NVIDIA GeForce FX 5800 Ultra Preview March 05, 2003 Brandon Bell

Introduction

Page:: ( 1 / 26 )

GeForce FX Variants

Which brings us to March 2003, just over two months since GeForce FX was announced. We now know that two GeForce FX models will be available, the GeForce FX 5800 Ultra which ships at 500MHz core/500MHz memory (1,000MHz effective) and features NVIDIA’s controversial FX Flow cooling system, and the GeForce FX 5800. Unlike the Ultra model, the GeForce FX 5800 is clocked at a more conservative 400MHz core/400MHz memory (800MHz effective) and doesn’t require FX Flow, although we’ve learned that many third party manufacturers will be using cooling units that consume the PCI slot adjacent to the AGP slot. The MSRP of the 5800 Ultra is $399 while the GeForce FX 5800 is priced at $299. In terms of availability, the only manufacturer to ship so far has been BFG Technologies, which began shipping their 5800 Ultra cards last week. From what we’ve heard from our sources, samples of Ultra cards are extremely limited right now, while the 5800 won’t be available for a few more weeks. This certainly isn’t good for NVIDIA, as ATI’s follow-up to the RADEON 9700 PRO is confirmed to be less than a month away from release.

In any case, we were able to get our hands on a GeForce FX 5800 Ultra sample and have run a gamut of tests with it against ATI’s RADEON 9700 PRO. In this preview article we’ll briefly go over the GeForce FX architecture as well as its drivers and the card itself, before diving into the real important stuff: image quality and performance.




SIDEBAR: This is the first in a series of articles we’ll be devoting to the GeForce FX 5800 family. Expect additional articles in the weeks ahead.

FX Delays/Pipelines

Page:: ( 2 / 26 )

0.13-micron Delay

NVIDIA’s GeForce FX is the company’s first foray into TSMC’s 0.13-micron process, which is often fingered as the source of GeForce FX’s numerous delays. It’s no so much the use of 0.13-micron (with copper interconnects) that’s the culprit; rather it’s believed that TSMC’s difficulties with low-k dielectric are the primary cause. Low-k dielectric is a material used to shield the wiring within a chip, reducing electrical crosstalk.

As more and more circuits are packed tighter within chips like GeForce FX, these circuits start to generate interference in one another. Low-k dielectric prevents this from occurring, allowing electronic signals to move cleaner and faster. When GeForce FX was originally conceived it’s widely rumored that it was intended to be built with TSMC’s 0.13-micron process with low-k dielectric. But when TSMC couldn’t get good yields out of its low-k dielectric process, NVIDIA had to scrap the idea. With GeForce FX’s 125 million transistors (versus 63 million in GeForce 4 and 110 million in RADEON 9700), the chip is the most complicated design to come out of TSMC. Complicating matters even further is the chip’s 500MHz core clock frequency, which is also higher than any other graphics processor to date.

It is widely believed that NVIDIA was counting on a 0.13-micron design with low-k dielectric to make GeForce FX’s breakthrough clock frequencies possible. When low-k dielectric wasn’t ready, NVIDIA pressed ahead without it. As a result, NVIDIA has had a hard time getting good yields out of GeForce FX 5800 Ultra parts, which explains why NVIDIA has had such a hard time supplying the retail and OEM segments with cards. Unless TSMC is able to implement low-k dielectric or improve its existing 0.13-micron process, GeForce FX 5800 Ultra may never make it beyond limited availability.

8-Pixel Pipeline or 4?

Another more recent source of controversy surrounding GeForce FX has been the number of pixel pipelines it contains. At launch NVIDIA was proud to boast GeForce FX’s “8 pixels/clock rendering pipeline”. Everyone assumed NVIDIA was referring to an 8-pixel pipeline with 1 texture unit per pipe configuration similar to what ATI has employed in its RADEON 9700 architecture. However, when GeForce FX cards made their way into reviewers hands it was discovered that GeForce FX performed similarly to the 4-pixel pipeline architecture used in GeForce4 in some situations.

