Nvidia GeForce GTX 200 GPU Reference guide

Technical Brief
NVIDIA GeForce®GTX 200 GPU
Architectural Overview
Second-Generation Unified GPU
Architecture for Visual Computing

Table of Contents
Introduction ...................................................................................................................4
GeForce GTX 200 Architectural Design Goals and Key Capabilities...............................5
Architectural Design Goals.................................................................................................5
Gaming Beyond: Dynamic 3D Realism................................................................................6
Gaming Beyond: Extreme HD.........................................................................................7
Gaming Beyond: SLI .....................................................................................................7
Beyond Gaming: High-Performance Visual Computing and Professional Computation.............. 8
GeForce GTX 200 GPU Architecture ...............................................................................9
More Processor Cores....................................................................................................9
Graphics Processing Architecture.................................................................................. 10
Parallel Computing Architecture.................................................................................... 12
SIMT Architecture....................................................................................................... 13
Greater Number of Threads in Flight............................................................................. 13
Larger Register File..................................................................................................... 14
Improved Dual Issue................................................................................................... 15
Double Precision Support............................................................................................. 15
Improved Texturing Performance................................................................................. 15
Higher Shader to Texture Ratio.................................................................................... 16
ROP Improvements..................................................................................................... 16
1 GB Framebuffer ....................................................................................................... 16
Geometry Shading and Stream Out .............................................................................. 17
512-bit Memory Interface............................................................................................ 17
Power Management Enhancements.............................................................................. 18
Additional Pipeline and Architecture Enhancements........................................................ 18
Summary ......................................................................................................................20
Appendix A: Retrospective...........................................................................................21
Appendix B: Figure 1 References .................................................................................22
2May, 2008 | TB-04044-001_v01

Figures
Figure 1: Realistic warrior from NVIDIA “Medusa” demo..........................................................6
Figure 2: Far Cry 2 – Extreme HD Dynamic Beauty! (Ubisoft)...................................................7
Figure 3: Significant Speedup Using GPU................................................................................8
Figure 4: GeForce GTX 280 GPU Graphics Processing Architecture.......................................... 10
Figure 5: GeForce GTX 280 GPU Parallel Computing Architecture ........................................... 12
Figure 6: TPC (Thread Processing Cluster) ........................................................................... 13
Figure 7: Local Register File 2× versus 1× ........................................................................... 14
Figure 8: Geometry Shading Performance ............................................................................ 17
Tables
Table 1: Number of GPU Processing Cores .............................................................................9
Table 2: GeForce 8800 GTX vs GeForce GTX 280.................................................................. 11
Table 3: Maximum Number of Threads ................................................................................ 14
Table 4: Theoretical vs Measured Texture Filtering Rates....................................................... 16
May 2008 | TB-04044-001_v01 3

Introduction
In this technical brief we introduce NVIDIA’s new GeForce®GTX 200 GPU
family, the first GPUs to implement NVIDIA’s second-generation unified graphics
and computing architecture. The high-end, enthusiast-class GeForce GTX 280
GPU and performance-oriented GeForce GTX 260 GPU are the first members of
the GeForce GTX 200 GPU family and deliver the ultimate visual computing and
extreme high-definition (HD) gaming experience.
We’ll begin by describing architectural design goals and key features, and then dive
into the technical implementation of the GeForce GTX 200 GPUs. We assume you
have a basic understanding of first-generation NVIDIA unified GPU architecture,
including unified shader design, scalar processing cores, decoupled texture and math
units, and other architectural features. If you are not well versed in NVIDIA unified
GPU architecture, we suggest you first read the Technical Brief titled NVIDIA
GeForce 8800 GPU Architecture Overview. You can also refer to Appendix A for a
historical retrospective.
4May, 2008 | TB-04044-001_v01

