Nvidia GeForce GTX 980 Kurzanleitung

Whitepaper

NVIDIA GeForce GTX 980

Featuring Maxwell, The Most Advanced GPU Ever Made.

V1.1

GeForce GTX 980 Whitepaper

Table of Contents

Introduction .................................................................................................................................................. 3

Extraordinary Gaming Performance for the Latest Displays .................................................................... 3

Incredible Energy Efficiency...................................................................................................................... 4

Dramatic Leap Forward In Lighting with VXGI .......................................................................................... 5

GM204 Hardware Architecture In-Depth ..................................................................................................... 6

Maxwell Streaming Multiprocessor.......................................................................................................... 8

PolyMorph Engine 3.0............................................................................................................................... 9

GM204 Memory Subsystem ...................................................................................................................10

New Display and Video Engines..............................................................................................................11

Maxwell: Enabling The Next Frontier in PC Graphics .................................................................................13

Hardware Acceleration for VXGI –Multi-Projection and Conservative Raster ......................................21

Tiled Resources .......................................................................................................................................23

Raster Ordered View............................................................................................................................... 24

DirectX 12................................................................................................................................................ 25

Advancing the State-Of-The-Art in Image Quality ......................................................................................27

Dynamic Super Resolution......................................................................................................................29

Conclusion...................................................................................................................................................31

LEGAL NOTICE .............................................................................................................................................32

GeForce GTX 980 Whitepaper

INTRODUCTION

Introduction

In our 21-year quest to bring the most realistic 3D graphics to gamers, NVIDIA has introduced a number

of innovations. With its hardware transform and lighting engine, NVIDIA’s GeForce 256 ushered in the

era of the GPU in 1999, bringing T&L support to consumer graphics for the first time. In late 2006 with

the G80 GPU and the GeForce 8800 GTX graphics board, CUDA and the formalization of GPGPU

computing as we know it today was first brought to the world. More recently, in 2010 we launched the

GeForce GTX 480—our first GPU built using the Fermi graphics architecture. GeForce GTX 480

incorporated fifteen parallel tessellation units, enabling dramatic speedups in geometry processing

compared to prior GPUs. With great strides in architectural efficiency, Kepler GPUs delivered

significantly improved performance and power efficiency when introduced in 2012.

Earlier this year we launched the GeForce GTX 750 Ti, our first GPU from the Maxwell architectural

family. Maxwell GPUs were designed from the ground up for extreme power efficiency and exceptional

performance per watt consumed. In many DX11 applications, the GTX 750 Ti is capable of matching or

even beating the performance of our once flagship GeForce GTX 480, while consuming only a fourth the

power. Thanks to its remarkable power efficiency, our first generation Maxwell GPUs were ideal for use

in power-limited environments like notebooks and small form factor PCs, in addition to mainstream

desktops.

NVIDIA’s latest GPU, GM204, is the first to use the full realization of our 10th generation GPU

architecture, Maxwell. Our design goals for GM204 were to deliver:

Extraordinary Gaming Performance for the Latest Displays

Incredible Energy Efficiency

Dramatic Leap Forward In Lighting With VXGI

Extraordinary Gaming Performance for the Latest Displays

After years of little or no progress in the evolution of PC displays, the pace of new technology

introductions has dramatically increased. Screen resolutions are growing, with the first 4K displays for

gamers hitting the market a little over a year ago. While 4K was initially expensive, thanks to rapidly

falling prices, 4K adoption is now beginning to take off. In order to drive all these pixels, GM2xx GPUs

were designed to deliver more performance at the highest resolutions, and also support new display

tech for 4K like HDMI 2.0. In addition, with Dynamic Super Resolution, Maxwell introduces the ability to

significantly enhance display quality on sub-4K displays by rendering at higher resolution and then

applying an advanced resizing filter that leverages the high resolution information.

Virtual Reality displays are another area that is likely to drive a dramatic increase in rendering rate

requirements. Two displays for two eyes doubles the rendering workload, and ever higher resolution

and refresh rate capability will be crucial to eliminating rendering artifacts that break the illusion of

reality for the VR user.

GeForce GTX 980 Whitepaper

INTRODUCTION

Incredible Energy Efficiency

All GeForce GPUs are made to deliver performance leadership in their respective segment. But PC

gamers also expect efficiency—not noisy fans, excessive heat, and over-taxed power supplies. The

desktop PC gaming market is seeing increasing popularity of smaller and thinner form factors, and the

laptop PC gaming market has grown explosively in the past few years.