Everything really came into the spotlight when NVIDIA voiced its displeasure with the game tests used in FutureMark’s 3DMark 03 synthetic benchmark, specifically game test 1. To quote an NVIDIA whitepaper on the topic:

“For any given frame in the scene, up to 90% of the pixels in the frame are single textured. This occurs because the majority of the scene is the low poly-count, single textured “skybox”, painted to look like sky and clouds. The documentation for 3DMark03 gives some detail about the four-layer multitexture used on the airplanes. Regrettably, all this effort is lost on the fact that the airplanes cover so few pixels on the screen to make these four layers of multitexture completely insignificant. Game Test 1 is, essentially, a single texture fill rate test. No modern games, even DX7 games, are completely dominated by this kind of simple rendering technique.”

Pipline debate (cont’d)

Page:: ( 3 / 26 )

In the whitepaper, NVIDIA brings up a very valid point in that the first game test uses a significant amount of single texturing but with GeForce FX 5800 Ultra’s 500MHz core clock and 8-pixel pipelines NVIDIA theoretically has an advantage over RADEON 9700 PRO in these situations (4,000 Mpixels/sec in GeForce FX versus 2,600 in RADEON 9700) so why should they complain? Lets use 3DMark to see just what kind of fill rate GeForce FX actually delivers:

Fill rate Performance

As you can see in our test results, GeForce FX delivers similar results to what you’d expect from a 4-pixel pipeline architecture (like the GeForce4) in single-textured situations. NVIDIA’s gripe with game test 1 begins to make more sense. NVIDIA has responded by saying that the GeForce FX runs at 8 pixels per clock in:

a) z-rendering
b) stencil operations
c) texture operations
d) shader operations
For advanced applications (such as Doom3) *most* of the time is spent in these modes because of the advanced shadowing techniques that use shadow buffers, stencil testing and next-generation shaders that are longer and therefore make the apps "shading-bound" rather than "color fill-rate" bound.
Only color+Z rendering is done at 4 pixels per clock, all other modes (z, stencil, texture, shading) run at 8 pixels per clock.

Just wanted to write a word or two regarding the issue raised couple of days ago. Seems like the whole Internet community wants to crucify nVidia about the controversy of how many rendering pipelines GeForceFX realy has. Is it 8 pipelines with 1 texture unit, or 4 with 2, or ... uh... I don't know anymore. And it really DOESN'T matter that much!

The only thing that matters is how fast and how good it can render pixels. And both GeForceFX and Radeon9700 are great products, the kind of hardware that developers long for. So, personally, I don't care much what's "under the hood".

Don't get me wrong, I am into 3D-graphic hardware, but this pipeline thing really went out of proportion. Number of pipelines is a good hardware information, and that's all there's to it. It really doesn't need to reflect the speed of the hardware directly. Come to think of it... currently, there are no games that utilize even 1/3rd of nifty features these two boards have.

Oh, before I forget... I'm not "nVidiot" (and I'm not "fanATIc", either). I'm just a game developer who wants good and fast technology for the future. And both ATI and nVidia have it now!

GeForce FX Ultra Card

Page:: ( 4 / 26 )

First Impressions

Physically, the GeForce FX 5800 Ultra board is slightly shorter than the GeForce4 Ti 4600, but not by much. As a result, GeForce FX dwarfs RADEON 9700. Like GeForce4 the power circuitry is located on the back of the board while the Silicon Image Sil 164 (which is also used on GeForce4) DVI transmitter is placed on the underside of the card.

[image]

FX Flow

To keep the GeForce FX GPU cool, NVIDIA has implemented its FX Flow cooling system. Like ABIT’s OTES (which uses a similar Y.S. Tech cooling fan) and Sapphire’s Ultimate Edition, FX Flow uses heat pipe technology to keep the core cool. Like the OTES system a copper heat pipe is soldered to a copper plate, which is in turn attached to the GeForce FX core. You’ll also see that NVIDIA uses a rolled copper fin heatsink (with much larger fins than OTES) and adds a second duct for cool air to help the process along. The enclosure is housed in a plastic covering, creating a sealed environment, which also ensures better cooling.

