[ Print Article! ]

NVIDIA GeForce 6800 Ultra Performance Preview
April 14, 2004 Brandon Sandman Bell

Summary: NVIDIA has been confidently claiming revolutionary performance and features in the upcoming NV40 line. Brandon puts the flagship GeForce 6800 Ultra through the paces and then some. With benchmarks ranging from the venerable Quake III Arena, through Tomb Raider which proved so troublesome to NV3X, finally to the super-modern and demanding Far Cry, this NVIDIA GeForce 6800 Ultra performance preview article is as complete as you might wish. Come take a look at what's coming soon to desktops near you, and what ATI will have to face for the foreseeable future.


IntroductionPage:: ( 1 / 24 )


First introduced in 1995, Microsoft’s DirectX application programming interface (API) was designed to make life easier for developers by providing a standard platform for Windows-based PCs. Before the arrival of DirectX, developers had to program their software titles to take advantage of features found in individual hardware components. With the wealth of devices on the market, this could become a tedious, time-consuming process.

[image]
<% print_image("01"); %><% print_image("02"); %>
DirectX is composed of a suite of multimedia APIs such as graphics (both 2D and 3D), audio, and input devices. Each API provides developers access to all the features found in the latest hardware without having to program directly for it. This simplifies the development process for developers, which should in theory bring titles to market more quickly.

Competing with Direct3D (the API within DirectX that handles 3D graphics) is OpenGL. OpenGL provides the same capability with the added bonus of being cross-platform; this allows developers to offer their titles across multiple platforms such as the Mac and Linux communities.

Initially Direct3D got off to a slow start, OpenGL and 3dfx’s proprietary API, Glide, were used in many predominant games, but newer revisions of Direct3D have been adopted by an increasing number of software developers. As a result, the majority of today’s gaming titles now take advantage of DirectX rather than OpenGL.

[image]
<% print_image("03"); %><% print_image("04"); %><% print_image("05"); %>

Traditionally, NVIDIA has timed the introduction of its latest graphics cards with the release of each new revision of DirectX. When Microsoft released DirectX 7, NVIDIA was ready with GeForce 256. Similarly, when Microsoft unveiled DirectX 8, NVIDIA unleashed GeForce3. Each of these cards was the first of their generation to be released on the market, giving NVIDIA bragging rights and allowing the company to establish a track record that was unmatched for its era.

All this came to an end however with ATI’s RADEON 9700 PRO. ATI had not only beat NVIDIA to market, the RADEON 9700 PRO also predated Microsoft’s own DirectX 9 API. The RADEON 9700 PRO was a breakthrough product, not only supporting 2.0 pixel and vertex shaders but also featured an 8-pixel pipeline architecture with a 256-bit memory interface. This gave the RADEON 9700 PRO extraordinary performance in a wide variety of games and applications. NVIDIA’s most effective reply didn’t come until over six months later with the introduction of GeForce FX 5900 Ultra.

Now NVIDIA is back with its latest GPU, GeForce 6800 Ultra. And unlike the GeForce FX 5900 Ultra refresh, 5950 Ultra, GeForce 6800 Ultra is a true next generation product and is built on an entirely new architecture. NVIDIA has crammed this new graphics core with all kinds of goodies: 16 pixel pipelines, 1.1GHz GDDR3 memory, and an enhanced, rotated-grid AA implementation. Clearly NVIDIA didn’t pull any punches with this card.


SIDEBAR: NVIDIA just held a launch party in the Bay Area for GeForce 6800 Ultra.


SpecificationsPage:: ( 2 / 24 )

The list

CineFX 3.0 Shading Architecture
Vertex Shaders
Support for Microsoft DirectX 9.0 Vertex Shader 3.0
Displacement mapping
Vertex frequency stream divider
Infinite length vertex programs
Pixel Shaders
Support for DirectX 9.0 Pixel Shader 3.0
Full pixel branching support
Support for Multiple Render Targets (MRTs)
Infinite length pixel programs
Next generation texture engine
Up to 16 textures per rendering pass
Support for 16-bit floating point format and 32-bit floating point format
Support for non-power of two textures
Support for sRGB texture format for gamma textures
DirectX and S3TC texture compression
Full 128-bit studio-quality floating point

NVIDIA high-precision dynamic range (HPDR) technology
Full floating point support throughout entire pipeline
Floating point filtering improves the quality of images in motion
Floating point texture drives new levels of clarity and image detail
Floating point frame buffer blending gives detail to special effects like motion blur and Explosions
New rotated-grid anti-aliasing removes jagged edges for incredible edge quality

