Original Link: http://www.anandtech.com/show/37

AGP Performance Explained

by Anand Lal Shimpi on September 28, 1997 4:30 PM EST


Over a week ago I received my very first LX based Pentium II board, the FIC KL-6011, unfortunately I was restricted to test the amazing new LX chipset with a Matrox Millennium II PCI card. A few days ago I received my second LX based Pentium II board, the Chaintech 6LTM, however this time it came fully packaged with an AGP card for testing. The ATI 3D Rage Pro, a pumped up version of the ATI 3D Xpression+ with full AGP support in 2x mode! Immediately I realized the problems AGP testing posed to me, however after hours of struggling and quiet hacking I managed to get the 3D Rage Pro to produce some reliable benchmarks, not only reliable, but amazing as well...in most cases that is...

Before testing the AGP card I had a few predictions in mind, mostly dealing with the performance I could expect from an AGP enabled system. I can tell you this now, you shouldn't expect any huge performance boosts, especially in tests like Winstone which depend on the graphics adapter to provide a certain level of performance and no more in order to receive the absolute best scores. The areas where I expected AGP to truly shine were mainly 3D Rendering and Games, and surprisingly enough, I was right =) There are some interesting quirks that I happened to run across while testing the ATI card for the first time and I will mention them in its full review on the Video Accelerator Comparison Guide.

Now lets get to those benchmarks, for the test system I used an Intel Pentium II MMX 266 at 300MHz on the Chaintech 6LTM with 64MB of SDRAM and a Quantum Atlas II UW SCSI 3 HDD. I took the Matrox Millennium II, the fastest PCI video card I have tested, and pitted it against the ATI 3D Rage Pro, in order to level the playing field both cards tested were 4MB cards and the exact same system was used in both tests. The latest drivers from each manufacturer was used in order to insure the best possible results, the test suite included Winbench 97, 3D Winbench 97, Chris Dial's 3D Bench and more...

Windows 95 Performance Comparison - PCI vs AGP - Business Winstone 97

What do we see here? A very small margin of a performance increase provided by the AGP card vs the PCI video card in real world business tasks, not a big plus to go out and buy an AGP card right?

Windows 95 Performance Comparison - PCI vs AGP - High End Winstone 97

This is a bit of a surprise, the outstanding Matrox Millennium II succeeds in beating the "superior" AGP card in the High End Winstone tests, looks like another win for PCI over AGP in this case...but the next few results should be a pleasant site to most AGP advocates.

DOS Gaming Performance Comparison - PCI vs AGP - Chris Dial's SVGA Bench

Here AGP begins to strut its stuff by beating the powerful PCI Millennium II in Chris Dial's SVGA Benchmark utility. Although the ATI 3D Rage Pro only received about 9fps more than the Millennium II we can predict by these results what a more tweaked and better designed AGP card can produce. In a few months we might even be able to dethrone the Voodoofx chipset as the supreme gaming platform for the Desktop PC, we'll just have to wait and see.

DOS Gaming Performance Comparison - PCI vs AGP - Chris Dial's VGA Bench

Yet another disappointment for the ATI "AGP Powerhouse" card, its VGA performance is outstanding compared to most PCI cards but still not enough to compete with the might of the PCI Millennium II. I should mention here that AGP was in no way intended for use exclusively with VGA games and programs requiring VGA acceleration, that is like giving a Porsche to a driver that will never take it above 20 mph. When you look at the 3D Winbench 97 Scores below you'll begin to see exactly why the ATI 3D Rage Pro is a better solution for AGP equipped systems than the all-mighty Matrox Millennium II.

2D Windows Performance Comparison - PCI vs AGP - Winbench97

PCI vs AGP - WinBench 97 Version 1.1 Graphics Winmark Results
Video Card ATI 3D Rage Pro (AGP) Matrox Millennium II (PCI)
Resolution Business Graphics Winmark 97 High End Graphics Winmark 97 Business Graphics Winmark 97 High End Graphics Winmark 97
800 x 600 x 8 bit Color (256 Colors) 157 No Result 162 No Result
800 x 600 x 16 bit Color (65K Colors) 156 No Result 146 No Result
800 x 600 x 24 bit Color (16.7M Colors) 142 No Result 126 No Result
800 x 600 x 32 bit Color (4B Colors) 132 No Result 116 No Result
1024 x 768 x 8 bit Color (256 Colors) 143 61.9 154 68.4
1024 x 768 x 16 bit Color (65K Colors) 134 53.7 118 57.0
1024 x 768 x 24 bit Color (16.7M Colors) Not Run Not Run 104 52.1
1024 x 768 x 32 bit Color (4B Colors) Not Run Not Run 71.4 47.6
1280 x 1024 x 8 bit Color (256 Colors) Not Run Not Run 154 68.0
1280 x 1024 x 16 bit Color (65K Colors) Not Run Not Run 107 55.8
1280 x 1024 x 24 bit Color (16.7M Colors) Not Run Not Run 78.6 48.9

Here are some interesting figures. At lower color depths the Matrox Millennium II (PCI) consistently beats the ATI 3D Rage Pro (AGP), however once you up the color depth past 256 colors to 65K or 16.7M the AGP card takes over. It is a very interesting occurrence, I never expected AGP to have this great of an effect on normal everyday programs requiring 2D Windows acceleration. Considering the ATI 3D Rage Pro is no where near the fastest AGP card out, I am wondering what the faster RIVA 128 based AGP cards will show us. I will try to have more benchmarks up later today, even some startling comparisons at 640 x 480!