[image]

Board Production/Drivers

Page:: ( 5 / 26 )

NVIDIA has implemented a 12-layer board design with GeForce FX 5800 Ultra. In comparison, GeForce4 Ti 4200 utilized a 6-layer design while Ti 4400 and Ti 4600 were built on eight layers; ATI’s RADEON 9700 PRO is a 10-layer design. As you add additional layers you get cleaner signals, but manufacturing costs also increase. To put it simply, the GeForce FX 5800 Ultra board design is just as complicated as the graphics core itself.

In light of this, NVIDIA is handling GeForce FX 5800 Ultra board production in-house. NVIDIA CEO Jen Hsun Huang recently went on record discussing GeForce FX board production “At the initial stage, Nvidia will ship 1.5 million NV30 chips and will also make cards based on the top-end GeForce FX 5800 Ultra (NV30 core) itself. The lower-end GeForce FX 5800 chips, however, will be released to downstream card manufacturers for production.”

Gainward has stated that they are producing a GeForce FX 5800 Ultra board of their own that will be up to 7 decibels quieter than NVIDIA’s 5800 Ultra reference board but since this board hasn’t shipped we can’t verify these claims. With NVIDIA’s CEO stating that they’re handling all board production we’re not certain if Gainward has to submit their cooling solution to NVIDIA who then produces Gainward’s boards for them, or if Gainward is able to purchase their Ultra boards without the FX Flow cooling system.

Drivers

NVIDIA’s GeForce FX is now up to driver version 42.69, the drivers we used for this test. NVIDIA has devoted the latter (42.67 and up) drivers for 3DMark 03 optimizations, resulting in a small bug we ran into involving the dynamic clock frequency feature. During testing with Serious Sam we noticed our GeForce FX 5800 Ultra board stutter, once the test was complete we noticed that the score was much lower than we expected.
[image]

DVD/2D Visuals

Page:: ( 6 / 26 )

DVD Playback Quality

First off, visual quality is highly subjective, what looks flawless to one person may be ugly to another. In addition, NVIDIA-based card manufacturers have traditionally manufactured their own boards (and thus, may implement lower quality filters), so any visual quality observations made with NVIDIA’s reference card could be entirely different than real, shipping hardware. However, since NVIDIA is handling all GeForce FX 5800 Ultra board production we feel like we can make some comments regarding the visual quality of the Ultra reference card we received from NVIDIA.

[image]

2D Display

ATI’s 2D display still outshines NVIDIA in our eyes. ATI’s display appears to be warmer, resulting in a very soft display that’s pleasing to the eye. Text with both cards is razor sharp and colors are vibrant so we don’t think you’ll be disappointed with either cards display, but ATI’s desktop just looks a little better.

However, we would like to stress that we aren’t disappointed with NVIDIA’s 2D display the quality is top notch. Gone are the days of third-party NVIDIA manufacturers with grainy text and muddled 2D displays, GeForce FX is right up there with the best of them in image quality.






SIDEBAR: NVIDIA is still offering free trial versions of NVDVD 2.0, first announced at Comdex.

Anisotropic filtering

Page:: ( 7 / 26 )

ATI’s adaptive anisotropic filtering engine has always been one of its strongest suits, both the RADEON 8500 and RADEON 9700 were highly regarded for their anistropic filtering quality. As a result, NVIDIA has put more of an emphasis on its anisotropic engine, and contained within the latest Detonator driver is NVIDIA’s response.

While older Detonator drivers weren’t very flexible, the Detonator 42.6x series incorporates three anistropic filtering settings – balanced, application and aggressive. The balanced setting is default, and, according to NVIDIA, is meant to compete with ATI’s “quality” setting. As its name suggests, the aggressive setting uses lower quality samples than in turn results in higher performance at the cost of image quality. The application setting uses whatever aniso level is requested by the application.

To compare the visual quality of ATI and NVIDIA’s latest offerings (as well as GeForce4), we tested with Unreal Tournament 2003’s DM-Insidious level. We’ve included screenshots of our testing below, but if you really want to get an accurate representation of what was displayed on the screen, we highly suggest you download the original bitmaps. We’ve zipped them up into one file for your convenience.