Intellisample 3.0 technology
Advanced 16x anisotropic filtering
Blistering fast anti-aliasing and compression performance
Support for advanced lossless compression algorithms for color, texture, and z-data at
even higher resolutions and frame rates
Fast z-clear
High-resolution compression technology (HCT) increases performance at higher
resolutions through advances in compression technology

UltraShadow II technology
Designed to enhance the performance of shadow-intensive games, like id Software’s
DOOM III

Advanced Engineering
Over 220 million transistors
Designed for PCI Express x16
Supports PCI Express high-speed interconnect (HSI) technology for bidrectional interconnect protocol conversion
Full support of AGP 8X including fast writes and sideband addressing
Support for the industry’s fastest GDDR3 memory
256-bit advanced memory interface
0.13-micron process technology
Advanced thermal management and thermal monitoring
40mm x 40mm, BGA flip-chip package

Advanced Video and Display Functionality
Dedicated on-chip video processor
MPEG video encode and decode
WMV9 decode acceleration
Advanced adaptive de-interlacing
High-quality video scaling and filtering
Integrated NTSC/PAL TV encoder supporting resolutions up to 1024x768 without the need for panning with built-in Macrovision copy protection
DVD and HDTV-ready MPEG-2 decoding up to 1920x1080i resolutions
Dual integrated 400MHz RAMDACs for display resolutions up to and including 2048x1536 at 85Hz.
Dual DVI ports for interfacing to external TMDS transmitters and external TV encoders
Microsoft Video Mixing Renderer (VMR) supports multiple video windows with full video quality and features in each window
VIP 1.1 interface support for video-in function
Full NVIDIA nView multi-display technology compatibility

Notes

GeForce 6800 Ultra is the world’s first DX9.0c graphics card. Note, we said DX9.0c and not DX9.1, which was the designation incorrectly used to denote the addition of 3.0 pixel and vertex shaders into DX9. Another highlight in GeForce 6800 Ultra that will probably grab your attention is the chip’s transistor count: over 220 million! We were given a figure of 222 million at NVIDIA’s Editor’s Day. Basically, this is one big chip.

GeForce FX 6800 is also NVIDIA’s first native PCI Express graphics solution. Follow-up GeForce 6 series parts will also support PCI Express natively. NVIDIA has also integrated a video processor on the graphics core, circumventing the need for an external chip.

Lets take a closer look at a few of the more important features found in GeForce 6800 Ultra, starting with the new 3.0 pixel and vertex shaders.


SIDEBAR: NVIDIA’s codename for GeForce FX 6800 was NV40.


Shader Model 3.0Page:: ( 3 / 24 )

Before we discuss the changes in the 3.0 pixel and vertex shader models, we’ll start with the traits they both share in common.

[image]

<% print_image("06"); %><% print_image("07"); %>

More instructions

While shader model 2.0’s shaders have the potential to bring a new level of photorealism to the PC, the 2.0 shader model was limited in the length of shader instructions: shader model 2.0 was officially limited to a pixel shader length of 96 instructions. In shader model 3.0, this figure is increased to 32,768 instructions. This increase allows developers to implement more complex shader programs. If you recall NVIDIA’s NV3x architecture, one of the primary highlights NVIDIA touted was instruction limits beyond the shader 2.0 model on both pixel and vertex shaders, NVIDIA referred to this as “2.0+” shaders. Keep in mind however that this doesn’t mean NV3x series is shader model 3.0 compliant, as there are other new features that have been introduced.

For the GeForce 6 series (and thus, GeForce 6800 Ultra) NVIDIA boasts their GPUs support vertex and pixel shader programs of an infinite length, although obviously a cutoff point has to be established in order to ensure optimal performance, as longer programs take more time to finish than shorter programs.

Besides providing maximums, shader model 3.0 also increases the minimum length of shader programs as well. A minimum of 512 instructions is stipulated for both pixel and vertex shader 3.0.

Dynamic looping/branching (Flow Control)

One feature that has been implemented in shader model 3.0 that is designed to make writing shaders easier for developers is dynamic branching/looping (also referred to as dynamic flow control). With dynamic branching, developers can add loops to their shader
programs.

NVIDIA mentions an example involving multiple light sources. In previous shader models, the developer would have to write a shader for each light. Dynamic branching makes it possible for the developer to write one shader which loops through a certain number of vertex lights and exits once all the lights have been processed. This will help to reduce shader count complexity. Another potential advantage to branching is reducing the variety of shaders used (i.e. many different shaders versus one).

Besides eased development, shader model 3.0 also presents potential performance improvements. For example, developers can use dynamic branching to skip unnecessary processing and help speed up the shader.

Branching, if not used carefully, can introduce slower performance. At the moment branching in graphics isn’t as powerful as in CPUs, which have considerably more resources for handling branching. Therefore, there’s another school of thought that believes branching, in its current form in shader model 3.0 is a little early.


