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 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 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. it would require hundreds if not programmable. And we have CPU. This feature has attracted a seeing this with the difficult transition place while, at the same time, the thousands of CPU cores to match only seen the beginning lot of attention in the computing that CPUs are making from single- emergence of GPU Computing gives Graphics processors have arguably the performance of a single GPU. of this transformation. community. In fact, an entirely new core to multi-core. Due to problems programmers a new powerful tool. been multi-core processors for field of effort has been spawned with heat dissipation and power Today, few application programs almost ten years. In 1998, NVIDIA’s Continued on following page>> called GPGPU, or General Purpose consumption, it is no longer possible benefit from multi-core CPUs. In TNT product was built with two Processing on GPUs, reflecting the to create faster CPUs that simply run contrast, the GPU programming pixel pipelines and two texture

transformations // nvidia corporation // www.nvidia.com transformations // nvidia corporation // www.nvidia.com professional transformations // The Future of 3D Graphics solutions

We believe the most valuable architectures are those that extend rather than disrupt the installed base, such as x86, Windows, HTML, and TCP/ IP. Preserving the industry’s investments is important not only for maintaining productivity of current applications and developers but also for encouraging investment in future applications.

Is Ray Tracing Ready for Furthermore, ray tracing is not a objects the eye sees directly). Simple conceptually simple, it is more work The GPU Can Do It All Mainstream Use? panacea or really a goal in itself, but visibility is not enough, however. than is practical with modern CPUs. The old debate of ray tracing vs. Some recent blogs and press reports rather—potentially—a way to make We also would like anti-aliasing rasterization has been going on as have commented that the future better pictures more easily. Although (i.e., the smoothing of edges). Ray Parallel GPU hardware can trace long as there has been ray tracing of 3D graphics will be based on some people may say that ray tracing tracing can accomplish this effect by rays as well. Making a picture and rasterization. The debate has the feature known as “ray tracing,” is more accurate or “the right way,” simply tracing more rays, although that is noticeably better than what now morphed into ray tracing vs. and therefore the performance of both ray tracing and rasterization this is more expensive and slower rasterization can do today is a difficult GPUs. But, comparing an approach rasterization is unimportant. While are approximations of the physical than allowing the GPU hardware to but worthy goal. That is one of the (ray tracing) with a device (GPU) is a we are enormous fans of interactive phenomena of light reflection from perform this task through rasterization most exciting parts of the field of funny way to express the comparison. ray tracing (IRT), it still requires surfaces. Neither is inherently better and dedicated anti-aliasing hardware. —we’re never That’s like asking which is better, Authored by Dr. David Kirk, a massive amount of processing or worse—just different. Three done. There is always something more fuel or automobiles? Most likely, the Chief Scientist, NVIDIA power. IRT, along with many other possible reasons for adopting ray One visual effect that is difficult to to accomplish. The picture of the answer to both questions is both. Not ideas we are pursuing, will provide at tracing are ease of programming, do well with rasterization is shadows. cathedral in this article is shown in two only do GPUs perform rasterization least another decade of innovation faster visual effect, and the possibility It is complicated to render sharp- versions. Figure 1 could be rendered efficiently using the conventional opportunity for GPU designers. The of better visual effects. Let’s talk edged shadows without having either with ray tracing or rasterization. API-based programmable graphics sustainable innovation opportunity about these reasons separately. jagged edges, and there are no really There is no special effort involved in pipeline, but GPU computing has will keep the GPU industry vibrant. robust approaches for making soft- making shadows or correct global the promise of performing other But rather than a “start from scratch” Ease of programming is edge shadows or the corresponding illumination. Figure 2 is rendered using rendering approaches as well. It is architectural approach, we believe important for 3D graphics as well effects of multiple inter-reflections global illumination. This is the effect of likely that game developers, film we need to preserve the massive as for many other applications. of light. These are visual effects for light inter-reflecting between surfaces. studios, animators, and artists investments of all the industries that Ray tracing is believed to be which ray tracing is indeed a more The back of the cathedral—behind will prefer to take advantage of all are deeply invested in OpenGL and “easier” for programmers than general solution. Also, as light reflects the light—is lit by the reflection of of the benefits of ray tracing and DirectX. We believe the most valuable rasterization because ray tracing from each surface to shine on other the light off of the surfaces in front rasterization, as well as a variety of architectures are those that extend can do everything in a single unified surfaces, every single object in a of the light. By the way, both of other techniques, all at the same time. rather than disrupt the installed approach. Although it is true that scene is both a light source and an these images are rendered using Why choose when you can have it all? base, such as x86, Windows, HTML, rays can be traced for every possible occluder (an object that blocks light). a GPU. The bottom line is that a Interestingly, one of the best GPGPU and TCP/IP. Preserving these visual effect, this is not necessarily In order to make the “perfect” picture, GPU can now make any picture applications may be…3D graphics! investments is important not only the best and fastest approach. you would need to trace a ray from that a CPU can make, rasterized

for maintaining the productivity every point on every object in every or ray traced. The combination of Nick Stam and Kevin Krewell of NVIDIA LARGE HEADER IMAGE NVIDIA’s “Human Head” of current applications and Ray tracing will never be as fast direction. A lot of rays would be special purpose GPU hardware and contributed to this article. demo, created with the GeForce 8800 Ultra GPU, developers, but also for encouraging as hardware rasterization for the required to simulate all of that light APIs (DirectX and OpenGL) with delivers a leap forward in realism, rendering the interplay of light and human skin at a level never investment in future applications. purpose of visibility (i.e., which bouncing around. Although this is computational graphics is powerful. before seen in real-time. Special thanks to actor Doug Jones for allowing the use of his likeness.