The Maxwell architecture was designed to provide an extraordinary leap in power efficiency and deliver

unrivaled performance while simultaneously reducing power consumption from the previous

generation. With a combination of advances originally developed for Tegra K1, new architectural

approaches seen first in the GeForce GTX 750 Ti, and further refinement for GM204, Maxwell now

delivers 2x the performance per watt compared to Kepler products.

Figure 1: GM204 delivers 2X higher perf/watt compared to Kepler

GeForce GTX 980 Whitepaper

INTRODUCTION

Dramatic Leap Forward In Lighting with VXGI

Lighting is the biggest hurdle in computer graphics that must be overcome in order to achieve

photorealism. Global illumination gets us significantly closer to achieving this goal. Global illumination is

often used in the movie industry to produce CG scenes that look stunning, but rendering truly dynamic

global illumination in real time on the GPU has been impractical until now.

NVIDIA Voxel Global Illumination (VXGI) is the next big leap forward in lighting. VXGI is based on the

concept pioneered by NVIDIA researcher Cyril Crassin in 2011, of using a 3D data structure (“voxels”) to

capture coverage and lighting information at every point in the scene. This data structure can then be

traced during the final rendering stage to accurately determine the effect of light bouncing around in

the scene.

Cyril’s original implementation relied on voxels that were stored in an octree structure. While it was able

to run successfully on a GeForce GTX 680, it had limitations. We’ve spent the last three years developing

an implementation that can be accelerated natively by the GPU, as well as improving the algorithm. The

result of that effort is VXGI.

VXGI provides dramatically improved performance for global illumination. While VXGI’s software

algorithm will run on all GPUs, the performance benefits of VXGI hardware acceleration will only be

available on our latest Maxwell GPUs.

The first graphics cards to ship with our new GM204 GPU are the GeForce GTX 980 and GeForce GTX

970. For simplicity purposes we’ll be focusing solely on the GeForce GTX 980 in this document. With

2048 CUDA Cores and 4GB of GDDR5 memory, GM204 is the fastest GPU in the world, yet with a TDP of

just 165W, it’s also the most efficient.

GeForce GTX 980 Whitepaper

GM204 HARDWARE ARCHITECTURE

IN-DEPTH

GM204 Hardware Architecture In-Depth

Figure 2: GM204 Full-chip block diagram

Like Fermi and Kepler, GM204 is composed of an array of Graphics Processing Clusters (GPCs), Streaming

Multiprocessors (SMs), and memory controllers. GM204 consists of four GPCs, 16 Maxwell SMs (SMM),

and four memory controllers. GeForce GTX 980 uses the full complement of these architectural

components (if you are not well versed in these structures, we suggest you first read the Kepler and

Fermi whitepapers).

Another version of the chip, with 13 SMs, will ship concurrently and be called GeForce GTX 970. In the

future we plan to offer additional products based on GM204 that will ship with different combinations

of GPCs, SMs, and memory controllers to address various segments of the graphics market.

GeForce GTX 980 Whitepaper

GM204 HARDWARE ARCHITECTURE

IN-DEPTH

In GeForce GTX 980, each GPC ships with a dedicated raster engine and four SMMs. Each SMM has 128

CUDA cores, a PolyMorph Engine, and eight texture units. With 16 SMMs, the GeForce GTX 980 ships

with a total of 2048 CUDA cores and 128 texture units.

The GeForce GTX 980 features four 64-bit memory controllers (256-bit total). Tied to each memory

controller are 16 ROP units and 512KB of L2 cache. The full chip ships with a total of 64 ROPs and

2048KB of L2 cache (this compared to 32 ROPs and 512K L2 on GK104).

The following table provides a high-level comparison of Maxwell vs. our previous-generation GK104

GPU:

GPU

GeForce GTX 680 (Kepler)

GeForce GTX 980 (Maxwell)

SMs

CUDA Cores

1536

2048

Base Clock

1006 MHz

1126 MHz

GPU Boost Clock

1058 MHz

1216 MHz

GFLOPs

3090

46121

Texture Units

128

Texel fill-rate

128.8 Gigatexels/sec

144.1 Gigatexels/sec

Memory Clock

6000 MHz

7000 MHz

Memory Bandwidth

192 GB/sec

224 GB/sec

ROPs

L2 Cache Size

512KB

2048KB

TDP

195 Watts

165 Watts

Transistors

3.54 billion

5.2 billion

Die Size

294 mm²

398 mm²

Manufacturing Process

28-nm

The GeForce GTX 980 has double the SMs compared to the GK104 GPU used in the GeForce GTX 680

released two years ago. Because of the changes implemented in GTX 980’s new Maxwell SM, we were

able to integrate 2x more SMs without doubling the die size. With each SM also containing its own

dedicated PolyMorph Engine, GeForce GTX 980 also has twice the number of geometry units as its direct

predecessor. We’ll be discussing more details on the new SM design in the next section of the

whitepaper.

