radeon rx 480 press deck

Senior Vice President And Chief ArchitectRaja Koduri

Under embargo until June 29, 2016 at 9 a.m. EST.

BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife

Imagine a world where only the top 16% of PC users purchase graphics cards that provide premium VR and gaming experiences…

Based on AMD internal data.2



Where millions of gamers have been relegated to using outdated technology…

3



Where notebook gamers are often forced to compromise…

4



Where tens of millions more can only read about incredible PC VR experiences they can’t enjoy for themselves…

5



And they’re worried that if they upgrade now, their investment will be short lived…

6



That’s the world we live in today.

7




Gamers and consumers are being left behind

8



Incredible gaming technology should be within the reach of everyone

9



How do we make the nextgeneration of gamers

PC gamers?

CHALLENGE 1

10



Single monitor gamers are gaming at 1080p or below according to Steam™

95%

*See Endnotes for details.

11



The overwhelming majority of gamers buy $100-300 graphics cards

84%+$300

$100-$300




5 million gaming notebooks are sold every year

Why isn’t it 100 million?




Exceptionally low power with low-z height to drive

notebook design wins

Engineered to deliver unprecedented

performance

Flagship technology that effectively gives GPU

performance that rivals that of consoles

Flagship technology that effectively gives mobile users GPU performance that rivals

that of consoles

Making the GPU accessible withclose-to-the-metal DirectX® 12 API

14



How do we bring real VR to millions of consumers?

CHALLENGE 2

15



0.0 B

0.2 B

0.4 B

0.6 B

0.8 B

1.0 B

1.2 B

1.4 B

1.6 B

1.43 billion PCs

VR is amazing, but relatively few can enjoy it

Just 13 million PCs worldwidethis year will have the graphics capabilities needed to run VR

Only 1% of the 1.43 billion PCs used globally

Source: Gartner: Forecast: Wearable Electronic Devices, Worldwide, Feb 2016.*See Endnotes for details.

16



17



Source: Seidmon Associates, Wave One / April, 2016, National Online Survey Among 1,000 Adults 18-60

One of the biggest deterrents to widespread adoption of VR is cost

26%

37%

41%

68%

0% 10% 20% 30% 40% 50% 60% 70%

Equipment too expensive

Headset too bulky

Situations wanting to use VR would be limited

Not enough diff. btwn VR and seeing things normally

VR User Adoption Survey

18



Opportunity to drive the install baseof VR-ready PCs to 100 million

in the next few years

19



Opportunity to drive VR into retail in volume

20



Making VR game and applicationdevelopment viable on a large scale

21



Reducing the cost of entry to VR:Dramatically accelerating the pace of the VR ecosystem growth to ultimately help make HMDs more affordable

22



Further Drives VR into retail

Grows the addressable market for PC VR

Makes VR possible for anyone who

wants it

Helps make HMDs more affordable

A high-performance, low-power solution to democratize VR, delivering extraordinary

VR experiences23



How can we get the best and widest range of content to PC

gamers as fast as possible?

CHALLENGE 3

24



How do we cut port times for console games to next to nothing and ensure the highest quality,

highest performance experiences?

25



We very intentionally selected these two chip sizes for the Polaris architecture because we believe gaming content will be built around these two

levels of capability.

5+ TFLOPS

2+ TFLOPSOR

26



The acceleration of moreaffordable content development

5+ TFLOPS

5+ TFLOPS

27



POLARIS 1036 CUs

> 5 TFLOPS256-bit

4th Gen GCN4K Encode/Decode

DP 1.3/1.4 HDR

Up to 2.8X Perf/Watt

POLARIS 1116 CUs

> 2 TFLOPS128-bit

4th Gen GCN4K Encode/Decode

DP 1.3/1.4 HDR

Up to 2.8X Perf/Watt


28



Introducing thePolaris architecture-based

Series,that delivers on the promise of gaming and VR for everyonewith three “entitlements” for all.

Extraordinary VR experiences at price points

never offered before

Console-class GPU performance for thin and

light notebooks

Great game content delivered to PC gamers in

real time.

29




Cool and Efficient SolutionFor eSports Gaming

TFLOPS > 2

CUs 14

Mem Size 2 GB GDDR5

Mem Bus Size 128-bit

Power connector none

Encode / Decode 4K H.264 / HEVC

Display Port 1.3/1.4 HDR

30



Brilliant HD Gaming

TFLOPS > 4CUs 32Memory Size 4GB GDDR5Memory Bus Size 256-bitPower connector 1x6-pinGCN 4th GenAMD FreeSync™ YESDisplay Port 1.3/1.4 HDR

31



Beyond HD GamingPremium VR Experiences

TFLOPS > 5CUs 36Memory Bandwidth 224/256* GB/sMemory Size 4/8 GB GDDR5Memory Bus Size 256-bitPower 150 WVR Premium YESAMD FreeSync™ YESDisplay Port 1.3/1.4 HDR

*AIBs may have different configurations in market 32


Premium HD GamingmGPU Scaling

with AMD CrossFire™ TechnologyDirectX™ 12 Gaming Featuring

Asynchronous ComputeSmooth Frame-rates

With AMD FreeSync™ Technology


1440p Premium Gaming Radeon RX 480 8GB VS NV GTX 970 4GB

1X

42%

21%

21%

25%

Percentage values are up to, see endnotes for details


33


Enabling Premium VR Experiences at Low Power

Radeon RX 480 8GB VS R9 380

Ready for Oculus™ Rift and

Steam® VR

Performance Test

3.66

6.3

Under embargo until June 29 2016 at 9 a.m. EST.


34



Polaris Platform for Gamers

Game-Changing ControlPremium Gaming Immersive VR

ExperiencesFuture-Proof

Technologies*

Under embargo until June 29, 2016 at 9 a.m. EST.*See endnotes for details 35


Gamers Deserve the Latest Game Technologies

+ 45%+ 63%

Premium Gaming

Single DX 12 mGPU OpenGL Vulkan

Under embargo until June 29, 2016 at 9 a.m. EST.See endnotes for details

Radeon™ R9 390X Radeon™ RX 480

36


Gamers Want Immersive VR Experiences

Radeon™ RX Graphics feature Asynchronous Shaders and QRQ

Immersive VR Experiences

With Asynchronous Time Warp

Judder reduction of 20-100x*

Frame 1

80

90

100

Frame 2 Frame 3 Frame 4

Without Asynchronous Time Warp

Frame 1

Missing ~5% frames*

80

90

100

Frame 2 Frame 3 Frame 4

MissedFrame

MissedFrame


*Results part of Oculus’s internal testing of ATW, graphics hardware not specified. Source: https://developer.oculus.com/blog/asynchronous-timewarp-on-oculus-rift/ Charts for illustrative purposes only.