3D Windows Performance Comparison - PCI vs AGP - 3D Winbench97

When it was originally introduced, AGP was pointed out as being a high end 3D solution for desktop PCs, and as you might be able to tell from the above scores, that does hold true. Completely shattering the competition, the ATI 3D Rage Pro based AGP card achieved a 3D Score over two times greater than the Matrox Millennium II which supposedly improved on the lack of 3D capabilities in the original Millennium. However, one must take into consideration that not all of the 3D features used by the 3D Winbench 97 Test Suite are supported by the Matrox Millennium II, in fact most of the features aren't supported and therefore caused the Millennium II to produce a few awful scores when compared to the ATI card, which supports all but one of the 3D features used. In order to clear up the confusion as to which specific 3D features were used, below is a list comparing the features supported and used by the ATI 3D Rage Pro to those of the Matrox Millennium II.

Supported 3D Features - PCI vs AGP
Feature ATI 3D Rage Pro (AGP) Matrox Millennium II (PCI)
3D Quality/Fog Vertex Capable Not Capable
3D Quality/Fog Table Not Capable Not Capable
3D Quality/Specular Highlights Capable Not Capable
3D Quality/Color Key Transparency Capable Not Capable
3D Quality/Alpha Transparency Capable Not Capable
3D Quality/Linear Capable Not Capable
3D Quality/Mipmap Linear Capable Not Capable
3D Quality/Dithering Capable Not Capable
3D Quality/Perspective Correction Capable Not Capable
3D Quality/Fog Vertex and Color Key Capable Not Capable
3D Quality/Fog Vertex and Alpha Capable Not Capable

Does this explain the difference in the performance of the ATI 3D Rage Pro when compared to the Matrox Millennium II under 3D Winbench 97? Simply put, yes! Although there are still quite a few more variables influencing the performance of the AGP card over the PCI card in 3D Windows situations, the supported 3D features list of the AGP card considerably outnumber those of the inferior PCI card. How does this effect the individual tests of 3D Winbench 97? Take a look at the below comparisons:

3D Winbench 97 - PCI vs AGP
Test ATI 3D Rage Pro (AGP) Matrox Millennium II (PCI)
3D WinMark/ 1/Stations1,N,2 56.9 27.9
3D WinMark/ 2/Race Track,N,6,T 28.7 22.5
3D WinMark/ 3/Chapel,N,6 8.97 8.85
3D WinMark/ 4/Stations2,L,2 34.5 7.76
3D WinMark/ 5/Stations3,L,2,S 25.7 6.61
3D WinMark/ 6/Islands,L,6,T,F 7.96 1.04
3D WinMark/ 7/Chapel,L,6 8.95 3.99
3D WinMark/ 8/Islands,NML,6,T 10.2 1
3D WinMark/ 9/Chapel,NML,6 8.89 0.746
3D WinMark/10/Stations4,NML,2 13.6 0.993
3D Triangle/640x480x8, Flat    
5 Pixel Triangle Setup 539 219
50 Pixel Triangle Setup 533 209
50 Pixel Triangle Fill 26.7 10.5
1000 Pixel Triangle Fill 76 197
3D Triangle/640x480x16, Flat    
5 Pixel Triangle Setup 538 220
50 Pixel Triangle Setup 539 209
50 Pixel Triangle Fill 26.9 10.4
1000 Pixel Triangle Fill 76 197
3D Triangle/640x480x16, Gouraud    
5 Pixel Triangle Setup 544 192
50 Pixel Triangle Setup 545 182
50 Pixel Triangle Fill 27.2 9.11
1000 Pixel Triangle Fill 76 49
3D Triangle/640x480x8, Z,Flat    
5 Pixel Triangle Setup 538 215
50 Pixel Triangle Setup 539 206
50 Pixel Triangle Fill 26.9 10.3
1000 Pixel Triangle Fill 76 195
3D Triangle/640x480x16, Z,Flat    
5 Pixel Triangle Setup 505 189
50 Pixel Triangle Setup 349 181
50 Pixel Triangle Fill 17.4 9.05
1000 Pixel Triangle Fill 45 26.6
3D Triangle/640x480x16, Z,Gouraud    
5 Pixel Triangle Setup 505 170
50 Pixel Triangle Setup 349 164
50 Pixel Triangle Fill 17.5 8.21
1000 Pixel Triangle Fill 45 26.5

In most cases the ATI card smoked the PCI Millennium II, however there are a few areas, particularly when dealing with Triangle Fills of 1000 pixels. The only explanation for this being that an excellent design cannot be beat even if a superior technology is used, now a comparison of a Matrox Millennium II AGP and a Matrox Millennium II PCI could prove to give this new technology a fair chance.

Log in

Don't have an account? Sign up now