First, we have our scene with anisotropic filtering disabled (taken with GeForce FX 5800 Ultra):

[image]

Final Thoughts

NVIDIA has really come a long way with its latest Detonator drivers, anistropic filtering quality is drastically improved over older driver revisions. They deserve tons of credit for stepping up to the challenge, and as you can see they’ve delivered a very good package. However, we still have to give the nod to ATI, not only does their engine go one step further than NVIDIA with its 16x setting, ATI’s performance mode delivers better image quality. ATI better not rest on their laurels for too much longer though, as NVIDIA has made some great strides in anistropic filtering.





SIDEBAR: NVIDIA was recently awarded with MPR’s Best Graphics Processor of 2002 award but as far as we know the card wasn’t available at retail in 2002.

Anti-Aliasing

Page:: ( 8 / 26 )

To test anti-aliasing image quality, we used 3DMark 03’s image quality test. If you’ve followed some of our other preview articles, you’ve probably noticed that we prefer to use 3DMark for all image quality anti-aliasing tests. The reason we prefer to use 3DMark is simple, jaggies can creep up at a moment’s notice. One slight twitch of the mouse can make a difference. The only way to ensure that we’re getting the exact same image is to use 3DMark’s image quality test, which grabs a screen capture of a static display.

We chose game test 1 because it’s a bright map (and thus jaggies are easier to see), and also because flight simulations are notorious for jagged edges. The long, straight lines of a planes’wing are a great place to spot them, while the edge of the horizon is a perfect place to spot shimmering. In this case all you have to do is look along the leading edge of the B-17’s wing and tail to see the jagged edges. We used the default screen resolution of 1024x768 for testing.

First lets look at a shot with anti-aliasing turned off (taken with GeForce FX 5800 Ultra):

[image]

Test Systems

Page:: ( 9 / 26 )

System Setup

Intel Pentium 4 3.06GHz

ASUS P4G8X Deluxe (E7205 “Granite Bay”)

512MB Mushkin PC3200 (operating at DDR266) SDRAM

NVIDIA GeForce4 Ti 4600
NVIDIA GeForce FX 5800 Ultra
Detonator 42.69
ATI RADEON 9700 PRO
Driver version Catalyst 3.1

30GB IBM Deskstar DTLA 307030 ATA/100 Hard Drive

Windows XP Professional

DirectX 8.1
DirectX 9.0

Benchmarks

3DMark 2001 SE
3DMark03
Quake III: Arena version 1.17
Serious Sam: The Second Encounter (Elephant Atrium demo)
Unreal Tournament 2003
Comanche 4
FableMark
ChameleonMark
Codecult Codecreatures

Test Notes

We decided to include test results with both NVIDIA modes (aggressive as well as balanced) in order to provide a more complete picture of the card’s performance. To keep things as similar as possible, we also decided to do the same with the RADEON 9700 PRO. However, in some cases we did have to manually adjust the quality sliders for the ATI card in order to maintain adequate visual quality in Open GL applications. Therefore, we’re keeping all of our official comments solely to the default settings used by both drivers.



SIDEBAR: For more performance, we ran our system with CAS Latency 1.5 timings.

T&L Performance/Bump mapping

Page:: ( 10 / 26 )

T&L Performance

Notes

While RADEON 9700 PRO is only an evolutionary step above GeForce4, GeForce FX 5800 Ultra demonstrates remarkable performance in polygon testing with 3DMark 2001 SE. GeForce FX 5800 Ultra delivers twice the performance of RADEON 9700 PRO in more intensive tests with eight light sources, while GeForce FX 5800 Ultra delivers 1.5 times the performance of RADEON 9700 PRO in testing with one light.

Bump mapping

Notes

In bump mapping tests, GeForce FX 5800 trails RADEON 9700 PRO by 3% in environment mapped bump mapping, but finishes seven percentage points ahead of the ATI card in dot3 bump mapping.




SIDEBAR: We’ll continue to incorporate a mixture of 3DMark 2001SE and 3DMark 03 for now, as both definitely have their place in our suite of benchmarks.