Oculus Rift Asynchronos Timewarp (ATW)

37


https://developer.oculus.com/blog/asynchronous-timewarp-on-oculus-rift/

Gamers Deserve the Chance for Better Upgrades

• 1.3 HBR3 Ready• 1.4 HDR Ready

• Synchronous panel refresh• Elimination of video tearing

• 2.0b Enabled

AMD FREESYNC™ TECHNOLOGY


38



Gamers Deserve Great Gaming Experiences

Better Experiences

Enabling Developers Improving User Experience


39


Gamers Want Great Streaming

High Quality Game Streaming forGamers With Radeon™ RX Graphics

H.264

Open Broadcaster Software

HEVC

Game at Max Settings Record With Your Favorite Software

Encode With High Quality Codecs Share

Under embargo until June 29, 2016 at 9 a.m. EST. *See Endnotes for details.

40


Cool & Efficient E-Sports Gaming

The First in a Family of Solutions

Beyond HD Gaming& VR Premium Brilliant HD Gaming

41



EndnotesSlide 11: http://store.steampowered.com/hwsurvey

Slide 16: 13 million PCs worldwide next year will have the graphics capabilities needed to run VR according to Gartner as cited in http://www.bloomberg.com/news/articles/2015-12-30/few-computers-are-powerful-enough-to-support-virtual-reality

Slide 28: Polaris 11 2.8x performance per watt : Testing conducted by AMD Performance Labs as of May 10, 2016 on 3DMark 11 and 3DMark Firestrikeusing a test system comprising of an i7-4600M, 8GB, AMD Radeon driver 16.20. AMD Radeon R9 M280X (14CUs) scored 5700 and 3500 with a board power of 82W. AMD Radeon RX 480M (16CUs) scored 7200 and 4070 with a board power of 35W. Using Performance/Board power, the resulting average across the 2 different titles was a perf per watt of 2.8X vs the Radeon R9 M280X . RX-5

Polaris 10 2.8x performance per watt: Testing conducted by AMD Performance Labs as of May 10, 2016 on the AMD Radeon™ RX 470 (110w) and AMD Radeon™ R9 270X (180w),ona test system comprising i7 5960X @ 3.0 GHz 16GB memory, AMD Radeon Software driver 16.20 and Windows 10.. Using 3DMark Fire Strike preset 1080p the scores were 9090 and 5787 respectively. Using Ashes of the Singularity 1080P High, the scores were 46 fps and 28.1 fps respectively. Using Hitman 1080p High, the scores were 60 fps and 27.6 fps respectively. Using Overwatch 1080p Max settings, the scores were 121 fps and 76 fps respectively. Using Performance/Board power, the resulting average across the 4 different titles was a perf per watt of 2.8X vs the Radeon R9 270X. Test results are not average and may vary. RX-6

Slide 33: Testing conducted by AMD Performance Labs as of June 10, 2016 on the AMD Radeon™ RX 480 8GB and NV GTX 970 4GB, ona test system comprising i7 5960X @ 3.0 GHz 16GB DDR4-2666 memory, Gigabyte X99-UD4, AMD Radeon Software Crimson Edition driver 16.5, NV driver 365.10 and Windows 10. Using Ashes of the Singularity High 1440p 2xMSAA, Gears of War High 1440p 4xAF, Hitman Max 1440p 16xAF, Quantum Break 1440p Medium AA ON, Radeon ™ RX 480 scored 45.23 fps, 57.26 fps, 52.53 fps, 50.45 fps respectively. NV GTX 970 scored 37.47 fps, 47.19 fps, 36.85 fps, and 40.43 respectively. Performance may vary based on use of latest driver versions. Test results are not average and may vary. RX-16

Slide 34: Testing conducted by AMD Performance Labs as of May 19, 2016 on the Radeon™ RX 480 and Radeon™ R9 380 on a test systemcomprising Intel i7-6700K, 2x4GB DDR4-2666 RAM, Z170 motherboard, Radeon Software Crimson Edition 16.1.1 and Win10 64-bit. Using the Steam VR Performance test, Radeon RX 480 scored 6.3 and Radeon R9 380 scored 3.6. RX-2

Slide 35: Statement of “future-proof” refers to support of current and upcoming technology standards including 14nm FinFET process technology, DirectX 12 and VulkanAPI support, new display technology, and experiences such as VR. “Future-proof” statement is not meant to serve as a warranty or indicate that users will never have to upgrade their graphics technology again.Support of current and upcoming technology standards described above has the potential to reduce frequency of graphics upgrades for some users.

Slide 36: DX12 Testing conducted by AMD Performance Labs as of May 10, 2016 on the AMD Radeon™ R9 390X and Warhammer DX12 at1440p. Single card scored 60 fps, CrossFire™ configuration scored 97.8 fps. Test results are not average and may vary. Vulkantesting conducted by AMD Performance Labs as of May 10, 2016 on Doom at High Settings, tested at 1080P with an Intel Core i7 5960X with 16GB DDR4, Radeon™ RX 480 scored 55 fps in OpenGL and 80 fps in Vulkan. Test results are not average and may vary. RX-10

FreeSync is an AMD technology designed to eliminate stuttering and/or tearing in games and videos by locking a display’s refresh rate tothe framerate of the graphics card. Requires Monitor, AMD Radeon™ graphics and/or AMD A-Series APU compliant with DisplayPort™ Adaptive-Sync 1.2 (or newer). AMD Catalyst™ driver 15.2 Beta (or newer) required. Adaptive refresh rates varyby display; check with your monitor manufacturer for specific capabilities. Only select AMD Radeon GPUs and A-Series APUs supported; see www.amd.com/freesync for full details. GD-67

Slide 40: HEVC acceleration is subject to inclusion/installation of compatible HEVC players.

42



http://store.steampowered.com/hwsurvey

http://www.bloomberg.com/news/articles/2015-12-30/few-computers-are-powerful-enough-to-support-virtual-reality

http://www.amd.com/freesync%20for%20full%20details.%20GD-67

Disclaimers and Attributions

43

The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors.

The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing manufacturers, software changes, BIOS flashes, firmware upgrades, or the like. AMD assumes no obligation to update or otherwise correct or revise this information. However, AMD reserves the right to revise this information and to make changes from time to time to the content hereof without obligation of AMD to notify any person of such revisions or changes.

AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION.

AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

Ashes of the Singularity images and logos Copyright © 2016 Oxide Games. Ashes of the Singularity is a trademark of Stardock Entertainment. All rights reserved. The Division images and logos © 2015 Ubisoft Entertainment. All Rights Reserved. Tom Clancy’s, The Division logo, the Soldier Icon, Uplay, the Uplay logo, Ubi.com, Ubisoft, and the Ubisoft logo are trademarks of Ubisoft Entertainment in the US and/or other countries. Gears of War is a registered trademark of Microsoft Corporation in the US and other countries. Hitman images and logos ©2016 Io-Interactive A/S. All rights reserved. IO-INTERACTIVE and the IO logo are trademarks of Io-Interactive A/S. HITMAN and the HITMAN logo are trademarks of Square Enix Limited. SQUARE ENIX and the SQUARE ENIX logo are registered trademarks or trademarks of Square Enix Co. Holdings Ltd.Warhammer images and logos © Copyright Games Workshop Limited 2016. Warhammer, the Warhammer logo, GW, Games Workshop, The Game of Fantasy Battles, the twin-tailed comet logo, and all associated logos, illustrations, images, names, creatures, races, vehicles, locations, weapons, characters, and the distinctive likeness thereof, are either ® or TM, and/or © Games Workshop Limited, variably registered around the world, and used under license. Developed by Creative Assembly and published by SEGA. Creative Assembly, the Creative Assembly logo, Total War and the Total War logo are either registered trade marks or trade marks of The Creative Assembly Limited. SEGA and the SEGA logo are either registered trademarks or trademarks of SEGA Holdings Co., Ltd. or its affiliates. All rights reserved. SEGA is registered in the U.S. Patent and Trademark Office. All other trademarks, logos and copyrights are property of their respective owners.Battlefield 1 images and logos Copyright © 2016 EA. Battlefield 1 is a trademark of EA. All rights reserved. Far Cry®3 images and logos © 2016 Ubisoft Entertainment. All Rights Reserved. Far Cry, Ubisoft, and the Ubisoft logo are trademarks of Ubisoft Entertainment in the US and/or other countries. Based on Crytek’s original Far Cry directed by Cevat Yerli. Powered by Crytek’s technology “CryEngine”.Quantum Break images and logos are trademarks or registered trademarks of Microsoft Corporation.

ATTRIBUTION© 2016 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, CrossFire, FreeSync, Radeon and combinations thereof are trademarks of Advanced Micro Devices, Inc. in the United States and/or other jurisdictions. DirectX is a registered trademark of Microsoft Corporation in the US and other jurisdictions. HDMI, the HDMI logo and High-Definition Multimedia Interface are trademarks or registered trademarks of HDMI Licensing, LLC in the United States and other countries. Vulkan and the Vulkan logo are trademarks of Khronos Group Inc. Other names are for informational purposes only and may be trademarks of their respective owners.



Under embargo until June 29, 2016 at 9 a.m. EST.Under embargo until June 29, 2016 at 9 a.m. EST.

Radeon Software

Terry MakedonDirector of Software Strategy and UX



GPUOpen enables developers through open source

Laser focused on delivering the best GPU user experience




GPUOpenAMD’s initiative to enable developers through open source

14 major updates in the last 30 days*

41 developer blogs in last 4 months*

61 SDK/Samples/Libraries/Tools posted since launch – Jan 26 2016

We are the first IHV to release an SPIR-V extension - the intermediate shaderlanguage used in Vulkan (we released support for GCN instructions)

*Status as of May 28



Open source effects, SDKs and libraries

New EffectsShadowFX (DX12)

GeometryFX improvements (DX11)

TressFX 3.1 (DX11)

New Libraries / SDK / SamplesMultithreaded Rendering (DX12)

Out of order rasterization (Vulkan)

FireRays (Vulkan and OCL)

CrossFire™ API (DX11)



Open sourcing tools for developers

Compressonator for Textures

GPUPerfAPI Analysis

CodeXL 2.0Profile, debug and analyze performance/power across CPU and GPU graphics and computeEfficiencies - DX 12 CPU and GPU timelines

Enabling improved performance with more open source tools



Radeon Support for OpenVX™

OpenVX™ is an open standard for cross platform acceleration of computer vision applications & algorithms

Radeon support for OpenVX™ – Open source API support

RunVX – Efficient Runtime tool

Graph optimizer – Automatic optimization

Radeon support for OpenVX ™is a fully open source stack designed for speed and accuracy



So… what about virtual reality?



Virtual Reality Adoption

2014 2015 2016 FUTURE

Multiple HMD VendorsUp to 1080x1200/eye

<10 ms latency

90 Hz

First GenerationHeadsets

Oculus DK21K/eye

<20 ms latency

75 Hz

Generation ZeroHeadsets

Multiple HMD VendorsUp to 2K/eye

100Hz+

Second GenerationHeadsets

Oculus DK1640x800/eye

50+ms latency

PrototypeHeadsets



VR Application Categories

Google Street View

BASIC

Minecraft

ESSENTIAL

EVE: Valkyrie

PREMIUM

Unreal Engine 4 Paris Demo

ADVANCED



TrueAudio Next

Compute Unit Reservation

Quick Response Queue

Variable Rate Shading

DirectX®12 and Vulkan™

LiquidVR™ – New FeaturesBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


TrueAudio Next

Real Time GPU Audio Physics ProcessingRay Tracing + Convolution for many audio sourcesScalableLow latency

Asynchronous ComputeExecutes concurrently with other graphics & compute tasksUses queues with customized scheduling

Open source library will be available on GPUOpen.com

Optimized for superior VR experiences

Immersive audio in real time



Dedicate a number of CUs for a specific task queue

Immediate execution with a fixed resource budget

Works with any type of shader(graphics or compute)

Compute Unit Reservation

Graphics

Audio



Solving VR time-critical task executionFrames that take too long to render result in“judder” that destroys sense presence

Quick Response Queue (QRQ)More precise control of timingLeverages asynchronous computeLower latency

Asynchronous Time Warp on Oculus Riftuses QRQ to avoid dropped frames

Quick Response QueueBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Multi-viewport processing APIAssign viewport ID to each polygonIndependent resolution/quality settings for each viewport

Used to implement foveated renderingHigh resolution for frame centerReduced resolution for peripheryEye tracking support

Variable Rate ShadingBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


DirectX®12 is ideal for VR developmentIncreased draw call throughputLow CPU overheadNative support for asynchronous computeand multi-GPULow-level GPU hardware access

Supported by Oculus SDK andBasemark VR Score benchmark

LiquidVR® DirectX®12 samples and documentationwill be available at GPUOpen.com

LiquidVR™ and DirectX®12BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


GPUOpen enables developers through open source

Laser focused on delivering the best GPU user experience




ExperiencesStability



For many years we did monthly WHQL driver updates.

