Attack Out Of The Blind Spot:
Matrox Parhelia-512

Introduction

Intro

It's always the same thing when a new graphics card is launched: prior to the launch, you get a test sample, and at the same time you have to sign an NDA (Non-Disclosure Agreement), which is supposed to ensure that you don't leak out the information received before a certain date. And it was the same story with the Matrox Parhelia card, which has been eagerly awaited by many readers. However, Matrox didn't send out the card until quite late - to be exact, Monday June 17 was when they sent out a test sample from Canada, via express mail, to our lab in Munich, Germany. Apparently, they were under the impression that this was plenty of time to review the product before the NDA was lifted today (June 25).

To make things even more exciting, UPS Germany received the card the next day on June 18, but they left it lying around at the airport in Cologne until June 20, and were unable to tell us why. And so, the test sample finally got to our lab on Friday, June 21. The initial tests got off to a good start, when our system choked on a VESA VGA bug in Parhelia, which appeared in connection with the DOS boot manager (XOSL) in our testing systems. More on that later.

Needless to say, all these events made for an extremely tight time limit, hindering extensive testing of the card. Therefore, we will be presenting the Parhelia test in two parts. In Part I, we bring you a detailed look at the 3D performance in comparison to an NVIDIA GeForce 4 Ti4600 card (which, at a street price of about $350, belongs to about the same price class) and the ATI Radeon 8500 128MB card (which is clearly less expensive, priced at approx. $200).

We will be bringing you a test of the Parhelia's 2D characteristics at a later time, when we've gained more practical experience with day-to-day use of the Parhelia in the multiple-screen mode. The theoretical capabilities have already been discussed in our May article, "Matrox Parhelia - The Challenger." If you haven't read this yet, you should take a quick look at it in order to get an overview of what the Parhelia can do.

Matrox Parhelia - The Card

For starters, Matrox will be offering the market two versions of the Parhelia. One of them will be a fully equipped retail version with a $399 price tag, as well as a bulk version for $349 (estimated). Both cards come with 128 MB, but with different clock speeds for the GPU and the memory. Later this year, 64 MB and 256 MB versions are expected to follow.

  • Bulk: 128MB DDR; GPU clock: 200 MHz; memory clock: 250 MHz (DDR); price: approx. $349
  • Retail: 128MB DDR; GPU clock: 220 MHz; memory clock: 275 MHz (DDR) ; price: approx. $399

The surprise here is the very low 220 MHz clock speed of the GPU, based on the 0.15-µm process, which Matrox claims is the price for the large number of transistors (80 million). After all, NVIDIA is able to clock its 4600 to 300 MHz and the memory to 650 MHz.

Matrox
Parhelia-512
NVIDIA
GF4 Ti 4600
ATI
Radeon 8500
Chip Technology 512-bit 256-bit 256-bit
Manufacturing Process 0.15 Micron 0.15 Micron 0.15 Micron
Transistor 80 Mio 63 Mio 60 Mio
Memory-Bus 256-bit DDR 128-bit DDR 128-bit DDR
Memory Bandwidth 18 GB/s 10,4 GB/s 8,8 GB/s
AGP Bus 1x/2x/4x/8x 1x/2x/4x 1x/2x/4x
Memory 128MB 128MB 128MB
GPU-Clock 220 MHz 300 MHz 250 MHz
Memory Clock 250 (500) MHz 375 (650) MHz 250 (500) MHz
BGA Memory 3,3ns 2,8ns 3,3ns
Vertex Shader 4 2 1
Pixel Pipelines 4 4 4
Texture Steps/Pipe 4 2 2
Pixel Shader Steps/Pipe 5 2 2
Textur Shader Steps 36 16 16
Vertex S. Version 2 1.1 1.1
Pixel S. Version 1.3 1.3 1.4
DirectX Generation 8.0 / 9.0 8 8.1
FSAA Modi Fragment /SuperSampling MultiSampling SuperSampling
Z-Date Compression - Yes Yes
Monitor Outputs 3 2 2
Chip Internel Ramdacs 2 2 2
Chip Externel Ramdacs 1 - -
Max. Dual Resolution 2048x1536 @ 32bpp 1600x1200 @ 32bpp 1600x1200 @ 32bpp
Max. Triple Resolution 3840x1024 @ 32bpp - -
Bits per Color Channels 10 8 8

Photos: The Board

Matrox Parhelia
The front of the card. The memory is arranged in pairs.

