web viewin other word, the more valuable ... pixel shader 2.0, and other opengl effect. ... based on...

23
Ma 1 Evolution of Graphic Card --- Development of Graphic Card since 2006 Jia Ming Simon Ma CMPS 161 Final Project

Upload: builiem

Post on 06-Feb-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 1

Evolution of Graphic Card ---Development of Graphic Card since

2006

Jia Ming Simon Ma

CMPS 161

Final Project

Page 2: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 2

Table of Contents

Abstract ---------------------- -------------------------------------------------------- 3 ~ 4

Introduction -------------------------------------------------------------------------- 4~6

How to Use------------------------------------------------------------------------------ 6

Benchmark Software --------------------------------------------------------------- 7~9

Discussion ----------------------------------------------------------------------------- 9~10

Initial Assumption --------------------------------------------------------------------- 10

Result --------------------------------------------------------------------------------- 11~14

Conclusion --------------------------------------------------------------------------- 15~16

Reference ---------------------------------------------------------------------------- 16~ 17

Page 3: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 3

Abstract

The graphic card, or precisely the GPU, was not popularized until 1999. Before

that, CPU was responsible to process most of the data. However as the digital graphics

spreads like wild fire, a processing unit specifically for graphic is necessary. This is

because the modern CPUs strongly favor lower latency of operations with clock cycles in

the nanoseconds. The trade-off area where our CPUs were not performing well was that

of massive fine grain parallelism [1]. In other words, CPUs focus on the throughput of the

pipelines, but not an individual operation. Therefore, CPU yielded rather inefficient

system for graphics processing any more.

On the other hand, GPUs are specifically designed to execute literally billions for

small user-written programs per second [1]. That means GPUs can handle the

complicated graphical data with a much faster speed comparing to CPU.

Unfortunately, it seems that CPU technology has been near the bottleneck now.

Due to the limitation of semi-conductor, people start building a CPU with more cores

instead of having a higher clock speed. As the result of this, the improvement is no

longer obvious. On the other side, the GPU technology hasn’t met the bottleneck yet, so

it still has a lot room to develop.

Because of this, GPU starts sharing the normal data processing with CPU. For

example, the CUDA technology from NVIDIA is a parallel computing architecture. It

instructs the computer to “co-process” on the CPU and GPU [2]. Based on this

technology, the processing speed is increased more than 100 times faster as figure 1

Page 4: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 4

shown [2]. So,

GPUs are not

only a graphical

processing unit,

but also an

important part of the central processing pipeline now.

The two major GPU producers in nowadays are NVIDIA and AMD (formerly ATI).

Nvidia marketed the “GeForce 256”, which is labeled as the “the world’s first GPU” in

1999, while its rival ATI Technologies announced the term VPU (visual processing unit)

with “Radeon 9700” in 2002. Since then, the endless war between NVIDIA and ATI (now

AMD graphic) has officially begun.

Introduction

This project is going to visualize the development of desktop graphic card since

2006. The program is going to analyze and represent the architecture of the GPU, the

price, and the performance in several graphs. Users are able to control what elements

to show on the graph by control panel.

Figure 2: nVidia, ATI, and AMD graphic logo

Figure 1: Processing time [2]

Page 5: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 5

There are three types of graphs. The

first one is regular X-Y Axis 2D scatter plot with

changeable Y axis as figure 3. The y axis could

be transistors count, memory clock rate,

fabrication process, fillrate, power

consumption (known as TDP), or release price.

The X axis is the timeline from 2006 to 2012.

The second type of graph is a

performance benchmark score chart. This

chart is a line graph for 2 brands, 6 categories of graphic

cards. The categories include entry-level graphic, mid-range graphic, and high-end

graphic for both AMD and nVidia.

The third type of graph is a

Performance/Price Ratio Chart based on time and

release price as figure 4 shown. The X axis is time

from 2006 to 2012, and Y axis is the release Price.

The size of each square represents the

performance/price ratio. In other word, the more

valuable the graphic card is, its square will be

bigger. Also, all graphic cards will be colored

based on their brand and targeting market.

Figure 3: scatter plot

Figure 4: P/P ratio chart

Page 6: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 6

This project is intended to help users to find the best selection of graphic cards.

More importantly, users are able to visualize the history, and predict the future of GPU

based on the trend.

How to Use

This program REQUIRES: FLTK library (v 1.3 tested), GLUT library, C++ compiler

(Visual C++ 2011 tested).

1) Press “Click here to start” on the menu bar, click “Select data” to open a file

browser. Select the data file which should be named as “data.txt”. After loading

the data, “Finish reading data” should be displayed in the window.

2) Press the corresponding button to display the chart you want. For the first chart

“scatter plot of specifications”, select what Y axis represents with the dropdown

menu, Press “Apply” button to display. For other two graph,

just Press the button and the chart will be displayed.

3) For all three charts, the color mapping is located on the top of each chart.

4) For price/performance ratio chart”, user can click on the window and the

information of those video cards in that region will be displayed on the screen.

