NVIDIA GPU Update Safe harbor

Except for the historical information contained herein, certain matters set forth in todayyp’s presentation includin g, but not limited to, statements as to: the performance, benefits, capabilities, advantages of, and possible uses and applications for GeForce GTX 200 GPUs, PhysX by NVIDIA, NVIDIA SLI, CUDA, Tesla, NVIDIA technology, HybridPower; and PCs; purposes and uses of enterprise and visual ; the optimized PC; and parallel and graphics computing as well as other predictions and estimates are forward-looking statements within the meaning of the Private Securities Reform Act of 1995. These forward-looking statements and any other forward-looking statements that go beyond historical facts that are made today during the presentation or in response to questions are subject to risks and uncertainties that may cause actual results to differ materially. For a complete discussion of risk factors that could affect our present and future results , please refer to our Form 10-Q for the period ended April 27, 2008 and from time to time in the reports we file with the Securities and Exchange Commission. All forward- looking statements are made as of the date of this presentation. Except as requidblired by law, we assume no o blititbligation to up dtdate any suc htth statemen ts. Agenda

GPU Beyond

Gaming Beyond

Beyond Gaming Agenda

GPU Beyond

Gaming Beyond

Beyond Gaming Enterprise Computing

Productivity Creativity Today’s PCs Are More Visual Than Ever

UI (Windows Vista / Leopard)Directions (Google Earth)

Photos (Windows Photo Gallery/Picasa)Web Browsing Traditional Approach

See Appendix for system specs. Optimized PC Approach

See Appendix for system specs. Agenda

GPU Beyond

Gaming Beyond

Beyond Gaming Introducing GeForce® GTX 200 GPUs

1.4 billion transistors 933 GGigaFLOPsigaFLOPs of Architecture processing power Graphics Processing 240 processing cores Architecture GeForce GTX 200 GPUs: GiBGaming Beyon d

2nd Generation Unified AhittArchitecture

240 processing cores

Next gen geometry performance

Improved dual-issue

2x registers

Race Driver GRID, © 2007 The Codemasters Software Company Limited ( " Codemasters" ). All rights reserved .

Comparisons relative to GeForce 8800 Ultra GeForce GTX 200 GPUs: GiBGaming Beyon d

1GB / 512-bit

3x ROP blending performance

Advanced compression technology tuned for extreme resolutions

33--wayway NVIDIA SLI™

GPU support for future 10-10-bitbit color & 120Hz LCD panels

Comparisons relative to GeForce 8800 Ultra GeForce GTX 200 GPUs: GiBGaming Beyon d

Radeon HD 3870 X2 GeForce GTX 260 GeForce GTX 280

2.4

2.2

2.0

1.8

1.6

1.4

1.2

1.0

0.8 3DMark Crysis Oblivion Call of Duty Call of CoH ET: Quake FEAR UT3 World in Vantage 4 Juarez Wars Conflict

Vantage run at extreme settings. Games run at 25x16. R177 and Cat 2.3 drivers.

GeForce GTX 200 GPUs: GiBGaming Beyon d Cross-Platform Support Industry’s Most Advanced Engine

Rigid body dynamics Collision-detection Anti-tunneling, Joints, Springs and Motors Cloth / Metallic deformation Soft Bodies Fluids Over 140 Titles GeForce GTX 200 GPUs: GiBGaming Beyond

Heterogeneous Computing

Core 2 Quad GeForce (Quad Core) GTX 280 Cores 4 240

GFLOPS 96 933

Fluids 1 15x Soft 1 12x Bodies Cloth 1 13x

Source: NVIDIA estimates See appendix for system specs Agenda

GPU Beyond

Gaming Beyond

Beyond Gaming GPU Encoding

Currently under development. Demo only. Not final UI. GPU Encoding – Adob e P remi ere® Pro

