power of GPUs cores, but each core can run many for uses beyond classic 3D threads, or program copies, at a graphics processing. GPUs time. The GeForce 8800 processes have evolved far beyond over 12,000 threads at once—each thread processing pixels, vertices, simply implementing a fixed or triangles! Imagine achieving that function graphics pipeline kind of parallelism and throughput to becoming flexible, FIGURE 1 FIGURE 2 with dual-or quad-core CPUs. It’s programmable, massively GPUs Lead Evolution to desire to use the power of the at ever higher clock rates. However, model aided by graphics APIs and just not possible. But, that’s not Many-Core Processing GPU for broader applications it’s quite easy to add multiple CPU the C programming language is all. In addition to the 12,000+ pixel parallel computers. Similarly, The programmable and flexible than just graphics. More recently, cores to a single chip—but that’s straightforward and easy to use. or vertex threads, there are many 3D graphics has evolved modern GPU is one of the most this broader effort, which we call where the simplicity ends. It is Although applying GPUs to a variety thousands of other concurrent to encompass many powerful computing devices on GPU Computing, has been made difficult for programmers to grasp of parallel computing tasks is a operations being processed by the forms of visual computing the planet. Since the year 2000, easier by the introduction of how to program multi-core CPUs natural evolution, trying to process GPU. Texture map calculations, applications. GPUs are now the individual processing cores NVIDIA’s CUDA (Compute Unified effectively. Also, for the first time in demonstrably parallel graphics rasterization, Z-buffer hidden- within GPUs have processed data Device Architecture) programming several decades, programmers can workloads with multi-core CPUs is surface-removal, color blending considered “computational using IEEE floating-point precision, environment. CUDA allows GPUs to no longer simply wait 18-24 months inherently challenging—because for transparency, and anti-aliasing graphics” engines, as just like standard CPUs (aka “real be programmed using the C language for their single-threaded programs simply grouping together many CPUs (edge smoothing) are all happening many of the fixed function computers”). The raw floating-point for non-graphics applications. to double in speed as processor will not produce an integrated parallel simultaneously. Without the special- parts of the graphics processing power of a modern GPU clock speeds increase. An industry processor. A GPU consists of many purpose hardware included in every pipeline have become is much larger and growing faster All processors evolve and change wide effort to “refactor” algorithms parallel processor cores integrated to GPU to perform these operations, than even the latest multi-core with time, not just GPUs. We are to run on multi-core CPUs is taking work together from the ground up.
Load more

Recommended publications
  • Conservation Cores: Reducing the Energy of Mature Computations
    Conservation Cores: Reducing the Energy of Mature Computations Ganesh Venkatesh Jack Sampson Nathan Goulding Saturnino Garcia Vladyslav Bryksin Jose Lugo-Martinez Steven Swanson Michael Bedford Taylor Department of Computer Science & Engineering University of California, San Diego fgvenkatesh,jsampson,ngouldin,sat,vbryksin,jlugomar,swanson,[email protected] Abstract power. Consequently, the rate at which we can switch transistors Growing transistor counts, limited power budgets, and the break- is far outpacing our ability to dissipate the heat created by those down of voltage scaling are currently conspiring to create a utiliza- transistors. tion wall that limits the fraction of a chip that can run at full speed The result is a technology-imposed utilization wall that limits at one time. In this regime, specialized, energy-efficient processors the fraction of the chip we can use at full speed at one time. Our experiments with a 45 nm TSMC process show that we can can increase parallelism by reducing the per-computation power re- 2 quirements and allowing more computations to execute under the switch less than 7% of a 300mm die at full frequency within an same power budget. To pursue this goal, this paper introduces con- 80W power budget. ITRS roadmap projections and CMOS scaling servation cores. Conservation cores, or c-cores, are specialized pro- theory suggests that this percentage will decrease to less than 3.5% cessors that focus on reducing energy and energy-delay instead of in 32 nm, and will continue to decrease by almost half with each increasing performance. This focus on energy makes c-cores an ex- process generation—and even further with 3-D integration.