Press the mouse again to restore the chart.

Benchmark Software

Figure 5

Page 7: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Figure 6: PassMark screenshot[4]

Ma 7

I used three different benchmark score in order to represent a all-around

performance for different rendering situation.

PassMark Video Benchmarks :

PassMark video card

benchmarks has been designed to

benchmark the how well your video

card performs when using the most

common features of DirectX. It renders

a number of scenes to the screen in

windowed or full screen mode. As such, Performance Test requires DirectX version 9 or

above. With an extra DirectX 10 test being available if you have Vista and a DirectX10

compatible video card[4]. This benchmark has three different tests, including simple text,

medium test, and complex test.

This program focus on Vertex Shader 2.0, Pixel Shader 2.0, and other OpenGL

effect. This is a representation of basic rendering perfromance.

3DMARK 06 :

3DMark06 is a benchmark from Futuremark.

Page 8: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 8

3DMark06 is a PC

benchmark suite designed

to test the DirectX9

performance of your

graphics card. A 3DMark

score is an overall measure

of your system's 3D gaming

capabilities, based on

comprehensive real-time 3D graphics and processor tests. By comparing your score with

those submitted by millions of other gamers you can see how your gaming rig performs,

making it easier to choose the most effective upgrades or finding other ways to optimize

your system[5].

This benchmark shows the ability to process game/software based on DirectX 9.0

as well as SM2.0/ SM3.0. This is a representation for last generation game performance.

3DMARK 11 :

A newer version of

3DMARK 06 which support

newest DirectX 11.

This software is

designed to measure your PC’s

gaming performance 3DMark 11

Figure 8 :3DMARK 11[6]

Figure 7: 3DMARK 06[5]

Page 9: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 9

makes extensive use of all the new features in DirectX 11 including tessellation, compute

shaders and multi-threading. Trusted by gamers worldwide to give accurate and unbiased

results, 3DMark 11 is the best way to consistently and reliably test DirectX 11 under game-like

loads[6].

This benchmark focus on the future compatibility and DirectX 11 rendering ability. It is

the representation of new game performance.

Discussion

My goal is to analyze the history of graphic card development, and predict the

future. So, the fabrication process is a great representation of "developing". GPU has a

very similar developing history with CPU, so it is normal to assume the GPU will meet

the bottleneck of developing due to the limitation of fabrication process.

When the fabrication was still high, people could try to integrate more

transistors into one chip in order to achieve better processing power. So they lower the

fabrication process, and put more and more transistors into one chip that has the same

size.

So the transistors number is another critical component for measuring and

estimating the power of a GPU. I assume the transistors count has been increasing since

several years ago, and it will still go up because of its direct relation with performance.

Page 10: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 10

On the other hand, memory clock affects the processing ability. Besides the GPU,

memory controls the output thread, so some companies start to boost up the memory

clock for faster processing speed. I can assume that the memory clock has been

increased a lot.

The price of video cards are always categorized into certain levels. So it is

suppose to be equally spread along time.

For the performance measurement, how could we get the overall performance

score with three benchmark score? I thing we shall focus on the "sustainability" of a

graphic. In other words, a graphic card that support newest technology shall have higher

score since it will still perform well in the future. Some high-end last generation graphic

card can still get high score due to its high core clock and memory clock. But its actual

game performance may be worse because they do not support the new technology

which most future games may require.

As the result of this, I give 3DMARK 11 score 50% weight. 3DMARK06 and

PassMark shares the other 50% of score. I think this score reflects the future of a video

card, but also show the ability to handle old game and software.

Initial Assumption

I assume the data set is already sorted into old released date to new released

date, and the first part has nVidia video card only while the second part has AMD only.

Page 11: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Figure 10:Memory Clock

Ma 11

Result

The first graph is the transistors

count chart(figure 9). Obviously, the

overall trend of transistors number

increment is proportional to time. Also,

we can see the difference between last

year's top model and this year's top

model is really huge. The main stream is

increasing with a rate of 500 to 600 transistors per year.

The memory clock chart shows

an unexpected result. The memory clock

of AMD's produces do not change a lot.

On the other side, nVidia increases their

clock rate every 2 years for a better

performance. It seems that AMD does

not focus on memory overclocking

comparing to nVidia. NVidia is increasing

their memory clock by around 100% every two years.

Figure 9 :Transistors

Page 12: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Figure 10:fab process

Figure 11: fillrate

Ma 12

This is the fabrication

process chart. As my expect, the

fabrication development has been

really fast from 2006 to 2008. It

takes only 2 year to step up from 90

nm to 55 nm. However, it remains

the same since 2009. This is caused

by the difficulty of solving physical

problem of semi-conductor. Also,

AMD is always leading this

improvement.

The fillrate is another

representation of performance. The

higher the fillrate is, the better the

performance is. From the chart, we

can see the fillrate has similar trend as

transistors,

with roughly 20 to 30 more

GigaTexture/s for one year.