Users thought it was too much.



Then for a few years we toned down our release frequency.

Users thought it was too little.



Frequency

4 WHQL and 9 releases

2016

3 WHQL and 9 Beta Releases

2015

CURRENT PLAN

6 full WHQL certified recommended drivers a year as many optional or game specific drivers as needed

(target all of them to be WHQL)



Game-Day 0 supported with Radeon Software Crimson Edition

Crimson 16.1 Crimson 16.2 Crimson 16.2.1 Crimson 16.3 Crimson 16.4.1



Game-Day 0 supported with Radeon Software Crimson Edition

RSCE 16.5.2 RSCE 16.5.3RSCE 16.5.1 RSCE 16.5.3



User overall satisfaction

3.6

4.14.2

4.4 4.4 4.3 4.3 4.3 4.3

4.1

4.3

4.1 4.14.2 4.3

4.4 4.3 4.3 4.34.2

4.4

3.03.23.43.63.84.04.24.44.64.8

9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5

3 4 1 2 3 4 1 2

2014 2015 2016

September 2014 – May 2016Source: Internal AMD Data 4.4/5 Approval Rating



Radeon Software Beta ProgramLaunching today

Users can email for [email protected]

Fully integrated with our QA department and run by our lead beta tester of over 15 years



ExperiencesPerformance



UP TO UP TO UP TO

Radeon™ Fury X GPU: Vulkan™ vs OGL 25%1 103%2 27%3

Radeon™ R9 390 GPU: Vulkan™ vs OGL 4%⁴ 88%5 23%6

Radeon™ R9 380 GPU: Vulkan™ vs OGL 20%7 83%8 16%9

Paving the way with Vulkan™

Radeon Software 16.5.3

*Test results are not average and may vary



UP TO UP TO UP TO UP TO UP TO

Radeon™ Fury X vs. GTX 980Ti 21%11 23%12 18%13 2%14 0%15

Radeon™ R9 390 vs. GTX 970 57%16 81%17 40%18 19%19 13%20

Radeon™ R9 380 vs. GTX 960 9%21 70%22 39%23 12%24 38%25

Performance leadership in DirectX® 12 games

Radeon Software 16.5.2 vs. Nvidia 365.10

*Test results are not average and may vary



Smooth gaming experience on Radeon SoftwareRadeon Software offers smooth MGPU DX12 titles

Low variance in frame times results in a better gaming experience 10

0

10

20

30

40

50

60

1 15 29 43 57 71 85 99 113

127

141

155

169

183

197

211

225

239

253

267

281

295

309

323

337

351

365

379

393

407

421

435

449

463

477

491

505

519

533

547

561

575

589

603

Milli

seco

nds p

er fr

ame

Time



ExperiencesFeatures



MultiGPU

Scaling and high FPS don’t improve experience above playable frame rate

CrossFire™ API (DX11) is on GPUOpen



AFR frame pacingSupport for DX12 – planned support in a future Radeon Software driver update

Insert delay on the GPU before each present to the display

GPU1

GPU2

Frame 1

Frame 2

Frame 3

Frame 4

D P

D

D

P

P

D P

D = Delay P = Present

Consistent present times



Polaris support for Linux® distributions

All Open Driver (all open source components)Currently upstreaming

PRO Driver (Mix of open and closed components)Supported for DOTA 2 Vulkan on Day 1 of releaseBundled with SteamOS Beta 2.8 – Enabling Vulkan support



XConnect™ software features

Safe Unplug – available now!

Display through XConnect™ planned for a future release



Radeon Settings9 new Crimson features

Crossfire™ Global Control

HDMI™ Scaling Control

Color Temperature Control

Power Efficiency Global Control

Display Scaling Per-Application

2 Display Quick Eyefinity Support

Quick OS Launcher

UI Language Selector



X

Introducing Radeon WattMan*Take complete control of your GPU

Available at launch for Radeon™ RX400 Series GPUs

*formerly AMD OverDrive™



AMD Overdrive™Years of our users overclocking

Now we’re going to redefine that experience with Radeon WattMan



The new redefined overclocking experienceSimple GPU and Memory Control

Power Limit Control

New features

Dynamic GPU Frequency Curve

Voltage Control Curve

Acoustic Limit Control

Advanced Temperature Control

Advanced Fan Control

Histogram Interface

Crash Recovery Mechanism



New dynamic GPU frequency curveWith the new dynamic GPU Frequency curve, you can take direct control your

GPU clock, adjusting frequency and voltage per state.

7 customizable states on Polaris



New histogram interfaceRecord per-application data with only that application open. Easily track peak

and average activity, clocks, temperature and fan speed.

5 trackable data-points



Hialgo acquisitionLays groundwork for gaming innovations

Chill:Power savingEnforced FPS limit only when movement is low

Boost: Dynamic Image Quality ControlDelivers higher FPS during high action; improving smoothness and reduces lag



SummarySOFTWARE RELEASES

13 New releases for 2016, including 4 WHQL releases

15+ games supported and optimized (performance and quality)

FEATURES

New API and SDK Support Vulkan™ API, DirectX® 12 Quick Response Queue, Oculus Rift™ SDK

New Product Support Polaris 10, Polaris 11, Oculus Rift™, HTC Vive™, AMD XConnect™ technology,

AMD Radeon™ Pro Duo graphics

New Features Introduced for Radeon™ Settings Radeon WattMan, Global AMD Crossfire™ control, HDMI™ Scaling, Color

Temperature Controls, Per-Game Display Scaling, Language Menu, Two Display

Eyefinity, AMD Crossfire™ Status Indicator, Updated Social Links based on

Region, Power Efficiency Toggle

STABILITY

Over 140 fixes and customer end user issues resolved in the last 6 months

4.4 User Satisfaction currently

PERFORMANCE

Sample Crimson improvements comparing 15.11 to 16.5.3

27% 12% 26% 26% 35% 60%



Enabling the Polaris Experience


Features, technologies and process The Polaris Architecture



Best Graphics Core Yet State-of-the-ArtMultimedia IP

Future-ProofMonitor Support Cool & Quiet Gaming

Four Principles of the New Polaris Architecture




1st GenerationDirectX® 11

OpenGL 4.2

OpenCL 1.2

EQAA

Partially Resident Textures (PRT)

Hardware Tessellation

MLAA

Updates Over Time:DirectX® 11.1

DirectX® 12 FL 11_1OpenGL™ 4.5

MantleVulkan™

AMD LiquidVR™Asynchronous ComputeVirtual Super Resolution

2nd GenerationImproved compute task scheduling

(DirectX® 11)