    [Show full text]
  • Nvidia® Gelato™ 1.0 Hardware
    NVIDIA GELATO PRODUCT OVERVIEW APRIL04v01 NVIDIA® GELATO™ 1.0 HARDWARE- Key to this doctrine of no compromises is Gelato’s new shading language incorporates a ACCELERATED FINAL-FRAME RENDERER Gelato’s use of NVIDIA graphics hardware. simple and streamlined syntax based on C, Gelato is breakthrough, rendering software Gelato uses the NVIDIA Quadro FX as a second making it familiar and easy for most from NVIDIA, designed with a new architecture floating-point processor, taking advantage of programmers to learn and allowing for state- that leverages advances in mainstream graphics the 3D engine in ways far beyond gameplay. of-the-art shader-specific types and hardware to accelerate film-quality rendering. Gelato is one of the first in a wave of software functions. Gelato ships with an extensive This software renderer takes advantage of the applications that use the graphics hardware as set of shader libraries and examples. programmability, precision, performance, and an off-line processor, a “floating-point Gelato is available with a world-class quality of NVIDIA Quadro® FX professional supercomputer on a chip,” and not simply to support package, backed by NVIDIA, graphics solutions to render imagery of manage the display. the global leader in 3D graphics. uncompromising quality at unheard-of speeds. FAST AND GETTING FASTER The annual support package includes Gelato offers all the features film and television all product updates and upgrades. customers demand today and is flexible and Gelato unleashes the processing power of the extensible enough to satisfy their future graphics hardware that currently sits idle on LOOKING TO THE FUTURE requirements.
    [Show full text]
  • FCM 61 Italiano
    Full Circle LA RIVISTA INDIPENDENTE PER LA COMUNITÀ LINUX UBUNTU Numero #61 - Maggio 2012 AUDIO FLUX NUOVA SEZIONE MUSICA GRATIS IN CC foto: downhilldom1984 (Flickr.com) CCOOPPIIAA EE CCOODDIIFFIICCAA DDII DDVVDD QQUUAATTTTRROO SSIISSTTEEMMII CCRROONNOOMMEETTRRAATTII EE PPRROOVVAATTII full circle magazine n.61 1 Full Circle magazine non è affiliata né sostenuta da Canonical Ltd. indice ^ HowTo Full Circle Opinioni LA RIVISTA INDIPENDENTE PER LA COMUNITÀ LINUX UBUNTU Python-Parte33 p.07 Rubriche LaMiaStoria p.38 UsareilcomandoTOP p.10 NotizieLinux p.04 AudioFlux p.52 LaMiaOpinione p.42 VirtualBoxNetworking p.15 Comanda&Conquista p.05 GiochiUbuntu p.53 IoPensoChe... p.43 GIMP-BeanstalkParte 2 p.21 LinuxLabs p.29 D&R p.50 RecensioneLibro p.45 Torna Prossimo Mese Inkscape-Parte1 p.24 DonneUbuntu p.XX Chiuderele«Finestre» p.32 Lettere p.46 Grafica Gli articoli contenuti in questa rivista sono stati rilasciati sotto la licenza Creative Commons Attribuzione - Non commerciale - Condividi allo stesso modo 3.0. Ciò significa che potete adattare, copiare, distribuire e inviare gli articoli ma solo sotto le seguenti condizioni: dovete attribuire il lavoro all'autore originale in una qualche forma (almeno un nome, un'email o un indirizzo Internet) e a questa rivista col suo nome ("Full Circle Magazine") e con il suo indirizzo Internet www.fullcirclemagazine.org (ma non attribuire il/gli articolo/i in alcun modo che lasci intendere che gli autori e la rivista abbiano esplicitamente autorizzato voi o l'uso che fate dell'opera). Se alterate, trasformate o create un'opera su questo lavoro dovete distribuire il lavoro risultante con la stessa licenza o una simile o compatibile.