Shader Performance

Page:: ( 11 / 26 )

Pixel Shader Performance

Vertex Shader Performance

ChameleonMark

Notes

We ran pixel shader tests with both 3DMark 2001SE (depicted as “01” in the graphs) and 3DMark 03 (the pixel shader 2.0 figures in the graphs). The advanced pixel shader test comes from 3DMark 2001SE. In the simple pixel shader test we see that the GeForce FX 5800 Ultra outperforms RADEON 9700 PRO by 3%, but once 2.0 pixel shaders are used RADEON 9700 PRO nearly delivers 1.5 times the performance of GeForce FX 5800 Ultra. In advanced pixel shader testing we see similar results, the margin is actually slightly greater.

In vertex shader testing, RADEON 9700 PRO comes out on top in all tests, although the margins are much closer (1.1 vertex shaders are used). And while the ragtroll test doesn’t directly test vertex shader performance, it is still a good indicator of vertex shader performance as the trolls in the test are skinned with 1.1 vertex shaders. Again, RADEON 9700 PRO outperforms GeForce FX 5800 Ultra, the margin increases to 18%.

Finally, even NVIDIA’s own shader test, ChameleonMark, gives the edge to ATI. We averaged the three tests (shiny, glass and real) to obtain the final result, with ATI’s RADEON 9700 PRO offering roughly 24% higher performance than GeForce FX 5800 Ultra.



SIDEBAR: We used ChameleonMark 1.0 for our tests. As far as we know this is the latest version of the benchmark available.

Occlusion Culling

Page:: ( 12 / 26 )

FableMark

Notes

PowerVR’s fablemark benchmark essentially takes its popular villagemark test to the next level by incorporating an updated graphics engine with soft shadows. The end result is a more stressful benchmark. GeForce FX 5800 Ultra and RADEON 9700 PRO run neck-and-neck in this test with both cards offering up to 2.5 times the performance of GeForce4. At 1600x1200 GeForce FX 5800 Ultra is able to take a substantial 16% lead, we ran a few quick tests with VillageMark and noticed an even larger performance advantage for GeForce FX 5800 Ultra.






SIDEBAR: You can find more details on FableMark at PowerVR’s website

Codecreatures

Page:: ( 13 / 26 )

Codecult Codecreatures

Notes

From here on you’ll see the use of NVIDIA’s balanced “bal” and aggressive “aggr” modes, while we used ATI’s performance “perf” and quality settings “qual” to provide as complete a picture as possible of the performance of these cards. We’ll leave our official comments however to their default settings, as we had to make some custom driver adjustments to the 9700 PRO in OpenGL titles to maintain adequate visual quality. Besides, that’s the setting most gamers are likely to use.

Codecreatures benchmark pro uses vertex and pixel shaders heavily, so based on the synthetic tests we saw earlier we expected the RADEON 9700 PRO would come ahead of GeForce FX 5800 Ultra in this test. However, as you can see NVIDIA scored a clean sweep, the margin remains a consistent 14% across all resolutions.




SIDEBAR: Codecult used the Codecreatures engine to model a high-speed train track in Shanghai.

3D Mark 2001 SE

Page:: ( 14 / 26 )

3D Mark 2001 SE v.330

Notes

The margin between RADEON 9700 PRO and GeForce FX 5800 Ultra is just 2% at lower resolutions in 3DMark 2001SE, but once we crank up the screen resolution we see that GeForce FX 5800 Ultra is able to pull ahead. Lets take a closer look at the scores though.



SIDEBAR: NVIDIA’s GeForce FX still hasn’t made its debut on Futuremark’s list of top graphics cards.

3D Mark 2001 SE – Frame Rates

Page:: ( 15 / 26 )

3DMark 2001 - Car Chase

3DMark 2001 - Dragothic

3DMark 2001 - Lobby

3DMark 2001 - Nature

Notes

FutureMark’s car chase test uses physics calculations for the car’s suspension and movements, the truck itself has three texture layers, while the wrecked cars and some houses in the scene use two textures. In low detail mode, two textures layers are used, while high detail adds an additional texture layer (as well as additional objects with more detail) and dynamic shadowing. In low detail mode GeForce FX 5800 Ultra outpaces RADEON 9700 PRO by 7%, while RADEON 9700 PRO takes a 16% lead in high detail mode. Dragothic also makes extensive use of multiple textures and vertex shaders are used for morphing the animations of the dragon, bowmen, and the people in the village.