SIDEBAR: NVIDIA’s splash page for NV40 read: prepare to drench your senses.


Shader Model 3.0 (cont’d)Page:: ( 4 / 24 )

Centroid sampling support

If you think back to last summer’s Half-Life 2 AA fiasco that erupted on the halflife2.net forums, chances are you’ve heard of centroid sampling. Just in case though we’ll provide a quick refresher.

While multisample anti-aliasing does a good job of cleaning up jaggies without to much of an impact on performance, one of the limitations is that it takes its texture sample from the center of a pixel. As a result, the wrong texel could be sampled, causing artifacts on the polygon boundaries where this occurs. This is a problem in any game that uses lightmaps with anti-aliasing turned on, including the entire Quake series and the original Half-Life.

With centroid sampling, if the triangle fails to cross the pixel’s texture sampling position, it takes the sample from within the centroid of the covered samples, hence the name centroid sampling.

ATI first introduced this feature in RADEON 9700, now it has become a standard feature in pixel shader 3.0 and is supported by NVIDIA’s GeForce 6 series.

FP32 minimum

Another topic that was hotly contested last year was NVIDIA’s decision to adopt FP32 whereas ATI relied on FP24 for its series of DX9 RADEON cards. In shader model 3.0, FP32 is now the required minimum precision for compliance, all others are considered partial precision. The end result is higher fidelity particularly when dealing with effects like high dynamic range lighting and fog.

On GeForce FX 6800, programmers can choose between the core’s native 32-bit mode or, if memory storage is a concern, 16-bit mode. Other formats are also supported, although not FP24.

Multiple Render Targets

Like centroid sampling, multiple render targets is another feature first introduced in RADEON 9700. Multiple render targets allows the pixel shader to save pixel data into buffers. This data can then be used to produce complex lighting effects after the geometry is rendered without having to run multiple passes through the scene.

Vertex processing with textures

The key addition in vertex shader 3.0 (besides the aforementioned aspects like dynamic branching and instruction count) is vertex textures. This allows vertex shaders to do texture lookups.

This feature makes displacement mapping possible in GeForce FX 6800. With this technique, texture data can be mapped onto vertices, providing more control over the shape of objects and surfaces. Previously developers used bump maps to add detail. Bump maps looked good, but with displacement maps, developers can create objects with ridges, peaks and valleys, and more detailed textures.
[image]

<% print_image("08"); %>

High Dynamic Range lighting

NVIDIA has integrated high dynamic range lighting in its GeForce 6 series of GPUs. If you recall, previous designs were limited to just 256 levels for each color, with one of those 256 hues being black. This is particularly important in rendering a scene with wild lighting variations: an area where light from the sun can be very bright, while the shadows cast can be very dark. Without HDR, these variations can’t adequately be rendered, resulting in less lifelike images. Valve has demonstrated HDR in action in Half-Life 2.



SIDEBAR: Centroid sampling is also supported in pixel shader 2.0b.


Core details/Architecture/VideoPage:: ( 5 / 24 )
[image]

<% print_image("09"); %>
Arguably one of the worst kept secrets regarding GeForce 6800 Ultra is its 16-pipeline architecture. That’s right, 16 pixel pipelines! Coupled with the GeForce 6800 Ultra’s 16 pixel pipelines is one texture unit for a 16x1 architecture, this is double that of RADEON 9700/9800, which features an 8x1 architecture. The GeForce 6800 Ultra graphics core is clocked at 400MHz, producing a peak texel fill rate of 6.4Gigatexels/second. This figure is nearly twice that of RADEON 9800 XT’s 3.3Gigatexels/second and just over 40% greater than GeForce FX 5950 Ultra. Of course, pixel pipeline and fill rate figures will become less important in time, but for now it’s still an important performance metric.

The GeForce 6800 Ultra graphics core is built on IBM’s 0.13-micron manufacturing process and, as we mentioned previously, contains a whopping 222 million transistors. In comparison, Intel’s Pentium 4 Extreme Edition contains 178 million transistors. This means the chip is more expensive to produce for NVIDIA, assuming equal yields. Like GeForce FX 5950 Ultra, NVIDIA went straight into production from A01 silicon, indicating NVIDIA and IBM have a knack for getting their design right the first time.

[image]
<% print_image("10"); %><% print_image("11"); %><% print_image("12"); %>

GeForce 6800 Ultra’s vertex processing engine consists of six vertex units, another aspect which compares favorably for NVIDIA in comparison to ATI’s high-end DX9 cards such as the RADEON 9800 XT, which only sport four vertex units.