Matrox Parhelia
The back of the card.

Matrox Parhelia
A general view of the card. There are two DVI-I outs on the bracket.

Matrox Parhelia
A 3.3ns BGA memory module.

Matrox Parhelia
Matrox integrates two Chrontel CH7301A-T chips on the card, which are only responsible for DVI-out. The TV-out function on the chips is not used. Instead, Matrox integrates circuits on the Parhelia for this purpose.

Matrox Parhelia
The Parhelia chip has turned out to be quite large.

Matrox Parhelia
Adaptors for CRT monitors and TV-Out

The Driver

The new Matrox "Power Desk" Display Utility is the central element of the Matrox driver. This is what allows you to create all settings, from the monitor setup to the 3D options.

Matrox Driver
Matrox Power Desk Utility

3D Option

Happily, the 3D options are easy to set, even though it would have been nice to have a few more advanced settings at your fingertips.

Version

In the tests, the driver ran quite stably, except for two blue screens that weren't reproducible. Only in Aquanox could we find any image deficiencies, which, in this case, were encountered with the rendering of water plants. ATI's Radeon 8500 had also suffered from this problem for a long time. Another problem is the display of VESA graphics under DOS. Parhelia cannot display VESA graphics that are not 100% compatible, for example, in a high-resolution BIOS of a motherboard. Matrox will be offering an additional BIOS to remedy this problem.

Bildfehler Aquanox

A positive aspect is the signal quality of the Parhelia. In 1600 x 1200 @ 85 Hz, the card delivered a crystal clear image on our 22" test monitor. We'll be bringing you further details on this in Part II of the article, where we'll talk about all other characteristics of the Parhelia apart from 3D quality.

Matrox Demo Software

Matrox includes a few technical demos with the Parhelia, which demonstrate hardware displacement mapping and the performance of the 36 stage shader arrays. Here is a small selection:

Reef Demo

Reef Demo

Reef Demo

Reef Demo

This demo involves a very high polygon count. Four layers of high-resolution textures are used per fish.

Mars Demo

Mars Demo

This one demonstrates the use of hardware displacement mapping. The height of the terrain is controlled in real time using the "+/-" keys.

Surround Gaming

Surround Gaming

Parhelia supports up to three monitors at a maximum resolution of 3840 x 1024 @ 32bpp, and up to 2400 x 600 in games. This is the basis for Surround Gaming mode, in which games are divided up among the three monitors for a panoramic type of effect. However, this only works if the game allows variable settings for FOV (Field Of View). Here is a selection of current games that support this feature:

  • Flight Simulator 2002, by Microsoft(r) Corporation;
  • Quake III Arena(tm), by Id Software(tm)/Activision(r);
  • Return to Castle Wolfenstein, by Gray Matter and Id Software(tm) Inc./Activision(r);
  • Star Wars(r) Jedi Knight(r) II: Jedi Outcast(tm), by Raven Software/LucasArts(r) Entertainment Company LLC;
  • Unreal(r) Tournament 2003, by Digital Extremes and Epic Games, Inc./Atari(r);
  • Imperium Galactica 3: Genesis, by Mithis Games/Philos Laboratories/CDV Software Entertainment AG

Almost all of the games based on the Quake 3 Arena 3D engine can be played in Surround Gaming mode. Still, in order to actually use Surround Gaming, you need to push up your shirt sleeves and manually adjust the configuration files of the game as well as the registry. Here's how you do it, using Star Wars: Jedi Knight II as an example:

  1. Run game first time (required to create configuration file)
  2. Go to Jedi Knight 2 installed directory
  3. Go to \GameData\base subdirectory
  4. Edit file 'jk2config.cfg' (for single player gameplay)
  5. Find line 'seta r_customwidth,' set to "2400" (if not present, add 'seta r_customwidth "2400"')
  6. Find line 'seta r_customheight,' set to "600" (if not present, add 'seta r_customheight "600"')
  7. Find the line 'seta r_mode' set to "-1"
  8. Find the line 'seta r_mode,' set to "-1" Find line 'seta cg_fov,' set to "138" (if not present, add 'seta cg_fov "138"')
  9. Edit registry
  10. Find line \\HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\MTXPARH\Device0
  11. Add binary key named 'OGL.TripleHead' - set to '01 00 00 00'
  12. Run single player game (HUD should now be only displaying on central screen).
    Note: Do not change resolution in-game, or these changes will have to be repeated.