While GeForce FX offers similar levels of performance regardless of the mode its running in, we see that the margin between the RADEON’s two modes is significantly greater in low detail mode. It’s possible that we may see a similar situation with GeForce FX with mature drivers. In high detail mode bandwidth limitations constrict both cards but GeForce FX is able to maintain its lead, likely due to its multi-texturing fill rate advantage.

The lobby demo uses vertex shaders to skin the characters in the lobby demo, all of the characters have a single texture layer, although the room uses two layer textures extensively. Considering GeForce FX’s fill rate disadvantage with single texturing and the vertex shader performance we demonstrated in 3DMark 2001SE and 3DMark03 (as well as the ragtroll test), it isn’t surprising to see GeForce FX fall to RADEON 9700 PRO in the low detail lobby test. In high detail mode, the margin between both cards remains the same (5%), the big difference between the two tests are the reflections present in high detail mode as well as the dynamic shadows and damage done to the walls.

We expected similar results with the nature test, as vertex shaders are used extensively in addition to the pixel shader that is used for the water. And we all saw the pixel and vertex shader test results in 3DMark 2001, 03, and even NVIDIA’s own ChameleonMark test a few pages ago (all favoring RADEON 9700 PRO). However, just as Codecreatures favors GeForce FX, the same applies here. Our best guess is that the multiple texture layers that are used for the water, house, bridge and landscape give NVIDIA an edge. We saw an extensive performance advantage for GeForce FX 5800 Ultra in multi-texturing fill rate.



SIDEBAR: That’s probably the most we’ve commented on 3DMark 2001’s frame rate results in quite a while!

3DMark03

Page:: ( 16 / 26 )

3DMark03

Notes

FutureMark’s controversial 3DMark03 gives the nod to GeForce FX 5800 Ultra. We’ve seen how NVIDIA can play the numbers in the overall score with a quick driver update, so lets head straight to the frame rate results.



SIDEBAR: What do you think of the new 3DMark benchmark? Drop your thoughts in the news comments!

3DMark03 – Frame Rates

Page:: ( 17 / 26 )

3DMark03 – Wings of Fury

3DMark03 – Battle of Proxycon

3DMark03 – Troll’s Lair

3DMark03 – Mother Nature

Notes

While NVIDIA isn’t a big fan of game test 1 (wings of fury), due to its simple nature, GeForce FX 5800 Ultra actually performed well in this benchmark, outperforming RADEON 9700 PRO by 2%. Again, we’re not commenting on the “performance” and “aggressive” settings, we want to stick with the defaults to keep the analysis simple but we still wanted to provide the numbers for you to compare. Also keep in mind that ATI’s performance mode offers visual quality that is actually closer to NVIDIA’s balanced mode than NVIDIA’s performance-oriented “aggressive” mode.

NVIDIA’s arguments with game tests 2 (proxycon) and 3 (troll’s lair) are largely its inefficient coding and the use of 1.4 pixel shaders, which aren’t used as frequently as 1.1 and 1.3 pixel shaders in today’s games. However, we’re still seeing GeForce FX outperforming RADEON 9700 PRO, particularly in game test 2. We would like to add that we considered omitting the GeForce4 scores, as visual errors were apparent everywhere in tests 2, 3, and 4.

NVIDIA feels that game test 4 doesn’t go far enough, 2.0 pixel shaders are used for the leaves, water surface, and sky, while 1.1 pixel shaders are used for the grass and 1.4 pixel shaders for the ground. With such extensive use of pixel shaders, we’d again assume that GeForce FX 5800 Ultra would be at a disadvantage based on the pixel shader results we saw earlier in both 3DMark 2001SE and 3DMark03 but as you can see that isn’t the case.
We know that NVIDIA has taken a lot of time to optimize its performance in tests 2, 3, and 4 so we can’t speculate on what’s really happening. We do know that GeForce FX 5800 Ultra has finished behind RADEON 9700 PRO in synthetic pixel and vertex shader tests (including ChameleonMark), but once we throw synthetic benchmarks like 3DMark 2001, 03, and Codecreatures at GeForce FX, it comes out on top. Even in those tests from 3DMark that use shaders extensively. Lets try and settle this with some real games shall we?