transformations // nvidia corporation // www.nvidia.com transformations // nvidia corporation // www.nvidia.com professional solutions transformations // Solving the World's Hardest Problems

Since the introduction of , the company has seen a broad range of applications begin to harness Solving the World's Hardest Problems the massive computational power of the GPU and use that to achieve unprecedented speed increases.

Somewhere deep down in the While traditional techniques UIUC is leading the way in using Similarly, Evolved Machines of Palo In the field of finance, Hanweck These are just a few of the billions of cells that comprise observe the reactions of cells, computers built with a GPU to Alto, Calif. is using GPUs to reverse- Associates has developed a real- organizations using the massive some scientists have adopted a put the equivalent of a large array engineer the circuits of the human time options implied-volatility parallel processing performance of the human body, something new approach that goes beyond of computers on the desk of any brain, first to understand how they engine, Volera. Using a single PC the GPU to solve critical problems goes terribly wrong. A observation to simulating and researcher. UIUC researchers have work and then to use those principles containing three GPUs, Volera can in fields like medicine, molecular single cell starts to divide understanding what happens on been using their NAMD (Nanoscale to create machines capable of evaluate 150,000 options per second. biology, oil & gas exploration, and uncontrollably, launching a the inside, at the most basic level Molecular Dynamics) and VMD similar functions (such as seeing and With three Tesla S870 1U servers, finance. Offering orders of magnitude process that leads to cancer. of the cell. With new computing (Visual Molecular Dynamics) software smelling). The company has, to date, Hanweck is able to calculate the improvements in simulation, techniques and high-powered arrays running on NVIDIA GPUs to conduct achieved speed increases of more than entire US option market in real-time. NVIDIA’s HPC solutions represent Why does this happen? How of computers, scientists can simulate simulations of nano-devices that 100X and has observed that a single (See http://www.hanweckassoc.com/) a discontinuity that will dramatically can it be prevented? And the basic processes of cells, from can be used to sequence DNA in desktop system containing two GPUs change the landscape of medicine once it’s started, how can the atoms up, to understand how real-time and help reduce the cost of runs as fast as a 200-core cluster, at and science. it be stopped? Thousands cells work and how they can go genomic medicine. They are seeing a fraction of the cost and power. (See wrong. But the sheer scale of this 100-240X speed increases and, www.evolvedmachines.com) For more information, visit: www. of scientists are pursuing task—simulating the parts of a cell, more importantly, are now able to nvidia.com/object/tesla_computing_ these questions with the all the interactions of these parts, run these calculations at their desks, solutions.html traditional tools of biological and how they respond to external rather than queuing up to use large science: recombinant DNA, factors—requires some of the most clusters in remote server rooms and animal studies, and chemical powerful computers in the world. waiting weeks for the results. (See www.ks.uiuc.edu/Research/vmd/) experiments—all performed New computing techniques, like to divine the nature of cells those being pioneered at the and their internal machinery. University of Illinois at Urbana- Champaign (UIUC), seek not only to expand the scope of simulation to more sophisticated situations, but also to provide a powerful new tool to any scientist in the pursuit of FAR LEFT UIUC researchers are simulating complex molecular systems (Image courtesy answers to how our bodies work. University of Illinois at Urbana-Champaign)

THIS PAGE Evolved Machines is reverse-engi- neering brain circuits to develop a new paradigm for device technology (Image courtesy Evolved Machines)

transformations // nvidia corporation // www.nvidia.com transformations // nvidia corporation // www.nvidia.com professional solutions transformationstransformations // // Welcome Lorem Ipsum to Bollywood est Dolor