At first glance, when you consider that you need three monitors, Surround Gaming seems quite unpractical. On the other hand, the least expensive of the 17" monitors are no longer unaffordable, so some of you might actually consider making the investment. Fans of Flight Simulator should be particularly be tempted by what Surround Gaming offers.

3D Performance

A few more comments before we begin with the 3D benchmarks. Matrox does not see the Parhelia as being the GeForce 4 killer. Instead of going for maximum benchmark scores, the Canadian company focuses on excellent performance at maximum image quality, or, to be more exact, on the use of anisotropic filtering and anti-aliasing (FSAA). This is where the true strength of the Parhelia is supposed to be found. In order to analyze the Parhelia from this angle, we made a second run of the benchmark tests, in addition to the standard run. In the first run, we looked at its performance during standard operation, or with tri-linear filtering and without FSAA. The reason for this is that many users make their evaluations based on frame rate above all. And for $399, you can't blame the user for having high expectations.

In Part II, we'll be looking at the Parhelia's image quality and its performance at maximum quality, with a comparison to an NVIDIA GeForce 4 Ti4600. Here, we won't be including ATI's Radeon 8500, because its anisotropic filtering, which involves various driver tricks, is somewhat controversial and would make a comparison unnecessarily complicated.

Test Setup

Hardware
Processor Intel Pentium 4 2,2 GHz (100 MHz)
Memory 2 x 256 MB, PC 266, CL2
Graphic Cards Matrox Parhelia 512 128MB Retail
NVIDIA GeForce 4 Ti Ti 4600
ATI Radeon 8500 128MB
Mainboard MSI 845G Max
Drivers & Software
Graphics Driver NVIDIA - v29.42
ATI - v6.13.10.6094
Matrox - v1.00.25
DirectX Version 8.1
OS Windows XP Professional
Benchmarks & Settings
Quake III Arena Retail Version 1.17
Benchmark using 'Q3DEMO1'
Star Wars Retail Version v1.02
Jedi Knight II Benchmark using 'jk2ffa'
Max Payne Retail Version v1.05
Benchmark using 'Shooting Alex'
Giants Retail Version v1.04
Aquanox Retail Version v1.17
Dungeon Siege Retail Version v1.00
Comanche 4 Benchmark Demo V 1.0.0.1.18
3D Mark 2001 SE Pro Version, Build 330

Benchmarks

Giants

Giants

Giants

Giants

Giants

Here, the Parhelia clearly lags behind the Ti 4600 and Radeon 8500.

Max Payne

Max Payne

Max Payne

Max Payne

Max Payne

Again, the GeForce 4 manages to put a significant distance between itself and the others. Parhelia performs about the same as ATI's Radeon 8500.

Aquanox

Aquanox

Aquanox

Aquanox

Aquanox

In Aquanox, the situation is similar to the previous one. ATI and Matrox are at about the same level, while NVIDIA clearly leads.

Comanche 4

Comanche 4

Comanche 4

Comanche 4

Comanche 4

In Comanche 4, ATI's Radeon 8500 clearly falls behind the Parhelia.

Dungeon Siege

Dungeon Siege

Dungeon Siege

Dungeon Siege

Dungeon Siege

In this role-playing game, Parhelia is able to hold its own against the Radeon 8500. However, it is once again beaten by the GF4.

Star Wars: Jedi Knight II

Star Wars: Jedi Knight II

Star Wars: Jedi Knight II

Star Wars: Jedi Knight II

Star Wars: Jedi Knight II

JK2 uses the newest version of the Q3A 3D engine. Here, ATI and NVIDIA hold the trump cards.

Quake 3

Quake 3

Quake 3

Quake 3

Quake 3

Parhelia's performance in Quake 3 is a bit disappointing. However, in light of the high frame rate level, the performance results are perfectly fine.