    [Show full text]
  • The Utilization Wall
    UC San Diego UC San Diego Electronic Theses and Dissertations Title Configurable energy-efficient co-processors to scale the utilization wall Permalink https://escholarship.org/uc/item/3g99v4qd Author Venkatesh, Ganesh Publication Date 2011 Peer reviewed|Thesis/dissertation eScholarship.org Powered by the California Digital Library University of California UNIVERSITY OF CALIFORNIA, SAN DIEGO Configurable Energy-efficient Co-processors to Scale the Utilization Wall A dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Computer Science by Ganesh Venkatesh Committee in charge: Professor Steven Swanson, Co-Chair Professor Michael Taylor, Co-Chair Professor Pamela Cosman Professor Rajesh Gupta Professor Dean Tullsen 2011 Copyright Ganesh Venkatesh, 2011 All rights reserved. The dissertation of Ganesh Venkatesh is approved, and it is acceptable in quality and form for publication on microfilm and electronically: Co-Chair Co-Chair University of California, San Diego 2011 iii DEDICATION To my dear parents and my loving wife. iv EPIGRAPH The lurking suspicion that something could be simplified is the world's richest source of rewarding challenges. |Edsger Dijkstra v TABLE OF CONTENTS Signature Page.................................. iii Dedication..................................... iv Epigraph.....................................v Table of Contents................................. vi List of Figures.................................. ix List of Tables................................... xi Acknowledgements...............................
    [Show full text]
  • 3D Computer Graphics Compiled By: H
    animation Charge-coupled device Charts on SO(3) chemistry chirality chromatic aberration chrominance Cinema 4D cinematography CinePaint Circle circumference ClanLib Class of the Titans clean room design Clifford algebra Clip Mapping Clipping (computer graphics) Clipping_(computer_graphics) Cocoa (API) CODE V collinear collision detection color color buffer comic book Comm. ACM Command & Conquer: Tiberian series Commutative operation Compact disc Comparison of Direct3D and OpenGL compiler Compiz complement (set theory) complex analysis complex number complex polygon Component Object Model composite pattern compositing Compression artifacts computationReverse computational Catmull-Clark fluid dynamics computational geometry subdivision Computational_geometry computed surface axial tomography Cel-shaded Computed tomography computer animation Computer Aided Design computerCg andprogramming video games Computer animation computer cluster computer display computer file computer game computer games computer generated image computer graphics Computer hardware Computer History Museum Computer keyboard Computer mouse computer program Computer programming computer science computer software computer storage Computer-aided design Computer-aided design#Capabilities computer-aided manufacturing computer-generated imagery concave cone (solid)language Cone tracing Conjugacy_class#Conjugacy_as_group_action Clipmap COLLADA consortium constraints Comparison Constructive solid geometry of continuous Direct3D function contrast ratioand conversion OpenGL between
    [Show full text]
  • VMD User's Guide
    VMD User’s Guide Version 1.9.4a48 October 13, 2020 NIH Biomedical Research Center for Macromolecular Modeling and Bioinformatics Theoretical and Computational Biophysics Group1 Beckman Institute for Advanced Science and Technology University of Illinois at Urbana-Champaign 405 N. Mathews Urbana, IL 61801 http://www.ks.uiuc.edu/Research/vmd/ Description The VMD User’s Guide describes how to run and use the molecular visualization and analysis program VMD. This guide documents the user interfaces displaying and grapically manipulating molecules, and describes how to use the scripting interfaces for analysis and to customize the behavior of VMD. VMD development is supported by the National Institutes of Health grant numbers NIH P41- GM104601. 1http://www.ks.uiuc.edu/ Contents 1 Introduction 11 1.1 Contactingtheauthors. ....... 12 1.2 RegisteringVMD.................................. 12 1.3 CitationReference ............................... ...... 12 1.4 Acknowledgments................................. ..... 13 1.5 Copyright and Disclaimer Notices . .......... 13 1.6 For information on our other software . .......... 15 2 Hardware and Software Requirements 17 2.1 Basic Hardware and Software Requirements . ........... 17 2.2 Multi-core CPUs and GPU Acceleration . ......... 17 2.3 Parallel Computing on Clusters and Supercomputers . .............. 18 3 Tutorials 19 3.1 RapidIntroductiontoVMD. ...... 19 3.2 Viewing a molecule: Myoglobin . ........ 19 3.3 RenderinganImage ................................ 21 3.4 AQuickAnimation................................. 21 3.5 An Introduction to Atom Selection . ......... 22 3.6 ComparingTwoStructures . ...... 22 3.7 SomeNiceRepresenations . ....... 23 3.8 Savingyourwork.................................. 24 3.9 Tracking Script Command Versions of the GUI Actions . ............ 24 4 Loading A Molecule 26 4.1 Notes on common molecular file formats . ......... 26 4.2 Whathappenswhenafileisloaded? . ....... 27 4.3 Babelinterface .................................