Increased geometry throughput

AMD TrueAudio

OpenCL™ 1.2

OpenGL™ 4.3

HSA

Updates Over Time:AMD FreeSync™

DirectX® 11.2DirectX® 12 FL 12_0

OpenCL™ 2.0OpenGL™ 4.5

AMD MantleVulkan™

AMD LiquidVR™Asynchronous ComputeVirtual Super Resolution

3rd GenerationHigh Bandwidth Memory support

Improved tessellation performance

Delta Color Compression (DCC)

Hardware scheduler (HWS) for asynchronous compute

Native FP16/Int16 shader ops (APU)

Data parallel processing instructions

4th GenerationImproved geometry processing

AMD LiquidVR™ API for variable resolution rendering

DCC & updated memory controller

Shader instruction pre-fetch and enhanced buffering

Real-Time and prioritized asynccompute with spatial and temporal

scheduling

dGPU support for native FP16/Int16 Ops

AMD TrueAudio Next

H.265 Main10 decode acceleration

4K60 HEVC encode acceleration

HDMI® 2.0b

DisplayPort™ 1.3 HBR3

DisplayPort™ 1.4-HDR

HDR display supportUnder embargo until June 29, 2016 at 9 a.m. EST.



14nm FinFET process

1 Graphics Command Processor

4 ACE

2 HWS

36 Compute Units

4 Geometry Processors

32 Pixels Output/Clock

144 Texture Units

576 32b Load/Store Units

2 MB L2 Cache

256-bit GDDR5

AMD CrossFire™ technology

DisplayPort™ 1.4-HDR, HDMI® 2.0b

Video Encode/Decode acceleration

PCIe® 3.0

Radeon™ RX 480 at a glanceBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


14nm FinFET process


4 ACE

2 HWS

16 Compute Units



64 Texture Units


1 MB L2 Cache

128-bit GDDR5




PCIe® 3.0

Radeon™ RX 460 at a glanceBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


4th-Gen GraphicsA Philosophy of Efficiency



Primitive Discard Accelerator

Culls triangles early in the pipeline with zero area or

no inclusive sample points

Returns grow as MSAA increases

New Index cache for small instanced geometry

Reduces data movement

Frees internal bandwidth resources

Maintains or improves primitive throughput during

instancing

Enhanced geometry engines

0%

50%

100%

150%

200%

250%

300%

350%

400%

Rela

tive

Perf

orm

ance

Tessellation With 4xAA3

Prim Discard Off Prim Discard On



Instruction prefetch

Improves efficiency by reducing pipeline stalls

Makes instruction caching more efficient

Increased per wave instruction buffer size

Improve single threaded performance

Tuned L2 cache behavior

Client cache request grouping

Improved cache and memory access efficiency

Native FP16 and Int16 support

Reduced memory/register footprint and bandwidth

Lower power execution

Graphics, computer vision, data learning

Improved shader efficiency

Branch & Message Unit

Scalar Unit

Vector Units(4x SIMD-16)

Vector Registers(4x 64KB)

Texture Filter Units (4)

Local Data Share(64KB)

L1 Cache(16KB)

SchedulerTexture Fetch

Load / Store Units (16)

Scalar Registers(4KB)

Polaris Architecture Compute Unit



Improved shader efficiency

+15%PERFORMANCE PER CU

RADEON™ RX 480 VS. RADEON™ R9 2901

Instruction prefetchImproves efficiency by reducing pipeline stallsMakes instruction caching more efficient

Increased per wave instruction buffer size

Improve single threaded performance

Tuned L2 cache behavior Client cache request grouping Improved cache and memory access efficiency

Native FP16 and Int16 supportReduced memory/register footprint and bandwidthLower power executionGraphics, computer vision, data learning

UP TO



GPUOpen shader intrinsic functionsDirect access to inline ISA

Data parallel primitive accessDetect dynamically uniform conditions at runtime

Enables performance-critical functionsExecution exactly as developers intend

Available now: DirectX® 11, DirectX® 12 and Vulkan™

Console style programming enabled on PC to extract more performance from the GPU

Console to PC



Updated memory controller & PHYSupports up to 8Gbps GDDR5 memory devicesUp to 256GB/s memory bandwidth

Lossless DCC Full 2/4/8 : 1 compression ratios supportedDriver optimized for utilization of DCCProvides perf/bandwidth and power advantages Extends the life, viability and efficiency of GDDR5

Efficient balance of resourcesYields higher effective peak bandwidth

Memory & delta color compression engines

0%

20%

40%

60%

80%

100%

120%

140%

160%

Radeon™ R9 290X Radeon™ R9 Fury X Radeon™ RX 480

Rela

tive

Mem

ory

Band

wid

th

Effective Bandwidth Gain With DCC4

Uncompressible Data Compressible Data



320GB/s

256GB/s

0%

20%

40%

60%

80%

100%

120%

140%

160%

R9 290 Polaris 10

Energy Efficient Memory

Bandwidth Performance / bit

Large L2 cacheDoubled L2 cache

Reduces external memory bandwidth requirementsImproves power efficiencyIncreases benefits of DCC

Up to 40% power savings on memory transactions5



Other Graphics Architectures In High-Level APIs

Radeon™ Graphics in DirectX® 12, Vulkan™, and VR



* Diagrams for illustrative purposes only.

Shad

er U

tiliz

atio

n

Time

Compute Preemption of Graphics Quick Response Queue

Graphics Compute

Asynchronous Compute

Radeon™ asynchronous compute is incredibly flexibleBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife

ACE ACE HWS GRAPHICS COMMAND PROCESSOR HWS ACE ACE

4

Hardware scheduler (HWS) for asynchronous compute

CPU kernel mode driver scheduling task offload Real time and prioritized queues (audio/VR) Temporal and spatial resource management Concurrent task or process scheduling Coordination and load balancing with graphics Functionality can be updated with microcode



Polaris GFX Architecture Summary

+15%PERFORMANCE PER CU

RADEON™ RX 480 VS. RADEON™ R9 2902

2.8xPERFORMANCE PER WATT

RADEON™ RX 470 VS. RADEON™ R9 270X1

FinFET 14 Technology | Micro–Architecture Tuned for Gaming | Physical Design Optimizations | DPM Techniques

UP TO UP TO



Future-Proof Display Support



DisplayPort™ 1.3 HBR3 and DisplayPort™ 1.4-HDR

Polaris is ready for DisplayPort™ 1.3 HBR3Existing cables and connectors utilizedUp to 32.4Gbps of bandwidth+80% bandwidth over HDMI® 2.0b

Polaris is ready for DisplayPort™ 1.4-HDRUp to 10-bit 4K96CTA-861.3 HDR metadata transportRec.2020 color space supportSMPTE 2084 EOTF