SIDEBAR: NVIDIA recently dropped out of Futuremark’s beta program. If ATI, S3, SiS, Matrox, PowerVR, and Trident are in there we’d like to see NVIDIA get back in the game.

Serious Sam SE

Page:: ( 18 / 26 )

Serious Sam SE (Elephant Atrium) – OpenGL

Notes

The RADEON 9700 PRO has historically had a more difficult time with the Serious Sam engine, that continues to hold true to this day. At 1600x1200x32 we’re getting very playable frame rates out of both GeForce FX 5800 Ultra and RADEON 9700 PRO, although the GeForce FX card does hold a thin 1% lead at that resolution.




SIDEBAR: With its impressive flexibility and individual game tests designed to stress difference PC components, Serious Sam has always been one of our favorite benchmarks.

Quake III: Arena

Page:: ( 19 / 26 )

Quake III - OpenGL

Notes

We eagerly await the arrival of Doom III, if only so we can finally retire Quake 3 from our stable of benchmarks: we’re seeing 200+ frames per second at 1600x1200! It’s truly a testament to the huge gap between the hardware and software out there today. At 1600x1200 GeForce FX holds a pretty substantial 17% lead, although with the performance we’re seeing in Q3 today you’d be crazy to play this game without AA and anistotropic filtering turned on.




SIDEBAR: In early launch documents NVIDIA claimed that GeForce FX would offer 3 times GeForce4’s performance but we just aren’t seeing it so far.

Comanche 4

Page:: ( 20 / 26 )

Comanche 4 – DirectX 8

Notes

While the Comanche 4 benchmark test uses pixel shaders extensively, it’s unfortunately a better processor benchmark than a graphics-related test. However, at 1600x1200 we do see that RADEON 9700 PRO finishes 5% ahead of GeForce FX 5800 Ultra.




SIDEBAR: Earlier Comanche titles actually used voxels, the graphics were excellent for their time but obviously the technology never took off.

Unreal Tournament 2003

Page:: ( 21 / 26 )

Unreal Tournament 2003 Flyby – DirectX 8

Unreal Tournament 2003 Botmatch – DirectX 8

Notes

We really like the Unreal Tournament 2003 flyby test for running video tests; botmatch (with its extensive use of AI) on the other hand is good for testing processor performance. With that being said, we see that GeForce FX 5800 Ultra outperforms RADEON 9700 PRO at the higher resolutions we’re likely to play it (by up to 18%), while RADEON 9700 PRO dominates the lower resolutions. While this is definitely a win for NVIDIA, we still have enough horsepower to spare with both cards that we can turn on AA and anistropic filtering, we’ll get to those results in a minute.




SIDEBAR: UT 2003 was one of the first games that we can recall that is a part of NVIDIA’s “Way It Was Meant To Be Played” program.

4x Anti-Aliasing

Page:: ( 22 / 26 )

Quake III – High Quality

Notes

Under the additional stress of 4x AA mode, we see that the GeForce FX 5800 Ultra loses its performance advantage it enjoyed earlier, trailing RADEON 9700 PRO by 5% at 1600x1200.




SIDEBAR: The upcoming Tomb Raider game is also a part of NVIDIA’s meant to be played program

Q3 4x Anti-Aliasing and 8x Anisotropic Filtering

Page:: ( 23 / 26 )

Quake III – High Quality

Notes

With 4xAA and 8x anisotropic filtering, GeForce FX 5800 Ultra is able to put up a better fight than it did with just 4xAA enabled, but it still isn’t enough to overcome RADEON 9700 PRO, which bests GeForce FX 5800 Ultra by 4% at 1600x1200.



SIDEBAR: PNY has taken up the professional workstation market for NVIDIA. The company is offering Quadro FX 2000 as well as 1000 products.