3D Mark 2001 SE (330)

3D Mark 2001 SE (330)

3D Mark 2001 SE (330)

3D Mark 2001 SE (330)

3D Mark 2001 SE (330)

If you believe that the theoretical memory bandwidth of about 18 GB would give you a card that could reach new heights in 3D mark scores, you're gonna be disappointed. Parhelia clearly lags behind the Radeon 8500 and GF4 Ti.

Quality Comparison

The Parhelia is not being positioned on the market as a card for brute performance in standard mode, but rather for maximum display quality. This simply begs for a quality comparison with the competing product from NVIDIA.

The benchmarks show that the GeForce 4 Ti 4600 is absolutely superior to the Parhelia in standard mode. Here, "standard" means that only trilinear texture filtering is involved, without anti-aliasing (FSAA). In order to increase image quality, you can filter the texture anisotropically, which, in conjunction with FSAA, leads to significantly higher image quality.

There are two questions to answer here: which product provides better quality, and at what performance level. Let's tackle them one at a time. First, we'll look at the question of image quality.

Note: all screenshots were made at a resolution of 1024 x 768 at 32-bit color. In order to get an accurate impression of the image quality, you should set your desktop resolution to 1024 x 768 when viewing. All screenshots can also be downloaded uncompressed and in full resolution, so that you can judge for yourselves. Again, you should set your computer to 1024 x 768 when viewing them.

Download Screenshots (15 MB)

Anisotropische Filtering - Without Anti-Aliasing

It's difficult to compare various graphics cards with anisotropic filtering, because virtually every manufacturer has its own individual interpretation of this filter technology. Here and there, various tricks are used, and each manufacturer has its own nomenclature for the different filtering levels. Matrox only gives you two settings options: on or off. Officially, NVIDIA only supports anisotropic filtering with OpenGL. Although the driver has long since supported it, even under Direct 3D, NVIDIA still hasn't planned any optional settings for it in the driver menu. This leaves you with no choice but to modify the registry manually, or use freeware tools such as Rivatuner or nVHardPage.

Because quality is central in this comparison, we used the maximum possible settings for both cards:

comp-anisotropic
Click here for an uncompressed version of the screenshot.

Here, you can see a fragment that contains a ground texture. In the distance, the details become significantly less defined with simple trilinear filtering (left). Anisotropic filtering makes sharper, crisper textures possible. NVIDIA's GF4 was tested at a setting of 2x (which corresponds to the Parhelia's filtering level), as well as at the maximum setting of 8x (64 tap).

As you can see, the Parhelia's filtering level is quite low compared to what's possible with the GeForce 4. In the top half of the image, the GeForce 4 seems to be much sharper. Matrox calls its filter mode 64Sample Anisotropic, which, according to Matrox, corresponds to 16tap anisotropic filtering. Oddly enough, NVIDIA's interpretation at 64 samples corresponds to 32tap. However, a direct comparison of the images shows that Parhelia's anisotropic filtering corresponds to the Level 2 setting in NVIDIA's drivers.

Anisotropic Filering Performance

In the end, NVIDIA clearly takes the crown for best filtering quality. Here's an overview of how the candidates perform in this mode:

Anisotropic Q3

Anisotropic Q3

Anisotropic Q3

Despite the fact that the GeForce 4 offers better filtering, it still delivers higher frame rates with anisotropic filtering than the Parhelia does without filtering. Thus, NVIDIA can hold its ground against the Parhelia with regard to quality as well as performance.

Anisotropic Filering Performance, Continued

Anisotropic Max Payne

Anisotropic Max Payne

Anisotropic Max Payne

In Direct 3D as well, the Ti 4600 with 16tap anisotropic filtering (LV2) is faster than the Parhelia in standard mode.

Under Direct 3D, you get a different picture. Parhelia is a nose faster than NVIDIA's GeForce 4. When anisotropic filtering is activated at 1024 x 768, NVIDIA's top product loses approximately 46% at LV8 and 28% at LV2. By contrast, Parhelia loses only 10%. Unfortunately, the frame rate of the Matrox card is too low, so you can't really take advantage of this.