Based on efficiency and workload analysis, and math vs. texture processing requirements of modern

games, NVIDIA engineers determined that eight texture units per SMM is the best architectural balance

for Maxwell; therefore, the total number of texture units is the same as Kepler, 128. However, thanks to

GeForce GTX 980’s higher clocks, texture fill rate improves by 12% from one generation to the next. To

improve performance in high AA/high resolution gaming scenarios, we doubled the number of ROPs

The GFLOPS and texel fill rates in this chart are based on GPU Base Clock

GeForce GTX 980 Whitepaper

GM204 HARDWARE ARCHITECTURE

IN-DEPTH

from 32 to 64. Again, thanks to the added benefit of higher clocks, pixel fill-rate is actually more than

double that of GTX 680: 72 Gpixels/sec for GTX 980 versus 32.2 Gpixels/sec for GTX 680.

The memory subsystem has also been significantly revamped. GTX 980’s memory clock is over 15%

higher than GTX 680, and GM204’s cache is larger and more efficient than Kepler’s design, reducing the

number of memory requests that have to be made to DRAM. Improvements in our implementation of

memory compression provide a further benefit in reducing DRAM traffic—effectively amplifying the raw

DRAM bandwidth in the system.

Maxwell Streaming Multiprocessor

The SM is the heart of our GPUs. Almost

every operation flows through the SM at

some point in the rendering pipeline.

Maxwell GPUs feature a new SM that’s

been designed to provide dramatically

improved performance per watt than prior

GeForce GPUs.

Compared to GPUs based on our Kepler

architecture, Maxwell’s new SMM design

has been reconfigured to improve

efficiency. Each SMM contains four warp

schedulers, and each warp scheduler is

capable of dispatching two instructions per

warp every clock. Compared to Kepler’s

scheduling logic, we’ve integrated a

number of improvements in the scheduler

to further reduce redundant re-

computation of scheduling decisions,

improving energy efficiency. We’ve also

integrated a completely new datapath

organization. Whereas Kepler’s SM shipped

with 192 CUDA Cores—a non-power-of-two

organization—the Maxwell SMM is

partitioned into four distinct 32-CUDA core

processing blocks (128 CUDA cores total

per SM), each with its own dedicated

resources for scheduling and instruction

buffering. This new configuration in

Maxwell aligns with warp size, making it

easier to utilize efficiently and saving area

Figure 3: GM204 SMM Diagram (GM204 also features 4 DP units per

SMM, which are not depicted on this diagram)

GeForce GTX 980 Whitepaper

GM204 HARDWARE ARCHITECTURE

IN-DEPTH

and power that had to be spent to manage data transfer in the more complex datapath organization

used by Kepler.

Compared to Kepler, the SMM’s memory hierarchy has also changed. Rather than implementing a

combined shared memory/L1 cache block as in Kepler SMX, Maxwell SMM units feature a 96KB

dedicated shared memory, while the L1 caching function has been moved to be shared with the texture

caching function.

As a result of these changes, each Maxwell CUDA core is able to deliver roughly 1.4x more performance

per core compared to a Kepler CUDA core, and 2x the performance per watt. At the SM level, with 33%

fewer total cores per SM, but 1.4x performance per core, each Maxwell SMM can deliver total per-SM

performance similar to Kepler’s SMX, and the area savings from this more efficient architecture enabled

us to then double up the total SM count, compared to GK104.

PolyMorph Engine 3.0

Tessellation was one of DirectX 11’s key features and will play a bigger role in the future as the next

generation of games are designed to use more tessellation. With the addition of more SMs in GM204,

GTX 980 also benefits from 2x the Polymorph Engines, compared to GTX 680. As a result, performance

on geometry heavy workloads is roughly doubled, and due to architectural improvements within the PE,

can achieve up to 3x performance improvement with high tessellation expansion factors.

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

1357911 13 15 17 19 21 23 25 27 29 31

Performance

Expansion Factor

Microsoft SubD11 SDK Test

GTX 980

GTX 680

GeForce GTX 980 Whitepaper

GM204 HARDWARE ARCHITECTURE

IN-DEPTH

GM204 Memory Subsystem

In GM204 one ROP partition contains 16 ROP units (compared to eight ROP units per partition in Kepler);

each ROP can process a single color sample. With four ROP partitions, a full GM204 has 64 ROPs, twice

that of its predecessor, GK104, dramatically improving ROP throughput.