GeForce GTX 200 Architectural Design Goals
and Key Capabilities
GeForce GTX 200 GPUs are massively multithreaded, many-core, visual computing
processors that incorporate both a second-generation unified graphics architecture
and an enhanced high-performance, parallel-computing architecture.
Two overarching themes drove GeForce GTX 200 architectural design and are
represented by two key phrases: “Beyond Gaming” and “Gaming Beyond.”
Beyond Gaming means the GPU has evolved beyond being used primarily for 3D
games and driving standard PC display capabilities. More and more, GPUs are
accelerating non-gaming, computationally-intensive applications for both
professionals and consumers.
Gaming Beyond means that the GeForce GTX 200 GPUs enable amazing new
gaming effects and dynamic realism, delivering much higher levels of scene and
character detail, more natural character motion, and very accurate and convincing
physics effects.
The GeForce GTX 200 GPUs are designed to be fully compliant with Microsoft
DirectX 10 and Open GL 2.1.
Architectural Design Goals
NVIDIA engineers specified the following design goals for the GeForce GTX 200
GPUs:
Design a processor with up to twice the performance of GeForce 8800
GTX
Rebalance the architecture for future games that use more complex
shaders and more memory
Improve architectural efficiency per watt and per square millimeter
Improve performance for DirectX 10 features such as geometry
shading and stream out
Provide significantly enhanced computation ability for high-
performance CUDA™applications and GPU physics
Deliver improved power management capability, including a substantial
reduction in idle power.
GeForce GTX 200 GPUs enable major new graphics and compute capabilities,
providing the most realistic 3D graphics effects ever rendered by GPUs to date,
while also providing nearly a teraflop of computational power.
May 2008 | TB-04044-001_v01 5

Gaming Beyond: Dynamic 3D Realism
While prior-generation GPUs could deliver real-time images that appeared true-to-
life in many cases, frame rates could drop to unplayable levels in complex scenes
with significant animation, numerous physical effects, and multiple characters. The
combination of the sheer shader processing power of GeForce GTX 200 GPUs and
NVIDIA’s new PhysX™technology facilitates many new high-end graphics effects
including:
Convincing facial and character animation
Multiple ultra-high polygon characters in complex environments
Advanced volumetric effects (smoke, fog, mist, etc.)
Fluid and cloth simulation
Fully simulated physical effects such as live debris, explosions, and
fires.
Physical weather effects such as accumulating snow and water, sand
storms, soaking, drying, dampening, overheating, and freezing
Better lighting for dramatic and spectacular effect, including ambient
occlusion, global illumination, soft shadows, color bleeding, indirect
lighting, and accurate reflections.
Figure 1: Realistic warrior from NVIDIA “Medusa” demo
6May, 2008 | TB-04044-001_v01

Gaming Beyond: Extreme HD
GeForce GTX 200 GPUs provide 50-100% more performance over prior-
generation GPUs, permitting increased frame rates and higher visual quality settings
at extreme resolutions, resulting in a truly cinematic gaming experience.
Figure 2: Far Cry 2 – Extreme HD Dynamic Beauty! (Ubisoft)
Support for the new DisplayPort interface allows resolutions beyond 2560 × 1600,
and 10-bit color support permits up to a billion different colors on screen (driver,
display, and application support is also required). Note that prior-generation GPUs
included internal 10-bit processing, but could only output 8-bit component colors
(RGB). GeForce GTX 200 GPUs permit both 10-bit internal processing and 10-bit
color output.
Gaming Beyond: SLI
NVIDIA’s SLI®technology is the industry’s leading multi-GPU technology, giving
you an easy, low-cost, high-impact performance upgrade. PC gaming simply doesn’t
get any faster or more realistic than running GeForce GTX 200 GPU-based boards
in SLI mode on the latest nForce®motherboards.
Two flavors of SLI are supported by the initial GeForce GTX 200 GPUs:
Standard SLI (two GPU boards), which typically boosts supported
game performance by 60-90% and permits higher quality settings
3-way SLI, which provides even higher frame rates and permits higher
quality settings for the ultimate experience in PC gaming when
connected to a high-end, high-resolution monitor.
GeForce GTX 200 GPUs process and display complex DirectX 10 and OpenGL
game environments with amazing graphics effects and high frame rates at extreme,
high-definition resolutions.
May 2008 | TB-04044-001_v01 7