Conclusion: GeForce 4 gives you significantly better quality with anisotropic filtering than Parhelia. Not only that, but in OpenGL it delivers better performance even despite its better display mode (64tap vs. 16tap of Parhelia). Tested at the same level of quality, the Ti 4600 is clearly superior.

FSAA Rendering Quality

To put it simply, the latest graphics cards have two separate 3D drivers: one for Direct 3D and one for OpenGL. Of course, this is also true for anti-aliasing. In the past, NVIDIA's graphics cards achieved very good frame rates under OpenGL, but broke down in Direct 3D. Ever since the newest generation of drivers and the introduction of the GF Ti models, this has been remedied. This example shows you why Direct 3D and OpenGL must be analyzed separately.

Matrox's new Fragment Anti-Aliasing technology (16x FAA) (http://www6.tomshardware.com/graphic/02q2/020514/parhelia-05.html) introduces a whole new factor, because, unfortunately, this mode doesn't work with all games. Under certain conditions, with the use of stencil buffers, for example, this mode is unable to find the edges of the objects. In this case, Parhelia must make use of the very slow 4x SuperSampling mode.

The following gives you a quality comparison in Max Payne:

FSAA Rendering Quality
Click here for an uncompressed version of the screenshot.

Parheliaґs 16x FAA mode does not find all of the edges in this game. Good ol' "Max" is indeed freed of his rougher edges, but the objects surrounding him, such as the ticket scanner, for instance, are rendered without anti-aliasing. In SuperSampling mode, however, all objects are filtered. It's interesting to note that the back part of the train station becomes much more clearly rendered in SuperSampling mode.

And now, on to the GeForce 4, where we use the Quincunx and 4x FSAA modes:

FSAA Rendering Quality
Click here for an uncompressed version of the screenshot.

With NVIDIA's various FSAA modes, all areas of the image are always filtered, so you don't have the problem of edges getting left out of the calculation. The image in the middle shows the Quincunx mode, where NVIDIA tries to attain 4x quality at 2x performance through a few tricks. However, the anti-aliasing comes at the cost of significant blurring, which is quite noticeable in the background of the image, as well as the word "Entry." At 4x, these objects are much sharper.

In Quake 3, by contrast, Parhelia's 16x Fragment mode leaves no room for complaint. The following shows a quality comparison:

FSAA Rendering Quality
Click here for an uncompressed version of the screenshot.

The differences between NVIDIA's Quincunx and 4x modes are marginal. It's only when you look very closely at the image that you can perceive any differences, since the blurriness that is typical for Quincunx mode does not have a big effect on this portion of the image. The advantage of Parhelia's Fragment mode, however, becomes all too clear. The jagged edges look much smoother, even at the first fleeting glance.

The conclusion in the area of quality definitely points in favor of Parhelia. Still, Matrox has some explaining to do as to why 16x FAA does not function properly in Max Payne.

And now for the second part of the comparison: anti-aliasing performance.

Anti-aliasing Performance

Let's start out with Max Payne.

FSAA Performance Max Payne

FSAA Performance Max Payne

FSAA Performance Max Payne

In standard quality, Parhelia can't keep up with the GeForce 4 Ti 4600. In 16x mode, however, Parhelia runs slightly faster. Strangely enough, at 1600 x 1200, the GF4 cannot render in 4x mode, despite 128 MB memory. It's strange because at the same resolution with 4x in OpenGL, there's no problem with rendering. Parhelia's SuperSampling mode is out of the running here, because it is quite unusable in this game. It's hard to say how the Parhelia's performance will be affected by omitting various objects in the calculations.

Anti-aliasing Performance, Continued

FSAA Performance Quake3

FSAA Performance Quake3

FSAA Performance Quake3

In OpenGL, you get a very different picture. In the beginning, the GeForce 4, which is the better performer in standard mode, can still dish up some stiff competition for the Parhelia when in 4x mode. However, as the resolution increases, Parhelia moves ahead. In Quincunx mode at low resolutions, the performance of the GF4 is almost twice as high as that of the Parhelia. Again, the Parhelia's SuperSampling mode only delivers a very unsatisfactory result.