GM204 has a 256-bit memory interface with 7Gbps GDDR5 memory, the fastest in the industry. GM204

also features a unified 2048KB L2 cache that is shared across the GPU. In addition, GM204 has made

significant enhancements to our memory compression implementation.

To reduce DRAM bandwidth demands, NVIDIA GPUs make use of lossless compression techniques as

data is written out to memory. The bandwidth savings from this compression is realized a second time

when clients such as the Texture Unit later read the data. As illustrated in the preceding figure, our

compression engine has multiple layers of compression algorithms. Any block going out to memory will

first be examined to see if 4x2 pixel regions within the block are constant, in which case the data will be

compressed 8:1 (i.e., from 256B to 32B of data, for 32b color). If that fails, but 2x2 pixel regions are

constant, we will compress the data 4:1.

These modes are very effective for AA surfaces, but less so for 1xAA rendering. Therefore, starting in

Fermi we also implemented support for a “delta color compression” mode. In this mode, we calculate

the difference between each pixel in the block and its neighbor, and then try to pack these different

values together using the minimum number of bits. For example if pixel A’s red value is 253 (8 bits) and

pixel B’s red value is 250 (also 8 bits), the difference is 3, which can be represented in only 2 bits.

Andere Handbücher für GeForce GTX 980

Inhaltsverzeichnis

Andere Nvidia Grafikkarte Handbücher

Nvidia

Nvidia geforce4 MX420 Bedienungsanleitung

Nvidia

Nvidia QUADRO SYNC II Bedienungsanleitung

Nvidia

Nvidia Quadro K6000 Bedienungsanleitung

Nvidia

Nvidia Quadro RTX Bedienungsanleitung

Nvidia

Nvidia GeForce GTX 200 GPU Referenzhandbuch

Nvidia

Nvidia Quadro FX 4700 X2 Bedienungs- und Sicherheitshinweise

Nvidia

Nvidia QUADRO SYNC II Bedienungsanleitung

Nvidia

Nvidia Quadro FX 3700 Gebrauchsanweisung

Nvidia

Nvidia MS-8802 Bedienungsanleitung

Nvidia

Nvidia GeForce GTX 770 Bedienungsanleitung

Nvidia

Nvidia 4000 - e-GeForce MX 128MB DDR PCI Graphics... Bedienungsanleitung

Nvidia

Nvidia Titan RTX Bedienungsanleitung

Nvidia

Nvidia Quadro Bedienungsanleitung

Nvidia

Nvidia Video-228PCI Series Bedienungsanleitung

Nvidia

Nvidia FX4500 - Apple MAC Pro QUADRO FX 4500 Video Card... Bedienungsanleitung

Nvidia

Nvidia Quadro K6000 Bedienungsanleitung

Nvidia

Nvidia GeForce GTX 570 Bedienungsanleitung

Nvidia

Nvidia GeForce RTX 2070 SUPER Bedienungsanleitung

Nvidia

Nvidia GeForce GTX Handbuch

Nvidia

Nvidia GEFORCE GTX 550 Ti Bedienungsanleitung

Nvidia

Nvidia Quadro GV100 NVLink Bridge Bedienungsanleitung

Nvidia

Nvidia GeForce MX Series Bedienungsanleitung

Nvidia

Nvidia VCQFX4500X2 Handbuch

Nvidia

Nvidia Inno3D Bedienungsanleitung

Beliebte Grafikkarte Handbücher anderer Marken

Gainward

Gainward GeForce GTX 690 4GB Bedienungsanleitung

ATI Technologies

ATI Technologies RADEON 137-40225-20 Bedienungsanleitung

MSI

MSI N220GT series Bedienungsanleitung

ATI Technologies

ATI Technologies RADEON 7000 MAC EDITION Bedienungsanleitung

SIIG

SIIG aurora Bedienungsanleitung

ClearSpeed

ClearSpeed ClearSpeed Advance e720 Bedienungsanleitung

MSI

MSI R7850 series Bedienungsanleitung

DVS

DVS Atomix HDMI Bedienungsanleitung

Aztech

Aztech VGS3D-AGP Bedienungsanleitung

Advoli

Advoli TB4 Standard Bedienungsanleitung

ELSA

ELSA ERAZOR III LT Bedienungsanleitung

Gigabyte

Gigabyte GV-R7500L Bedienungsanleitung

Themis

Themis TGA-7000 PMC Bedienungsanleitung

EMS

EMS XtremeRGB-Ex1 Bedienungsanleitung

Pioneer

Pioneer PDA-5003 Bedienungsanleitung

QDI

QDI N2 Bedienungsanleitung

NEC

NEC N8116-48 Bedienungsanleitung

MSI

MSI R7770 PoweR Edition Bedienungsanleitung