    [Show full text]
  • C 2009 Aqeel A. Mahesri TRADEOFFS in DESIGNING MASSIVELY PARALLEL ACCELERATOR ARCHITECTURES
    c 2009 Aqeel A. Mahesri TRADEOFFS IN DESIGNING MASSIVELY PARALLEL ACCELERATOR ARCHITECTURES BY AQEEL A. MAHESRI B.S., University of California at Berkeley, 2002 M.S., University of Illinois at Urbana-Champaign, 2004 DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Computer Science in the Graduate College of the University of Illinois at Urbana-Champaign, 2009 Urbana, Illinois Doctoral Committee: Associate Professor Sanjay J. Patel, Chair Professor Josep Torrellas Professor Wen-Mei Hwu Assistant Professor Craig Zilles ABSTRACT There is a large, emerging, and commercially relevant class of applications which stands to be enabled by a significant increase in parallel computing throughput. Moreover, continued scaling of semiconductor technology allows us the creation of architectures with tremendous throughput on a single chip. In this thesis, we examine the confluence of these emerging single-chip accelerators and the appli- cations they enable. We examine the tradeoffs associated with accelerator archi- tectures, working our way down the abstraction hierarchy of computing starting at the application level and concluding with the physical design of the circuits. Research into accelerator architectures is hampered by the lack of standard- ized, readily available benchmarks. Among these applications is what we refer to as visualization, interaction, and simulation (VIS). These applications are ide- ally suited for accelerators because of their parallelizability and demand for high throughput. We present VISBench, a benchmark suite to serve as an experimen- tal proxy for for VIS applications. VISBench contains a sampling of applications and application kernels from traditional visual computing areas such as graphics rendering and video encoding.
    [Show full text]
  • Enabling Compute-Communication Overlap in Distributed Deep
    Enabling Compute-Communication Overlap in Distributed Deep Learning Training Platforms Saeed Rashidi∗, Matthew Denton∗, Srinivas Sridharan†, Sudarshan Srinivasan‡, Amoghavarsha Suresh§, Jade Nie†, and Tushar Krishna∗ ∗Georgia Institute of Technology, Atlanta, USA †Facebook, Menlo Park, USA ‡Intel, Bangalore, India §Stony Brook University, Stony Brook, USA [email protected], [email protected], [email protected], [email protected] Abstract—Deep Learning (DL) training platforms are built From other server nodes by interconnecting multiple DL accelerators (e.g., GPU/TPU) Datacenter Network 1) Mellanox 2) Barefoot via fast, customized interconnects with 100s of gigabytes (GBs) 3) Inception of bandwidth. However, as we identify in this work, driving CPU CPU 4) sPIN this bandwidth is quite challenging. This is because there is a 5) Triggered- pernicious balance between using the accelerator’s compute and Based memory for both DL computations and communication. PCIe PCIe PCIe PCIe This work makes two key contributions. First, via real system Switch Switch Switch Switch measurements and detailed modeling, we provide an under- A 7) ACE A A standing of compute and memory bandwidth demands for DL NPU 0 F NPU 1 AFI F NPU 4 F NPU 5 I I I compute and comms. Second, we propose a novel DL collective ACE Accelerator communication accelerator called Accelerator Collectives Engine Fabric (AF) (ACE) that sits alongside the compute and networking engines at A A A A NPU 2 F NPU 3 F F NPU 6 F NPU 7 the accelerator endpoint. ACE frees up the endpoint’s compute I I 6) NVIDIA I I Switch-Based and memory resources for DL compute, which in turn reduces Datacenter AFI network PCIe link NIC the required memory BW by 3.5× on average to drive the same network link link network BW compared to state-of-the-art baselines.