Conclusion: If Parhelia's 16x FAA mode would work properly, then it would be superior with regard to quality. NVIDIA's 4x mode is slower and has poorer results. By contrast, the GeForce 4 reveals superior performance in Quincunx mode, which provides a good compromise. If you're looking for optimal FSAA quality, then Parhelia is definitely the first choice. However, the Parhelia's very poor SuperSampling performance is disappointing, because considering the high memory bandwidth, it should have done much better.

Maximum Quality: Anisotropic + Anti-Aliasing

In the last section of the quality comparison, the focus is on maximum possible image quality, which means anti-aliasing plus antisotropic filtering.

Maximum Quality: Anisotropic + Anti-Aliasing
Click here for an uncompressed version of the screenshot.

Parhelia makes a very good impression here. Unfortunately, the ground texture is still a bit unclear, which is evident when compared to NVIDIA's GeForce 4.

Maximum Quality: Anisotropic + Anti-Aliasing
Click here for an uncompressed version of the screenshot.

The quality of the textures is much better with the GeForce 4. Surprisingly, there's very little difference between Quincunx and 4x.

Parhelia is better at anti-aliasing, while GeForce 4 offers the better texture filtering. We won't attempt to say which aspect holds more weight in an evaluation - this is best determined for yourself. Don't forget, however, that, in practice, you seldom take a close look at the individual parts of the scene when a game is running.

Anisotropic + Anti-Aliasing Performance

Max Quality Quake 3

Max Quality Quake 3

Max Quality Quake 3

If you're satisfied with the Quincunx mode, then Parhelia has no chance in terms of performance. Moreover, the disadvantage of unclear textures in Quincunx mode is compensated for by anisotropic filtering. If you run the GeForce 4 with 4x at the same level of quality (LV2), then this card is significantly better than Parhelia. Parhelia only gains a slight advantage starting with the higher resolutions.

Anisotropic + Anti-Aliasing Performance, Continued

Max Quality Max Payne

Max Quality Max Payne

Max Quality Max Payne

It's a similar picture with Max Payne. At higher resolutions, however, the GeForce 4 with 4x FSAA falls further behind.

It's difficult to draw a conclusion from the quality comparison, since this depends heavily on one's personal preferences. You could suggest that it's unfair to evaluate the GeForce 4 at the highest level of anisotropic quality, because the Parhelia filters much less. Still, we decided that maximum image quality offered by both cards was a central point. By contrast, if you analyze the cards at the same level of filtering, then the GF4 Ti is significantly better.

With regard to FSAA quality, Parhelia is clearly ahead of the GeForce 4 - if Fragment AA functions, that is. If it doesn't, then it looks pretty bad for Parhelia, because the SuperSampling is rendered useless by the lame performance.

The GeForce 4 presents a good compromise between quality and performance with Quincunx and anisotropic filtering. However, if you're looking at maximum quality, then we consider Parhelia to be the winner, despite the poorer anisotropic filtering, because the edges look much smoother and you get a better overall image than with the GeForce 4, thanks to Fragment AA. And yet, the advantage here is only slight.

So that you can judge for yourselves, we've provided you with the screenshots for both cards, which you can download in the original size and uncompressed. When viewing them, you should set the desktop resolution to 1024x768-32.

Download Screenshots (15 MB)

3D Mark 2001 SE Details

The benchmark 3D Mark 2001 SE (Build 330) provides much more in the results than merely a score. The two cards differ from one another in many ways, described in the following.

Fill Rate

The fill rate tells you how many texels per second a graphics card can transmit through the graphic card's memory. To this end, 3D Mark tests both single- and multi-texturing:

  • Single-Texturing: There are 64 surfaces with one texture each. This means that the graphics hardware fills each of these objects separately, no matter how many texture layers that card is capable of drawing in a single pass.

    Single-Texturing

  • Multi-Texturing: Here, we drew 64 texture layers as fast as possible. This means that we took advantage of the fact that modern cards are usually capable of drawing multiple texture layers on a single object as fast as it would draw one single layer. 64 texture layers are distributed so that each surface in use has as many texture layers as that particular card can draw in a single pass. For example, if your card can draw eight texture layers in a single pass, then there will be eight objects with eight texture layers each. If your card is capable of doing six texture layers in a pass, there will be ten objects with six layers, and an 11th layer with the remaining four layers.