Page 13: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Figure 12: TDP

Figure 13: Price

Figure 14: Performance

Ma 13

The power consumption shows how much energy is needed for each graphic

card. Normally, a faster video card requires more energy. We can still see the top

models have obvious upward trend, but the mainstream of power consumption does

not increase a lot due to the increment of transistors number. Because of the new

power management system and the more effective components, the power supply unit

of GPU need less power to power up same amount of transistors. Therefore, even

though the total amount of transistors increased exponentially, the net power

consumption remains at a reasonable level.

The price scatter plot is equally

spread into different price range as I

expect.

The Performance score line

chart show the trend of performance

increment. We can see the total

performance is increasing by 30% to

Page 14: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Figure 15: P/P

Ma 14

40% per year. Referring to the price chart above, the price of each group of video card is

relatively stable. This cause the new video cards are always better and more valuable

than the previous generation.

The last chart is

Performance/Price Ratio chart. It

also shows a obvious trend.

As I observe in the

performance chart, the new video

card are more valuable than old

video cards. Especially for low mid-range to high entry-level video cards from both

companies, they provide a great performance/price ratio for normal daily users. As we

can see, the big performance/price ratio products are concentrated at the bottom-right

corner.

Another fact is that high-end and mid-range video cards are more valuable in

long-term condition. For those video cards before 2008, the entry-level product are not

able to keep their performance because of the limitation of hardware even though they

had really high p/p ratio when they were released. Compare to this, the mid-range and

high-end video cards are able to handle the newer rendering duty due to its high clock

speed and better architecture even though they do not support the new APIs or

languages.

Page 15: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 15

Conclusion

The first problem, which group of video cards are most valuable for us?

NVidia cooperates with a lot of gaming companies, so their video cards are

optimized for many games; AMD is far behind. From the graph, nVidia mid-range video

cards' performance are really close to AMD's high-end. More importantly, its price is

much lower than any of the high-end video card. So a mid-range nVidia video card

(gtx560) is the best buy for normal gamers with limited budget. Of course, AMD's mid-

ranges are also great, but they are lack of pre-optimization by game companies. If you

really want to buy a computer and will not upgrade it for 4 to 5 years, you shall buy an

AMD high-end video card (hd6970 or hd6990) to maximize the long-term performance.

The second question is to predict the future of GPU industry.

I have read a article "End of Moore's Law". It stated that computers are going to

be as fast as it is physically possible for them to be. Of course, we can still make more,

but they will take more space, time, and power, and will cost money, which, though at a

minimum, cannot be reduced any further[7]. This is the problem, and it is already

reflected into the slowness of fabrication process development in recent years.

I believe computers will continue to get faster, for a while, after the hardware

stops getting faster. But the incentives and focus will shift. They starts to put more GPU

on one board for faster performance, and it works for now (the top video cards with

most fillrate have been dominated by dual-core video cards since 2008). But the power

consumption and heat problem can't be solved easily. They also tried to get new

Page 16: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 16

technology to boost the GPU to run faster, but most of them are still in the research

process.

So, I think the companies will shift their focus from hardware to software; At the

same time, trying to innovate some new technology or material to get rid of the

bottleneck of Moore's Law.

Reference

[1]. "High Scalability - High Scalability - GPU vs CPU Smackdown : The Rise of Throughput-

Oriented Architectures." High Scalability. Web. 24 Feb. 2012.

<http://highscalability.com/blog/2010/12/3/gpu-vs-cpu-smackdown-the-rise-of-

throughput-oriented-archite.html>.

[2]. Laszweski, Gregor Von. "Towards Flow Cytometry Data Clustering on Graphics

Processing Units." By Jeremy Espenshade, Doug Roberts, and James Cavenaugh.

Rochester: Rochester Institute of Technology. Web. 24 Feb. 2012.

[3]. "NVIDIA Developer Zone." What Is CUDA. Web. 24 Feb. 2012.

<http://developer.nvidia.com/what-cuda>

[4]. PassMark CO. "Video Card Test Information." PassMark Software. Web. 16 Mar.

2012. <http://www.videocardbenchmark.net/gpu_test_info.html>.

[5]. "Futuremark." - World's Most Popular Benchmarks and PC Performance Tests. Web. 17

Mar. 2012. <http://www.futuremark.com/benchmarks/3dmark06/introduction/>.

Page 17: Web viewIn other word, the more valuable ... Pixel Shader 2.0, and other OpenGL effect. ... based on comprehensive real-time 3D graphics and processor tests

Ma 17

[6]. "3DMark 11 Video Card Benchmark for Windows 7 PC." 3DMark11 for Windows 7

DirectX 11 Benchmark Tests. Web. 17 Mar. 2012.

<http://www.3dmark.com/3dmark11/>.

[7]. "End Of Moores Law." Cunningham & Cunningham, Inc. 27 Dec. 2011. Web. 17 Mar.

2012. <http://c2.com/cgi/wiki?EndOfMooresLaw>.