HybridPower™ Advanced GPU Power Technologies

Dynamic Clock/ dGPU Voltage Scaling CPU Off Clock Gating mGPU dGPU renders off

70 60 atts

ter) 50 WW tt 40 30 20 (lower is be

Idle Power in Idle Power 10 0 GTX280 8800 Ultra 60% Less Idle Power Blu-ray Playback Power Mode GeForce GTX 260 Enthusiast Price / Performance

Processor Cores 192 Graphics Clock 576MHz Processor Clock 1242MHz Memory Clock 999MHz Memory 896MB GDDR3 Power Two 6-pin Connectors Length 10.5 inches Board Power 183W Thermal Dual Slot Fansink 2x DL-DVI + Outputs HDTV-out GeForce GTX 280 Ultimate Gaming Performance

Processor Cores 240 Graphics Clock 602MHz Processor Clock 1296MHz Memory Clock 1107MHz Memory 1GB GDDR3 Power 6-pin + 8-pin Connectors Length 10.5 inches Board Power 236W Thermal Dual Slot Fansink 2x DL-DVI + Outputs HDTV-out GeForce GTX 200 GPUs

Current New

GeForce GTX 280 1024MB GDDR3 $649 GeForce 9800 GX2 1024MB GDDR3 GeForce GTX 260 $399 896MB GDDR 3

GeForce 9800 GTX GeForce 9800 GTX 512MB GDDR3 512MB GDDR3

Prices are suggested retail prices Tl™Tesla™ Introducing the Tesla 10-Series Processor

1. 4 billion transistors 1 Teraflop of processing power 240 processing cores

…NVIDIA’s 2nd Generation CUDA™ Processor

Source: Performance calculated by NVIDIA Double the Performance Double the Memory Tesla T10

1 Teraflop 4 Gigabytes

500 Gigaflops 1.5 Gigabytes Tesla 10

Tesla 8

Tesla 8 Tesla 10 Double the Precision

Finance Science Design Numbers all relative to Tesla 8 Wide Developer Acceptance and Success

146X 36X 19X 17X 100X

Interactive Ionic placement for Transcoding HD Simulation in Astrophysics NN-- visualization of molecular video stream to Matlab using .mex body simulation volumetric white dynamics H.264 file CUDA function matter connectivity simulation on GPU

149X 47X 20X 24X 30X

Financial GLAME@lab: An Ultrasound Highly optimized Cmatch exact simulation of MM--scriptscript API for medical imaging object oriented string matching to LIBOR model with linear Algebra for cancer molecular find similar See swaptions operations on GPU diagnostics dynamics proteins and gene appendix sequences for details CUDA Runs on NVIDIA GPUs … Over 70 Million CUDA GPUs Deployed

GeForce® TeslaTM Quadro® Entertainment High-Performance Computing Design & Creation Application Software Industry Standard C Language

Numerics Engine cuFFT cuBLAS cuDPP

System CUDA Compiler CUDA Tools

1U PCIe Switch CFortran Debugger Profiler

4 cores How to Get to 100X?

Traditional Data Center Cluster

1000s of cores Quad-core 1000s of servers CPU

8 cores per server 2x Performance = 2x Number of Servers Linear Scaling with Multiple GPUs

Oil and Gas Computing: Reverse Time Migration Hand Optimized SSE Versus CUDA

200

160

120 FLOPS 80 GG

40

0 1 4 8 128 256 384

Number of Cores

X86 CPU NVIDIA GPU Source: See appendix Heterogeneous Computing Cluster

10,000s processors per cluster Hess NCSA / UIUC JFCOM SAIC University of Illinois University of North Carolina Max Plank Institute Rice University University of Maryland GGGusGus Eotvas University University of Wuppertal IPE/Chinese Academy of Sciences Cell phone manufacturers 1928 processors 1928 processors Tesla S1070 1U System

Processors 4 x Tesla T10

Number of cores 960

Core Clock 1.5 GHz

Performance 4 Teraflops

Total system memory 16.0 GB (4.0 GB per T10P)