    [Show full text]
  • Spatial Data Structures, Sorting and GPU Parallelism for Situated-Agent Simulation and Visualisation
    Spatial Data Structures, Sorting and GPU Parallelism for Situated-agent Simulation and Visualisation A.V. Husselmann and K.A. Hawick Computer Science, Institute for Information and Mathematical Sciences, Massey University, North Shore 102-904, Auckland, New Zealand email: f a.v.husselmann, k.a.hawick [email protected] Tel: +64 9 414 0800 Fax: +64 9 441 8181 March 2012 ABSTRACT Spatial data partitioning techniques are important for obtaining fast and efficient simulations of N-Body par- ticle and spatial agent based models where they con- siderably reduce redundant entity interaction computa- tion times. Highly parallel techniques based on con- current threading can be deployed to further speed up such simulations. We study the use of GPU accelerators and highly data parallel techniques which require more complex organisation of spatial datastructures and also sorting techniques to make best use of GPU capabili- ties. We report on a multiple-GPU (mGPU) solution to grid-boxing for accelerating interaction-based models. Our system is able to both simulate and also graphi- cally render in excess of 105 − 106 agents on desktop Figure 1: A visualisation of a uniform grid datastructure hardware in interactive-time. as generated by GPU, using NVIDIA’s CUDA. KEY WORDS thermore, a good scheme will also support - and not grid-boxing; sorting; GPU; thread concurrency; data impede - the introduction of parallelism into the com- parallelism putation so that an appropriate parallel computer archi- tecture [5] can be used to further reduce compute time per simulation step. As a vehicle, we use the Boids 1 Introduction model originally by Reynolds [6, 7].
    [Show full text]
  • July/August 2021
    July/August 2021 A Straight Path to the FreeBSD Desktop Human Interface Device (HID) Support in FreeBSD 13 The Panfrost Driver Updating FreeBSD from Git ® J O U R N A L LETTER E d i t o r i a l B o a r d from the Foundation John Baldwin FreeBSD Developer and Chair of ne of the myths surrounding FreeBSD is that it • FreeBSD Journal Editorial Board. is only useful in server environments or as the Justin Gibbs Founder of the FreeBSD Foundation, • President of the FreeBSD Foundation, foundation for appliances. The truth is FreeBSD and a Software Engineer at Facebook. O is also a desktop operating system. FreeBSD’s base sys- Daichi Goto Director at BSD Consulting Inc. tem and packages include device drivers for modern • (Tokyo). graphics adapters and input devices. Consistent with Tom Jones FreeBSD Developer, Internet Engineer FreeBSD’s role as a toolkit, FreeBSD supports a variety • and Researcher at the University of Aberdeen. of graphical interfaces ranging from minimalist window managers to full-featured desktop environments. The Dru Lavigne Author of BSD Hacks and • The Best of FreeBSD Basics. first article in this issue walks through several of these Michael W Lucas Author of more than 40 books including options explaining how users can tailor their desktop • Absolute FreeBSD, the FreeBSD to their needs. It also provides pointers to downstream Mastery series, and git commit murder. projects which build an integrated desktop system on Ed Maste Senior Director of Technology, top of FreeBSD. The next two articles dig into the details • FreeBSD Foundation and Member of the FreeBSD Core Team.