Beyond Gaming: High-Performance Visual
Computing and Professional Computation
With the power of CUDA technology and the new CUDA runtime for Windows
Vista, intensive computational tasks can be offloaded from the CPU to the GPU.
GeForce GTX 200 GPUs can accelerate numerous rich-media and computationally-
intensive applications such as video and audio transcoding, or running distributed
computing applications like Folding@home in the background while surfing the
web. Examples of GPU-enabled applications include the RapidHD video
transcoding application from Elemental and various video and photo editing
applications.
Many engineering, scientific, medical, and financial areas demand high-performance
computational horsepower for numerous applications.
Figure 3 shows the amazing speedups that can be achieved by using a GPU instead
of a CPU in a number of professional visual computing applications, in addition to
mainstream video transcoding. Appendix B lists references and details for these
applications.
Figure 3: Significant Speedup Using GPU
With an understanding of the GeForce GTX 200 GPU design goals and key
objectives, let’s delve deeper into its internal architecture, looking at both the
graphics and parallel processing capabilities.
8May, 2008 | TB-04044-001_v01

GeForce GTX 200 GPU Architecture
GeForce GTX 200 GPUs are the first to implement NVIDIA’s second-generation
unified shader and compute architecture. The GeForce GTX 200 GPUs include
significantly enhanced features and deliver, on average, 1.5× the performance of
GeForce 8 or 9 Series GPUs.
Manufactured using TSMC’s 65 nm fabrication process, GeForce GTX 200 GPUs
include 1.4 billion transistors and are the largest, most powerful, and most complex
GPU ever made. All GTX 200 GPUs are built to operate comfortably within the
power and heat specifications of high-end PCs.
You may recall that the first-generation NVIDIA unified visual computing
architecture in GeForce 8 and 9 Series GPUs was based on a Scalable Processor
Array (SPA) framework. The second-generation architecture in GeForce GTX 200
GPUs is based on a reengineered, enhanced, and extended SPA architecture.
The SPA architecture consists of a number of TPCs, which stands for “Texture
Processing Clusters” in graphics processing mode, and “Thread Processing
Clusters” in parallel compute mode. Each TPC is in turn made up of a number of
streaming multiprocessors (SMs), and each SM contains eight processor cores (also
called streaming processors (SPs) or thread processors). Every SM also includes
texture filtering processors used in graphics processing, but also useful for various
filtering operations in compute mode, such as filtering images as they are zoomed in
and out.
More Processor Cores
The new second-generation SPA architecture in the GeForce GTX 280 improves
performance compared to the prior generation G80 and G92 designs on two levels.
First, it increases the number of SMs per TPC from two to three. Second, it
increases the maximum number of TPCs per chip from 8 to 10. The effect is
multiplicative, resulting in 240 processor cores.
Chip TPCs SMs per
TPC
SPs per SM Total SPs
GeForce 8 & 9
Series
8 2 8 128
GeForce GTX
200 GPUs
10 3 8 240
Table 1: Number of GPU Processing Cores
May 2008 | TB-04044-001_v01 9

Based on traditional processing core designs that can perform integer and floating-
point math, memory operations, and logic operations, each processing core is a
hardware-multithreaded processor with multiple pipeline stages that execute an
instruction for each thread every clock.
Various types of threads exist, including pixel, vertex, geometry, and compute. For
graphics processing, threads execute a shader program and many related threads
often simultaneously execute the same shader program for greater efficiency.
All GeForce GTX 200 GPUs include a substantial portion of die area dedicated to
processing, unlike CPUs where a majority of die area is dedicated to onboard cache
memory. Rough estimates show 20% of the transistors of a CPU are dedicated to
computation, compared to 80% of GPU transistors. GPU processing is centered on
computation and throughput, where CPUs focus heavily on reducing latency and
keeping their pipelines busy (high cache hit rates and efficient branch prediction).
Graphics Processing Architecture
As mentioned earlier, the GeForce GTX 200 GPUs include two different
architectural personalities—graphics and computing. Figure 4 represents the
GeForce 280 GTX in graphics mode. You can see the shader thread dispatch logic
at the top, in addition to setup and raster units. The ten TPCs each include three
SMs, and each SM has 24 processing cores for a total of 240 scalar processing cores.
ROP (raster operations processors) and memory interface units are located at the
bottom.
Figure 4: GeForce GTX 280 GPU Graphics Processing Architecture
10 May, 2008 | TB-04044-001_v01