0

5

10

15

20

25

30

35

DisplayPort™ 1.1 DisplayPort™ 1.2 DisplayPort™ 1.3Ba

ndw

idth

(Gbp

s)

Evolution of DisplayPort™ bandwidth across generations

Link Bandwidth



Polaris can enable 4:4:4 4K displays at up to 120Hz

We anticipate DRR-ready 4K120 panels by 4Q2016

DRR 4K120 panels will enable 4K AMD FreeSync™ monitors compatible with 30-120 FPS coverage

DisplayPort™ 1.3 is a major leap for 4K FreeSync™BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Radeon™ RX 400 Series graphics will enable single-cable 4:4:4 5K60 resolution

5K is 78% more pixels than 3840x2160 (4K)

210ppi of 5K/28” provides incredible sharpnessand working space vs 142ppi of 4k/28”

We anticipate displays ready for single-cable 5K by year end

DisplayPort™ 1.3 enables single cable 5K60BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


X

DisplayPort™ 1.3 is future-proof for gaming monitors

Up to 240Hz SDR

Up to 200Hz HDR

Up to 240Hz SDR

Up to 240Hz HDR

Up to 190Hz SDR

Up to 155Hz HDR

Up to 120Hz SDR

Up to 96Hz HDR

Up to 60Hz SDR



The Polaris architecture is ready for home theater PCs


4K60



X

AMD FreeSync™ Technology over HDMI® 2.0bSupported on 28 displays as of May 28th



Polaris delivers better pixels with HDR

Larger Color Spaces Higher Contrast Ratios Higher Peak Luminance



0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Human Visual Range sRGB Rec. 2020 Digital Cinema Typical WCG Monitor

HDR drives more colors

Typical monitors are sRGB (blue triangle)

Short-term HDR goal is to reach P3 DigitalCinema (yellow triangle)

Long-term HDR goal is to reach Rec. 2020(red triangle)

2X visible colors vs. sRGB75% coverage of the human visual spectrumRed/gold/cyan notably improved



HDR replicates the human vision system

Full R

ange

of H

uman

Eye

Human Dynamic Range

Color Contrast CapturePost

Production Mastering Distribution Display Color Contrast

What Viewer Sees SDR

0.05-100 cd/m2

Full R

ange

of H

uman

Eye

Human Dynamic Range

Color Contrast CapturePost

Production Mastering Distribution Display Color Contrast

What Viewer Sees HDR

0.0005-10,000 cd/m2



HDR drives higher luminance

Sunlight 1,600,000,000 nits

Fluorescent light 10,000 nits

Highlights 1,000-10,000 nits

White Range 250-1,000 nits

Typical Objects 1-250 nits

Shadow Details 0.01-1 nit

Ultra Blacks 0-0.01 nit

* Image courtesy “Russellstreet” under CC BY-SA 2.0 license from Flickr. Scene comparison simulated.

10,000 nits

~1 nit


https://www.flickr.com/photos/russellstreet/

https://www.flickr.com/photos/russellstreet/7412970228/in/photolist-ci4pMw-dSpdJc-byoNe5-dXqqrf-bMitJ8-dXjKL6-btcJy5-bMivGn-dXqqkJ-ci4tRq-ci4hH9-ci4m4S-ci4rLh-6on7fC-APovSD-CRmjr-7NNB5u-7NNASf-pogGG5-p6fGAT-tQ96dq-gpSiPK-oNKULo-qfi5Vj-oBFhi4-fquEpj-pYTui5-wbr3zZ-q9yjFj-p6V5sg-pibyAW-fgn24H-yz1a9h-o2ozYJ-Aar5M6-xMjogw-qQEiph-udU1Xf-fTfRyX-e61F97-ivF8fE-xNNJ7v-fTfLZM-wsVSzF-sAsckA-o2ptdM-rv5iuA-yEwDqG-qbMhEd-fTfQsD


HDR is better at translating bits to photons

Electro-Optical Transfer Functions are how we convert digital signals to light

Today’s consumer content is designed around CRT television EOTFs

Based on work started in 1930sOnly considers up to 100 nits of brightnessWastes encoding bits and crushes black levels

10-bit ST 2084 is for HDR Designed for the next 20 years of displaysConsistent encoding 0-10,000 nitsDramatically improves detail/efficiency in black levels

0.1%

1.0%

10.0%

0.001 0.01 0.1 1 10 100 1000 10000

Min

imum

Con

tras

t Ste

p

Brightness Nits

10-bit ST 2084 8-bit Gamma (100 Nit) 12-bit ST 2084 10-bit Gamma (1000 Nit) Human Vision



Polaris is ready for 10-bit and 12-bit HDRFirst-gen HDR displays will be 10-bit12-bit will exceed human visual acuity

Polaris architecture display pipeDisplay color processing engine

Programmable gamut remappingProgrammable gamma controlFloating point processingEnables 1:1 mapping with display’s native capabilitiesEliminates in-display remapping latency

Display ControllerFrame Buffer

32bpp Fixed Point

64bpp Floating Point

Floating Point Color Processing

Output Bit Depth Converter

8/10/12bpc color

8/10/12bpc colorDisplayPort™

HDMI®



Tonemapping should be done by the game engineTonemapping in the display adds significant input lag

We poll color/contrast/nits from the displayPolling fed back to engine tonemapperOutput from GPU optimized for the connected display

Switching to HDR tonemapping not a large hurdlegame engines already do HDRR tonemapped to SDR

Available for developers now:DirectX® 11 HDR driver support available for developersDirectX® 12 support planned in a future driver updateHDR Video playback support available for developers

Radeon™ Photon SDK for HDR gamingBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Polaris architecture HDR display support

HDMI® 2.0b(with HDCP 2.2)

1920x1080 @ 192Hz

2560x1440 @ 96Hz

3840x2160 @ 60Hz (4:2:2)

DisplayPort™ 1.4-HDR(with HDCP 2.2)

1920x1080 @ 240Hz

2560x1440 @ 192Hz

3840x2160 @ 96Hz



Accelerated Video Encoding & Decoding



A history of AMD’s video encode accelerator

Radeon™ HD 7970H.264 up to 1080p75

Radeon™ R9 290XH.264 up to 1080p75

Radeon™ R9 285H.264 up to 4K60

Radeon™ R9 Fury XH.264 up to 4K30



Video encode acceleration with Polaris

H.264 HEVC

1080p120 1080p240

1440p60 1440p120

4K30 4K60



Improving stream quality with Polaris

Fast HQ encode is a tough problemRapid scene changesNon-uniform complexity distribution

Achieving fast quality with 2-pass encodingDownscaled pre-encode guides rate controlPicture-level analysis for efficient budget allocationMacroblock-level analysis for efficient QP selection

2-pass encoding yields up to 1dB Y-PSNR gainWorks with H.264 and HEVC content

Warhammer, the Warhammer logo, GW, Games Workshop, The Game of Fantasy Battles, the twin-tailed comet logo, and all associated logos, illustrations, images, names, creatures, races, vehicles, locations,

weapons, characters, and the distinctive likeness thereof, are either ® or TM, and/or © Games Workshop Limited, variably registered around the world, and used under licence. Developed by Creative Assembly

and published by SEGA. Creative Assembly, the Creative Assembly logo, Total War and the Total War logo are either registered trade marks or trade marks of The Creative Assembly Limited.