    [Show full text]
  • Graphics Hardware
    Department of Computer Engineering Graphics Hardware Ulf Assarsson Graphics hardware – why? l About 100x faster! l Another reason: about 100x faster! l Simple to pipeline and parallelize l Current hardware based on triangle rasterization with programmable shading (e.g., OpenGL acceleration) l Ray tracing: there are research architetures, and few commercial products – Renderdrive, RPU, (Gelato), NVIDIA OptiX – Or write your own GPU ray-tracer 2 3 Perspective-correct texturing l How is texture coordinates interpolated over a triangle? l Linearly? Linear interpolation Perspective-correct interpolation l Perspective-correct interpolation gives foreshortening effect! l Hardware does this for you, but you need to understand this anyway! 4 5 Recall the following l Before projection, v, and after p (p=Mv) l After projection pw is not 1! l Homogenization: (px /pw , py /pw , pz /pw , 1) l Gives (px´, py ´ , pz´ , 1) " v % " v % " 1 0 0 0 % x x $ '$ ' $ ' 0 1 0 0 $ vy ' $ vy ' p = Mv = $ ' = $ ' $ '$ ' 0 0 1 0 v $ vz ' $ '$ z ' # 0 0 −1/ d 0 &$ ' $ ' # 1 & −vz / d 6 # & Texture coordinate interpolation l Linear interpolation does not work l Rational linear interpolation does: – u(x)=(ax+b) / (cx+d) (along a scanline where y=constant) – a,b,c,d are computed from triangle’s vertices (x,y,z,w,u,v) l Not really efficient to compute a,b,c,d per scan line l Smarter: – Compute (u/w,v/w,1/w) per vertex – These quantities can be linearly interpolated! – Then at each pixel, compute 1/(1/w)=w – And obtain: (w*u/w,w*v/w)=(u,v) – The (u,v) are perspectively-correct interpolated l Need to interpolate shading this way too – Though, not as annoying as textures l Since linear interpolation now is OK, compute, e.g., Δ(u/w)/ Δx, and use this to update u/w when stepping in the x- direction (similarly for other parameters) 7 Put differently: l Linear interpolation in screen space does not work for u,v l Solution: – We have applied a non-linear transform to each vertex (x/w, y/w, z/w).
    [Show full text]
  • Gelato Pro 2.0 and Gelato 2.0 Gpu-Accelerated Final-Frame Renderer
    GELATO PRO 2.0 AND GELATO 2.0 GPU-ACCELERATED FINAL-FRAME RENDERER NVIDIA Sorbetto™ is a feature set Speed Without Gelato / Gelato Pro Sacrificing Quality that allows you to change any lighting parameter you wish and re-render the Selected Feature Comparison NVIDIA® Gelato® is built on the principle frame in record time. Best of all, Sorbetto Gelato of never compromising on the quality of performs this relighting on the final pixels. Features Gelato Pro the rendered image. Gelato uses NVIDIA The image you work with in Sorbetto is GPU acceleration graphics processing units (GPUs) as always identical to the final rendered image a general purpose GPU and not just and you do not need to alter or specially Highest quality images as a way to fill pixels on the display. It prepare your scene to use Sorbetto. renders images quickly and is robust Raytracing, incl. global Sorbetto uses the same geometry, and powerful enough to render the most illumination and ambient shaders, procedurals, and hardware complex scenes while avoiding the quality occlusion as a regular Gelato render. Sorbetto limitations of hardware-generated images. functionality is exposed in the Maya plug- High-order geometry support Gelato seamlessly leverages the advances in (and soon in the 3ds Max plug-in) and found in the latest GPU hardware while in the C++ API, so it can be used with any Fully programmable shading protecting your investment in production existing lighting tool in your pipeline. Sorbetto interactive tools and digital assets by insulating them Best of all, Gelato Pro is backed by the relighting from hardware changes.
    [Show full text]