Memory bandwidth 408 GB/sec peak (102 GB/sec per T10P) 2048-bit, 800MHz Memory I/O GDDR3 (512-bit per T10P) Form factor 1U (EIA 19” rack)

System I/O 2 PCIe x16 Gen2

Typical power 700 W Tesla C1060 Computing Processor

Processor 1 x Tesla T10 Number of cores 240

Core Clock 1.33 GHz

On-board memory 4.0 GB

Memory bandwidth 102 GB/sec peak

Memory I/O 512-bit, 800MHz GDDR3

Full ATX: 4.736” (H) x 10.5” Form factor (L) Dual slot wide System I/O PCIe x16 Gen2

Typical power 160 W More Than 250 Customers

Life MdilMedical Sciences Equipment Productivity Oil and gas EDA Finance

Evolved GE Healthcare OptiTex Hess Synopsys Symcor machines Siemens Tech-X TOTAL Nascentric Level 3 Smith-Waterman Techniscan LabView CGG/Veritas Gauda SciComp DNA sequencing Boston Scientific Weather Headwave CST Hanweck AutoDock Elemental Acceleware Agilent QuantCatalyst NAMD/VMD Technologies Seismic City Folding@Home Dimensional CMATCH string Imaging matching Manifold Digisens 70,000,000 GPUs

350,000 Driver Downloads Per Week

60,000 CUDA Downloads Folding@Home

Distributed comppguting to study yp protein folding Alzheimer’s Disease Huntington’s Disease Cancer Osteogensis imperfecta Parkinson’s Disease Antibiotics

CUDA client available soon! Molecular Dynamics : GROMACS

800 700 675 600 500

ns/day400 300 369 200

100 4100170 0

CPU PS3 RdRed GPUGPU TlTesla8 8 TlTesla10 10

Source: See appendix Life Sciences: Autodock for Cancer RhResearch National Cancer Institute reports 12x speedup

Wait for results reduced from 2 hours to 10 minutes

“We can only hope that in the long run, Silicon Informatics' efforts will accelerate the discov ery of new dru gs to treat a wide range of diseases, from cancer to Alzheimer's, HIV to malaria.”

Dr. Garrett Morris, Scripps, Author of AutoDock Science: National Center for Atmosp heri c R esearch

Weather Research and Forecast (WRF) model

4000+ registered users worldwide

20% speedup with 1% of WRF on CUDA

Saves 1 week analysis time

Source: NCAR EDA: Semiconductor Optics Correction bGby Gaud a

CPU

days

1000s CPUs

FPGA hours 10s GPUs

$100K $1 M $10 M

Cost

Source: Gauda Finance: Real-time Options Valuation

Hanweck Associates Volera real-time option valuation engine Value the entire U.S. listed options market in real-time using 3 S870s

GPUs CPUs Savings

Processors 12 600

Rack Space 6U 54U 9x Hardware $42,000 $262,000 6x Cost Annual Cost $140,000 $1,200,000 9x

Figures assume: • NVIDIA Tesla S870s with one 8-core host server per unit • CPUs are 8-core blade servers; 10 blades per 7U • $1,800/U/month rack and power charges • 5-year depreciation Desigggn: CAD Design For Apparel Cloth Physics 70M CUDA GPUs GPU CPU

Heterogeneous Computing

Oil & Gas Finance Medical Biophysics Numerics Audio Video Imaging Thank you Appendix I