    Fillrate Multitexturing

In single-texturing operation, the GeForce 4 Ti 4600 is clearly ahead. In multi-texturing operation, which is used by many of the latest games, the Parhelia moves ahead by a bit. Basically, the modest GPU clock speed of 220 MHz gets in the way of better results.

High Polygon Count

In this test, 3D Mark shows the throughput of triangles per second.

  • 1 light: For a high throughput, we have only diffuse materials and a single directional light in this run.
  • 8 lights: We have the same scene, but the dragons are of a shiny material giving a specular reflection. There are eight lights in all; one directional and seven point lights.

High Polygon count 1 + 8 Lights

In the simple test, the GF4 Ti can calculate almost double the amount of triangles as the Parhelia. In the more complicated test with eight lights, which involves additional reflections on the objects, both cards are at about the same level, with a slight advantage for the GF4. ATI's Radeon 8500 clearly falls behind.

Pixel Shader Performance

As with the GeForce 4, Parhelia supports pixel shaders based on version 1.3. 3D Mark 2001 SE provides a series of tests to evaluate the speed of the pixel shaders.

Game 4 - Nature

This part of the benchmarks is used by the pixel shader for the rendering of water:

This scene depends heavily on DirectX(r) 8 features. The trees, grass and butterflies are animated, transformed and lit using vertex shaders. The movements of the fisherman are made by morphing, also using a vertex shader. The surface of the lake uses vertex shaders, pixel shaders and cube maps. Due to the pixel shader use, this test cannot be run without DX8 compatible graphics hardware. The scene only has a high detail run and affects the 3DMark score accordingly.

Game 4 - Nature

Parhelia only achieves a disappointing 25 fps in this test, clearly beaten by the GF4 and the R8500.

Pixel Shader Speed

In this special test, 3D Mark measures the performance of the pixel shaders. In the standard test, pixel shaders based on version 1.3 are used. The Advanced Pixel Shader Text, however, tests version 1.4, which was introduced by ATI. Therefore, ATI cards should be faster than GF4 Ti and Parhelia in this test.

  • Pixel Shader
    Pixel shader is a technique for making custom per pixel lighting and other per pixel operations in a 3D scene. Bump mapping is a kind of per pixel operation, because it makes a surface look bumpy due to modifications of the pixel colors, without touching the scene geometry, or, in other words, the polygons of the scene. This test demonstrates a water surface drawn with a pixel shader doing environment bump mapping using a cube map. This test requires DirectX 8 compatible hardware with Pixel Shader 1.0 support. The graphics card needs to be capable of four texture layers in a single pass to produce this water surface in a single pass.
  • Advanced Pixel Shader
    This is a new test included in 3DMark2001 SE, and it uses Pixel Shader version 1.4, introduced in DirectX 8.1. The same effect can also be achieved using Pixel Shader 1.0, but then rendering the water surface requires two passes. Graphics hardware that supports Pixel Shader 1.4 (or higher) renders the water in a single pass. The water surface in this test is more realistic than in the test above. Instead of just an environment bump map reflecting a cube map, this test uses a ripple texture (traditional environment bump map), a reflection texture, a refraction texture and a Fresnel texture. The resulting effect is a water surface that has ripples from the wind, for example, shows a rippled reflection and a rippled refraction (view of objects beneath the water surface), while also adjusting the reflection intensity according to the camera's angle to the water surface. For example viewing the water's surface from a perpendicular angle will show only refracted light rays, or what's under the water surface, while looking almost parallel to the water's surface will show the reflection only.

Pixel Shader

Pixel Shader

Again, the Parhelia clearly lags behind the GeForce 4, which is doubly fast in the simple pixel shader test. Nor can the Parhelia catch up with the performance of the R8500. The surprising part is that the Parhelia is faster in the Advanced Pixel Shader Tests than in the standard tests, because this means more effort in calculation for cards with pixel shaders v1.3.

Vertex Shader Speed

This test determines the performance of vertex shading units. Parhelia has four vertex shading units, GeForce 4 Ti has two of them, and R8500 only has one.