Serious Sam SE AA/Aniso

Page:: ( 24 / 26 )

Serious Sam SE (Elephant Atrium) – OpenGL

Notes

Even with AA and anistropic filtering enabled, Serious Sam 2 continues to favor GeForce FX 5800 Ultra. In fact, with its 9% performance advantage at 1600x1200, the Ultra board was able to open up a little breathing room in this test.



SIDEBAR: Despite its recent setbacks, NVIDIA still holds the #1 position in desktop market share.

Unreal Tournament 4x Anti-Aliasing, 8x Anisotropic Filtering

Page:: ( 25 / 26 )

Unreal Tournament 2003 Flyby

Unreal Tournament 2003 Botmatch

Notes

While the scores remain close, ATI’s RADEON 9700 PRO is able to maintain a lead over GeForce FX 5800 Ultra in Unreal Tournament 2003 once anistropic filtering and anti-aliasing are turned on. At 1600x1200 the margin between both cards is at its greatest, 7%.




SIDEBAR: Thank goodness, almost done with these things!

Conclusion

Page:: ( 26 / 26 )

From a pure performance perspective, GeForce FX 5800 Ultra is not the performance champion we were lead to believe it would be back in November. Sure, GeForce FX 5800 Ultra wins its fair share of benchmarks (especially in non-AA situations), but for the most part ATI’s RADEON 9700 PRO was able to keep up with, our outperform it once the visuals were really cranked up a few notches.

NVIDIA deserves lots of credit for improving the flexibility of its anisotropic filtering. We’ve seen that in its balanced mode, it offers image quality that is dramatically improved over NVIDIA’s prior efforts. Fortunately this is a feature present in the newer Detonator drivers, so GeForce4 owners can also take advantage of this option if they chose to do so. With a little more fine-tuning NVIDIA could have a real advantage here.

Overall though, we can’t help but feel slightly disappointed with GeForce FX. While the 5800 Ultra model was announced over three months ago, these boards still can’t be found on store shelves and rumors are running rampant that you never will. When you consider that ATI’s RADEON 9700 PRO has been available for over six months now, it makes the situation even worse for NVIDIA.

Complicating matters is the heat output of the card and noise from the FX Flow cooling unit, both of these issues are turnoffs to many gamers. In addition, the scalable clock frequency “feature” can sometimes underclock your GeForce FX 5800 Ultra card right in the middle of gaming. We had to repeat multiple runs of Serious Sam and Quake 3 running with 4xAA/8xAniso enabled to get our final numbers, in some cases the margin between the scores was as high as 30%!

Of course, if you look back at NVIDIA’s last architecture change, the GeForce2 to GeForce3 transition, GeForce FX certainly looks better purely from NVIDIA’s historical perspective. While GeForce3 offered few performance gains over GeForce2 Ultra, GeForce FX 5800 Ultra demonstrates a remarkable improvement over GeForce4 Ti 4600. The key difference however, is two years ago GeForce3’s closest competitor was an NVIDIA product. This time around ATI beat NVIDIA to the punch by several months and built a highly competitive part.

Ultimately consumers will decide which product they prefer. But in order for NVIDIA to even have a chance of winning the high-end segment back they’ve got to ship more GeForce FX 5800 Ultra parts. And ATI isn’t sitting on their heels, R350 is right around the corner. Clearly NVIDIA’s days of dominance are over.

The GeForce FX 5800 Ultra is a great performer with today’s games and applications, but we’re going to reserve final judgment on the GeForce FX versus RADEON 9700 debate until more DirectX 9 titles arrive. Doom 3 (and the underlying games based on this game engine) is going to sell lots of graphics cards for ATI and NVIDIA, and right now it’s still unclear if ATI’s DX8/DX9 hyrbrid approach taken with RADEON 9700 is best, or if GeForce FX’s forward-looking design is ultimately proven to be the winning strategy. In any case, we’ll be here to report along the way, 2003 should prove to be a very interesting year for 3D graphics!



SIDEBAR: Are you waiting to pick up GeForce FX or do you plan on going with ATI for your next upgrade? Voice your thoughts on the 3D market, the GeForce FX 5800 Ultra, and this article in general in the news comments!