Slide 7, 8 Configurator. Sources: • Benchmarks run on G33 based with 2GB DDR2 system memory using Windows Vista Ultimate. Intel driver 8.3.0.1013. NVIDIA graphics driver 169.05. • PC#1 = GeForce 8300 GS + Intel Core2 Duo E6550 • PC#2 = GeForce 8600 GT + Intel Core2 Duo E4500 • 3DMark Vantage run in Performance mode. PCMark Vantage run at 10x7 setting. • Company of Heroes run at 12x10 & 4x8x setting, Sims2 run at 12x10 & 4x8x setting, and Bioshock run at 12x10 & 4x8x setting. • The Adobe Photoshop image processing test uses deconvolution algorithms to deblur a 1024x1024 RGB color image. The Traditional PC is tested using a CPU to run an ‘Interactive Deconvolution’ filter available in the Fovea Pro 4.0 software developed by Reindeer Graphics. The Optimized PC is tested using a GPU to run a ‘Lucy-Richard Deconvolution’ algorithm available in a CUDA based Photoshop plug-in developed by NVIDIA. • The HD Video Encoding test measures the time required to transcode a MPEG-2 1280x720x30 @ 20Mb/s video file to the iTunes Apple TV format. The CPU is tested with iTunes to do the transcoding. The GPU is tested using the RapiHD Transcoder from Elemental Technologies.

• Relative PC performance is calculated as a mean of all successfully running benchmarks of Optimized PC as compared to Traditional PC. Appendix II

Slide 16. Benchmarks: 1. PhysX CPU estimates based on a system with a Core 2 Quad Q6700. 2. GPU estimates based on PhysX running on one GeForce 9800 GTX with the graphics rendering on a separate GeForce 9800 GTX 3. Both running Windows XP Pro

Slide 27 - CUDA app lica tions. Sources: 1. Interactive Visualization of Volumetric White Matter Connectivity in DT-MRI Using a Parallel-Hardware Hamilton-Jacobi Solver paper by Won-Ki Jeong, P. Thomas Fletcher, Ran Tao and Ross T. Whitaker 2. GPU Acce lera tion o f Mo lecu lar Mo de ling App lica tions paper. 3. Video encoding uses iTunes on the CPU, and Elemental on the GPU running under Windows XP. CPUs tested were Intel Core 2 Duo 1.66GHz and Intel Core 2 Quad Extreme 3GHz. GPUs tested were GeForce 8800M on the Gateway P-Series FX noteb ook , an d GeF orce 8800 GTS 512MB. CPUs an d Ge Force 8800 GTS 512 were run on P5K-V motherboard (Intel G33 based) with 2GB DDR2 system memory. Based on an extrapolation of 1 min 50 sec 1280x720 HD movie clip. http://developer.nvidia.com/object/matlab_cuda.html Appendix III

4. High performance direct gravitational nbody simulations on graphics processing units paper. CitdbEPJdHlCommunicated by E.P.J. van den Heuvel 5. LIBOR paper by Mike Giles and Su Xiaoke. 6. FLAG@lab: An M-script API for Linear Algebra Operations on Graphics Processors paper 7. http: //www.t ech ni scanme dica lsys tems.com / 8. General Purpose Molecular Dynamics Simulations Fully Implemented on Graphics Processing Units paper by Joshua A. Anderson, Chris D. Lorenz and A. Travesset 9. Fast Exact String Matching On the GPU presentation by Michael C. Schatz and Cole TllTrapnell

Slide 31 – Linear scaling with GPUs. Sources: •“Experiences with Seismic Algorithms on GPUs”, presented by Hess and NVIDIA at EAGE Conference.

Slide 38 – GROMACS. Sources: • CPU, PS3 pppyerformance provided by Pande Laborator y, Stanford Universit y. 3870x2 numbers from NVIDIA testing (FAH uses only one of the two GPUs present). Legal information

Performance tests and ratings are measured using specific systems and/or components and reflect the approximate performance of NVIDIA products as measured by those tests. Any difference in system hardware or software design or configuration may affect actual performance. Buyers should consult other sources of information to evaluate the performance of systems or components they are considering purchasing.

Copyright © 2008 NVIDIA Corporation. All rights reserved. NVIDIA, The NVIDIA logo, GEFORCE, HYBRIDPOWER, SLI, TESLA, CUDA, , PHYSX and PUREVIDEO are trademarks or registered trademarks of NVIDIA Corporation in the U.S. and/or other countries.

Other names and brands may be claimed as the property of others.