  • Vertex Shader
    Vertex shader is like DirectX(r) 7 hardware transformation and lighting, but with a vertex shader, the developer can write a custom transformation and lighting model. While this test demonstrates a vertex shader that does skinning, vertex shaders can be used for countless other types of transformation and lighting, as well. There are 100 skinned characters from game test 3 running around and shooting each other. Vertex shaders can be efficiently emulated in software, therefore this test is run using a software vertex shader, if vertex shaders cannot be run in your graphics hardware.

Vertex Shader

Parhelia's vertex shader performance is just enough to beat ATI's 8500. However, NVIDIA's GeForce 4 Ti is not so far behind the R8500. In the end, when you consider that it has four vertex shaders, results for the Parhelia are rather disappointing.

Conclusion: With regard to the fill rate, Parhelia manages to shine in the important multi-texturing mode. On the other hand, the polygon performance and the poor pixel shader performance are disappointing. And, even with its Quad Vertex Shader Array, Parhelia is still unable to attain the vertex shader performance of the GeForce 4 Ti 4600, at least not in this test.

Matrox SharkMark

SharkMark

Programs that test vertex shader performance are as rare as those that test pixel shaders. Up till now 3D Mark 2001 SE has been the only program that provides a dependable test for these features. Now, Matrox has come out with its own vertex shader test, called SharkMark.

Matrox SharkMark

Here, Parhelia clearly moves ahead of the competition. According to Matrox, this is due to the Quad-Vertex Shader Arrays.

Conclusion

Conclusion

With respect to 3D performance, the Parhelia leaves us with mixed feelings. In contrast to the mediocre performance in the standard benchmarks, it turned up some really good results in the tests for quality, even though they sometimes represent only a very small lead over NVIDIA's GeForce 4 Ti. Another positive point is the excellent fill rate in multi-texturing mode. Here, the Parhelia might even be significantly faster than the GeForce 4 Ti 4600, thanks to its four pixel pipelines, but only if it had a faster GPU clock.

On the other hand, we were disappointed at how the Parhelia performed in the pixel shader tests from 3D Mark 2001 SE. Here, NVIDIA and ATI clearly have the upper hand. This is an aspect that becomes particularly critical when you consider the imminent generation of GPUs from ATI and NVIDIA, which are already positioned at the starting line and will be improving performance and features (DX9) in the area of pixel shading. The new DX9 features of the Parhelia, such as displacement mapping, are currently only used in Matrox demos.

When it comes to standard performance, Parhelia cannot keep apace with NVIDIA's GeForce 4 Ti. Matrox's argument that the strength of the Parhelia lies in the performance at maximum image quality is not entirely valid. In anisotropic filtering, it isn't able attain the same level as the GeForce Ti 4600. Parhelia takes in the high scores only with its high-quality Fragment Anti-Aliasing. The Surround Gaming feature is also quite convincing. If you're enough of an enthusiast, and if you can afford to buy three monitors, then you will be in for an impressive gaming experience.

Pitted against ATI's Radeon 8500 128 MB, the Parhelia offers tough competition in some of the standard tests. However, considering that the Parhelia's price tag is double that of the Radeon ($399 vs. $199), this doesn't exactly seem fitting. In the end, though, the Parhelia is not a top 3D performer, even though its proven 3D capabilities are still quite respectable.

In the 3D comparison, the results clearly point in favor of NVIDIA. Soon, we will be bringing you an evaluation of the Parhelia's 2D features and performance, again, compared to NVIDIA's GF4 and ATI's Radeon 8500. On paper, at least, the Parhelia seems to have the clear advantage. Together with its 2D features, Surround Gaming, 10-bit Gigacolor and good TV-out as well as signal quality, Parhelia can indeed be viewed as an excellent alternative to the established competition, which makes things pretty exciting.

We already presume that the Parhelia will find many friends. However, if you are one of those who expect the Parhelia to be the super-product to topple NVIDIA from the 3D performance throne, you're going to be rather disappointed. Rather, the 3D quality is to be found in the details.

КОНЕЦ СТАТЬИ


Координаты для связи с редакцией:

Общий адрес редакции: thg@thg.ru;
Размещение рекламы: Roman@thg.ru;
Другие координаты, в т.ч. адреса для отправки информации и пресс-релизов, приглашений на мероприятия и т.д. указаны на этой странице.


Все статьи: THG.ru

 

Rambler's Top100 Рейтинг@Mail.ru