2-Pass Encoding1-Pass Encoding



Leveraging the Polaris architecture encode accelerator for gamesBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


A history of AMD’s video decode accelerator

Radeon™ HD 7970H.264 up to 1080p60VC1 up to 1080p60

MP4-P2 up to 1080p60

Radeon™ R9 290XH.264 up to 1080p60VC1 up to 1080p60


Radeon™ R9 285H.264 up to 4K120VC1 up to 1080p60


Radeon™ R9 Fury XH.264 up to 4K120

Main HEVC up to 4K60VC1 up to 1080p60




Our most versatile decode accelerator yet

H.264Up to 4K120

HEVCUp to 4K60

Main-10

VP9Up to 4K

MP4-P2Up to 1080p60

VC1Up to 1080p60

MJPEGUp to 4K30



The Polaris architecture accelerates the best content

Beautiful HDR Movies(10-bit HEVC)

More Cat Videos(VP9)

Razor-Sharp Video Calling(HEVC & MJPEG)

Your Favorite Streamers(H.264)

* Logos for illustrative purposes only; no endorsement implied.



Optimizing FinFET 14



It’s been a long road

90nm2005

80nm2006

65nm2007

55nm2007

40nm2009

28nm2011

FinFET2016



Transistors per area have doubled effectively every 2 yearsSub-nm range, static leakage/area also doublingVoltage reductions from technology flattened

Many PD techniques can reduce leakage; mitigate poor scalingPower gatingVoltage islandsBack bias or advanced circuits

Unfortunately: a poor substitute for native process benefits

Drive to lower power consumptionBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


FinFET 14 performance vs. 28nm planar

FinFET offers product level performance advantage over 28A planar

Significant variation reduction and leakage power improvement

FinFETs have fundamentally lower variation than planar bulk

Performance

28nm FinFETLog

Leak

age



FinFET 14 has better variation

Leakage PowerPe

rfor

man

ce

Performance

FinFET 14

28nm

Lower variation improves product performance and power

Performance set by the slowest device, the required voltagePower set by the leakiest device

FinFETs have fundamentally lower variation than planar bulk

Lower power



FinFET performance/power benefit

Relative PowerRe

lativ

e Fm

ax

FinFET

28nm

Bottom lineFinFET devices provide fundamental performance / power benefit over planar



FinFETs are a very significant step Reduced operating voltagesReduced leakageFundamental improvement in power efficiency

FinFETs enable new product categoriesThin & light gaming notebooksSmall form factor desktopsDiscrete cards with fewer power connectors

Summarizing FinFET 14BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Design Approaches for Power Efficiency



All GPUs operate at low voltages with high currentChallenging to deliver quality voltage from packages and power supplies

Voltage variations can be 10%-15% of the nominal value

Average voltage must be raised to cover this variation, which wastes a lot of power

AMD’s adaptive clocking recovers that waste with up to a 25% power reduction

Adaptive clocking

Measured Droop Improvements on “Polaris 10”

0.8V 0.9V 0.9V 1.0V 1.0V 1.1V 1.1V 1.2V 1.2V 1.3V

Mea

sure

d Vo

ltage

Dro

op (V

set-

Vdev

)

Vset

CKS OFF CKS ON



Illustrating adaptive clocking

VDD DROPSBELOW THRESHOLD

RESPONSE TIME< 1 NANO-SECOND

TRADITIONAL MARGINED FREQUENCY

OPERATING FREQUENCYTEMPORARILY DECREASES

~ 5-10% VOLTAGE BENEFIT

VDD SUPPLY

VDD ABOVE THRESHOLD



Adaptive voltage & frequency scaling (AVFS)Reliably extract the true silicon speed capability of the GPU

Including: part-to-part processing, temperature and power deliveryHow: add a frequency sensor to existing power and temp sensors

Enables accurate setting of the optimal operating pointMaximum energy efficiency across the entire ASIC process, voltage, and temperature

Individual speed and voltage sensors distributed around the graphics core



Illustrating AVFS on the Polaris architecture

Voltage

Freq

uenc

y

Guard band required due to AC+DC voltage margins, temperature inversion and aging

True Si Fmax capability

Traditional DPM-states.

AVFS reduces the voltage required for the same

frequency

AVFS DPM-states.



Run voltage analysis code on a tester when the part is tested and binned

Log the voltage as seen by the integrated power supply monitors

Run the same code on PC boot and observe voltage

Dial board vregs to deliver the same voltage as observed on the tester

Eliminates waste power usually spent accommodating board/system variation

Boot time power supply calibration (BTC)

PSMs observe the minimum voltage on the tester

VVAVG_TST

VAVG_SYS

Same PSMs observe the voltage on each boot in the system

Differences are compensated out



GPUs require 2-3% clockspeed margin to accommodate transistor aging

Other aspects of the system also exhibit aging (e.g. lower voltage from system)

We want our parts to self-calibrate and adapt to changes over time – good or bad

More robust operation over timeOut-of-box performance is improved

Adaptive aging compensation

Exploit boot time calibration and aging-sensitive circuits in the AVFS modules

Vmin dialed in to compensate for

aging effectsV

VAVG_SYS



Specially-designed banks (MBFFs) that share circuitry

Critical clock sequence storage elements

There are approx. 21 million flip-flops in Polaris 10

They utilize about 15% of ASIC TDP

Save significant internal and sClk power4-5% TDP savings

Multi-bit flip-flop (MBFF)

Up to 40% total clock power savings

with quad MBFF



Reducing memory bandwidth dependence:Double L2 cacheImprove L2 utilization and efficiencyImprove LDCC algorithms and usage

Simultaneously reduce energy / bitFinFETs helpOptimized interface designPower gating, clock gating