Welcome to Bollywood

From villages in the north Pune, already an established world Content Creation (DCC) market space also deploying internal resources, like through involvement in ASIFA known for dairy production hub for IT industries, has also set worldwide recently launched an that of the NVIDIA Gelato development India, TASI, ABAI and other groups Bollywood Facts a more recent goal to become initiative called “Digital Bollywood.” team, to train and consult with within the DCC communities. to the idyllic beaches of the the animation and gaming hub of Indian studios in developing best • The 2006 release of summer blockbuster south, India is now witness to India and is well on its way with NVIDIA’s Digital Bollywood initiative practices for digital production. A little more than a year into the “Krrish” was the first film produced the booming industry of film the establishment of the DSK is comprised of several key elements, effort, Digital Bollywood is beginning where all the visual effects were done by India-based film studios. and game production. Often School of Animation and Gaming, including the sharing of expertise, Training to bear fruit. NVIDIA has compared to Los Angeles an offshoot of the world renowned training, and community development. Indian culture places a premium on over 95% market share in India Supinfocom animation school based education. Digital Bollywood’s training and the Gelato software renderer • Celebrated Hindi actor Shah Rukh and its madcap film industry in Valenciennes, France. Major players Expertise and education efforts occur on two is gaining traction. The recently Khan has appeared in over 60 films since growth in the 1980s, these such as Disney Feature Animation and Digital film production experience levels. First, NVIDIA has provided released film Resident Evil: Extinction 1992. His product endorsements are Indian markets are expanding Sony Pictures Imageworks are already in India has been largely limited to a Indian film schools and academies (a Sony/Screen Gems release) which some of the most valued in India today. as fast as studios can be built, present and working closely with few individuals who cut their teeth on with curricula and training materials took in over $25 million at the box their Indian colleagues, producing Hollywood and European productions on Gelato and other NVIDIA products. office during its opening weekend • Bollywood film trailers and music videos artists trained, and movies content for worldwide consumption. and returned to India to start their Second, NVIDIA has partnered with in the U.S., contained visual effects are the most popular mobile phone distributed to the theaters. own studios. Digital Bollywood is leading Indian film schools, like the rendered with Quadro and Gelato downloads after sports and fashion in India. While the number of films produced supplementing this expertise by DSK School of Animation and Gaming by Mumbai-based Anibrain. is large, film production techniques bringing digital filmmaking expertise and the Whistling Woods Academy in • The “Indian Oscars” are the Filmfare Producing over 800 films Awards, sponsored by Filmfare annually, Mumbai is the in India had remained traditionally to India and exposing more Indian Mumbai. The aim is to help develop a A key to the success of the Digital focused on 2D animation. Since the talent to the Hollywood experience. new generation of Indian filmmakers Bollywood initiative is to take the magazine, first given out in 1954. hub of all this excitement, explosion of growth, we are now One of the ways NVIDIA is doing this well-versed in digital technologies. long-term view. From the start, but production studios seeing 3D digital animation and visual is through the DCC Master’s Tour, NVIDIA realized that occasionally • The first Indian film was Raja Harishchandra, are springing up all over effects capability taking root and a quarterly speaking tour by leaders Community bringing in executives from North a silent movie made in 1913. India. The warm climes of starting to produce content for both in American and European digital Since NVIDIA does not compete with America would not suffice. Follow- Goa are calling to many, domestic and global consumption. content creation, having already any of the Indian studios and since through and long-term relationships Groups ranging from the India-based hosted the likes of Electronic Arts its products and technology are used are key to building the market. By as are Calcutta, Chennai, National Association of Software and and award-winning animator Andrew by virtually all of them, the company dedicating people knowledgeable Hyderabad, and New Delhi. Service Companies (NASSCOM) and Daffy. The talks were attended by is in a unique position to help foster about the film industry to the Combined with the very the Federation of Indian Chambers hundreds of young hopefuls and symbiotic relationships between project and leveraging substantial appealing tax incentives of Commerce and Industry (FICCI) professionals alike in numerous cities. the studios. Working with business technical capabilities resident in to the Los Angeles-based Visual In turn, NVIDIA is also identifying organizations like NASSCOM and India, NVIDIA is well-positioned to some Indian states are now Effects Society and ACM Siggraph promising students and young Indian FICCI, NVIDIA is active in sponsoring develop and sustain the effort. offering these studios, growth LARGE HEADER IMAGE Image created by Deepak Hargaonkar have all gotten onboard to help filmmakers and sponsoring their events and cooperation among of Indore, the first prize winner in the NVIDIA Gelato Imaging is certain to continue. encourage this blossoming industry. attendance at western film festivals, the studios. NVIDIA has also been Laura Dohrmann of NVIDIA Contest. SMALL HEADER IMAGES Mumbai-based Anibrain It is no wonder why NVIDIA, with where they can tap into expertise and actively involved in supporting contributed to this article. coupled NVIDIA Gelato film rendering software with NVIDIA Quadro GPUs to create realistic visual effects for Resident its significant share in the Digital build networks of contacts. NVIDIA is artistic and technical communities Evil: Extinction (a Sony/Screen Gems release). Images courtesy Anibrain.

transformations // nvidia corporation // www.nvidia.com Transformations is a quarterly publication of NVIDIA’s investor relations and communications group.

Please send feedback to [email protected].

NVIDIA Corporation | 2701 San Tomas Expressway | Santa Clara, CA 95050 | www.nvidia.com

Copyright © 2007 NVIDIA Corporation. All rights reserved. NVIDIA, the NVIDIA logo, NVIDIA GeForce, Tesla, and NVIDIA Quadro are trademarks or registered trademarks of NVIDIA Corporation in the U.S. and/or other countries. All company and/or product names may be trade names, trademarks and/or registered trademarks of the respective owners with which they are associated. Features, pricing, availability, and specifications are subject to change without notice.