Energy efficient memory interface

320GB/s

256GB/s

0%

20%

40%

60%

80%

100%

120%

140%

160%

R9 290 Polaris 10


Bandwidth Performance / bit



Reducing memory bandwidth dependence:Double L2 cacheImprove L2 utilization and efficiencyImprove LDCC algorithms and usage

Simultaneously reduce energy / bitFinFETs helpOptimized interface designPower gating, clock gating


0%

20%

40%

60%

80%

100%

120%

140%

160%

R9 290 Polaris 10


Energy / bit Performance / bit



320GB/s

256GB/s

0%

20%

40%

60%

80%

100%

120%

140%

160%

R9 290 Polaris 10


Energy / bit Memory power Bandwidth Performance / bit

Lower Energy per bitBetter performance per bit

Outcome:Memory bandwidth needs reducedTotal memory interface power reducedby up to 58%5


58%Power

Savings



14nm FinFET optimized by AMD

1.7xPerformance / Watt

With FinFET 146

2.8xPerformance / Watt

With AMD technologies2

UP TOUP TO

Baseline

Technology

Arch + Features

0%

50%

100%

150%

200%

250%

300%

Performance / Watt Gai ns

Baseline(28nm LPP)

FinFET 14 Δ

FinFET 14 + AMD

Technologies



24.5 x 24.529 x 29

Bonaire Polaris 11

L x W

Bonaire Polaris 11

1.9mm1.5mm

Z-Height

Radeon™ RX 460 graphicsBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


14nm FinFET process


4 ACE

2 HWS

36 Compute Units



144 Texture Units


2 MB L2 Cache

256-bit GDDR5




PCIe® 3.0

Radeon™ RX 480 graphicsBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Our best graphics core yet+15% generational performance per compute unit 1

Future-proof display supportReady for 10-bit and 12-bit HDR monitorsReady for next-gen high density and high refresh gaming monitors

State-of-the-art multimedia supportCPU-free streaming and recording of your favorite gamesReady for HDR streaming video servicesNew 2-pass video encoding

Cool and quiet gameplayFinFET 14 optimized by AMD technologiesUp to 2.8X generational perf/W uplift 2

Summarizing the Polaris Architecture

Polaris 10 GPU



1. Testing conducted by AMD Performance Labs as of May 10, 2016 on the AMD Radeon™ RX 470, AMD Radeon™ R9 270X. Using 3DMark FireStrike preset 1080p AMD Radeon™ RX 470 (110W) scored 9090, AMD Radeon™ R9 270X (180W) scored 5787. Using Ashes of the Singularity 1080P High, the AMD Radeon™ RX 470 (110W) scored 46 fps, AMD Radeon™ R9 270X (180W) scored 28.1. Using Hitman 1080p high at 1080p, the AMD Radeon™ RX 470 (110W) scored 60 fps, AMD Radeon™ R9 270X (180W) scored 27.6 fps. Using Overwatch 1080p Max settings, the AMD Radeon™ RX 470 (110W) scored 121 fps, AMD Radeon™ R9 270X (180W) scored 76 fps . Using Performance/Board power, the resulting average across the 4 different titles was a perf per watt of 2.8X vs the Radeon R9 270X . Test results are not average and may vary. RX-6

2. Testing conducted by AMD performance labs as of May 18, 2016 on the Radeon RX 480 and Radeon R9 290 on a test system comprisi ng Intel Core i7-5960X, 16GB DDR4-2666, Gigabyte X99-UD4, Windows 10 x64 (build 10586), Radeon Software Crimson Edition 16.5.2using Ashes of Singularity, GTA V, Project Cars, Witcher, and Assassin’s Creed Syndicate, All games tested at 1440p. Radeon RX 480 graphics (150W TGP/36 CU) vs. Radeon R9 290 graphics (275W TGP/40 CU) scores as follows: Ashes of the Singularity (44.19 FPS vs 46 FPS); GTA V (66.23 FPS vs. 66.44 FPS); Project Cars (48.99 FPS vs. 45.99 FPS); Witcher 3 (50.78 FPS vs. 50.13 FPS); Assassin’s Creed Syndicate (50.51 FPS vs. 45.78 FPS). Average FPS of above game scores: 52.14 (Radeon RX 480) vs. 50.06 (Radeon R9 290). Test results are not average and may vary. RX-3

3. Based on AMD internal small prim filter test. Primitive assembly rates with prim filter ON vs. OFF: 18 tri/px (3.947 vs. 1.255), 32 tri/px (3.901 vs. 1.773), 50 tri/px (3.760 vs. 1.402), 72 tri/px(3.303 vs. 1.187), 98 tri/px (3.928 vs. 1.171), 128 tri/px (3.870 vs. 3.639). System configuration: Radeon™ RX 480, Core i7-6700K, 16GB DDR4-2666, Windows 10 x64, Radeon™ Software 16.5.2.

4. Based on AMD internal memory bandwidth test. Radeon™ R9 290X: 263GB/s peak memory bandwidth. Radeon™ R9 Fury: 333 peak GB/s without DCC vs. 387 peak GB/s with DCC. Radeon™ RX 480: 186 peak GB/s without DCC vs. 251 peak GB/s with DCC. System configuration: Core i7-6700K, 16GB DDR4-2666, Windows 10 x64, Radeon™ Software 16.5.2.

5. Based on measurements of total memory interface power in watts conducted by the AMD performance labs as of 5/21/2016. System configuration: Radeon™ R9 290 vs. Radeon™ RX 480, Core i7-5960X, Gigabyte GA-X99-UD7, 16GB DDR4-2666, Windows 10 x64, Radeon™ Software 16.5.2.

6. Based on AMD internal data generated in AMD performance labs as of May 2016, measurements of capacitance, voltage frequency, leakage and power data show up to 1.7x performance/watt on 14nm vs 28nm FINFET technology. Final performance/watt results on AMD products using 14nm FinFET technology may vary and will depend on various factors including but not limited to clock speed, voltage, and various AMD proprietary technologies.

FootnotesBenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife BenchlifeBenchlifeBenchlifeBenchlifeBenchlifeBenchlife


Disclaimer & Attribution

The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors.

The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing manufacturers, software changes, BIOS flashes, firmware upgrades, or the like. AMD assumes no obligation to update or otherwise correct or revise this information. However, AMD reserves the right to revise this information and to make changes from time to time to the content hereof without obligation of AMD to notify any person of such revisions or changes.

AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION.

AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

ATTRIBUTION

©2016 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, [insert all other AMD trademarks used in the material here per AMD’s Checklist for Trademark Attribution] and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.

The terms HDMI and HDMI High-Definition Multimedia Interface, and the HDMI Logo are trademarks or registered trademarks of HDMI Licensing LLC in the United States and other countries.