The memory subsystem of GeForce 6800 Ultra is equally robust. NVIDIA has partnered with Samsung to provide GDDR3 memory on their GeForce FX 5700 Ultra, we previewed this card last month. Now NVIDIA is taking Samsung’s GDDR3 to the high-end of the market, as the GeForce 6800 Ultra utilizes 256MB of GDDR3 operating at 550MHz (1.1GHz effective). This provides up to 35.2GB/sec of peak memory bandwidth to the GeForce 6800 Ultra graphics core, the highest figure in the industry.

GeForce FX variants

Like previous GeForce cards, NVIDIA will be producing multiple GeForce 6800 variants. At the high-end of the GeForce 6800 family is GeForce 6800 Ultra. This is the card we’ve been focusing on up until now. The GeForce 6800 Ultra will occupy the same $499 price point previous GeForce FX cards have launched at in the past.

Just below the GeForce 6800 Ultra is GeForce 6800. GeForce 6800 offers the same key features as GeForce 6800 Ultra, namely 3.0 pixel and vertex shaders, only it has been designed to be less expensive for NVIDIA to manufacture.

Whereas GeForce 6800 Ultra sports 16 pixel pipelines, for GeForce 6800, NVIDIA has reduced that number to 12. This change reduces transistor count (although we weren’t given an exact figure), making the die size smaller and therefore NVIDIA is able to yield more chips per silicon wafer, ensuring lower production costs. NVIDIA has not announced clock speeds on the GeForce 6800, although it’s expected to debut around the same timeframe as GeForce 6800 Ultra, late May.

[image]

<% print_image("13"); %><% print_image("14"); %><% print_image("15"); %>

We’ve included pictures of a mockup GeForce 6800 Ultra board with single slot cooling. This cooler will ultimately ship on GeForce 6800 and not GeForce 6800 Ultra (athough board manufacturers may certainly implement their own single-slot GeForce 6800 Ultra boards, as ASUS has been doing for some time now). The power requirements for GeForce 6800 also differ from GeForce 6800 Ultra. The 6800 will ship with one Molex power connector rather than the two connectors used on 6800 Ultra.


SIDEBAR: GeForce FX 5950 Ultra contains 130 million transistors.


The boardPage:: ( 6 / 24 )
[image]

<% print_image("16"); %><% print_image("17"); %><% print_image("18"); %>
Like GeForce FX 5950 Ultra and the 5900 Ultra before it, GeForce 6800 Ultra ships with a tall cooler that will occupy the PCI slot adjacent to your AGP slot. And while the board looks massive, it’s actually a little lighter than GeForce FX 5950 Ultra. Both boards are about the same length, roughly as long as a GeForce4 Ti 4600, so we don’t think you’ll have any problems fitting a GeForce 6800 Ultra card in your case if you have enough room for the dual-slot GeForce FX 5950 Ultra.

One aspect that will concern most users however is power. The GeForce 6800 Ultra requires considerably more power than previous GeForce cards. In fact, NVIDIA recommends a 480-watt power supply for use with the Ultra board.

The main problem is that many power supplies don’t provide enough juice on the 12V rail. We detailed this in our power supply guide from last year. In order to provide some margin for compatibility with a wide variety of power supplies, NVIDIA set their guidelines at 480 watts, but we believe many good 430-watt power supplies provide enough power on the 12V rail to suffice for the 6800 Ultra. Of course, with processors getting faster and faster (and therefore requiring more voltage), it may not be a bad idea to go ahead and upgrade your power supply anyway as the power requirements to remain on the bleeding edge of technology are constantly going up.

Taking a closer look at the cooler itself, we see that NVIDIA has implemented a heat pipe cooling solution for GeForce 6800 Ultra. If you recall NVIDIA’s last heat pipe solution, the FX Flow cooler found on GeForce FX 5800 Ultra, you’re probably wondering how loud and how cool NVIDIA’s latest heat pipe cooler operates. Fortunately this new cooler is not a reprise of FX Flow. While the fan does operate in two modes, we never ran into a situation where the card’s fan ran at the higher mode other than during system bootup.

[image]
<% print_image("19"); %><% print_image("20"); %><% print_image("21"); %>

At the higher setting, the new cooler is loud, but not FX Flow loud. Whereas FX Flow’s motor contributed to a lot of the fan’s noise, this isn’t the case for the 6800 Ultra’s cooler. Instead the noise you hear is the whooshing sound of air moving at a very high velocity (I’ll update the article with MP3s once I get back from Toronto – Brandon). In other words, it’s a loud noise, but not nearly as annoying as FX Flow was.

The moderate setting is much more bearable. It’s still louder than RADEON 9800 XT and GeForce FX 5950 Ultra, but not loud enough to be distracting. The card’s fan never went above the moderate setting, even during extended testing sessions, and the 6800 Ultra board actually runs cooler than 5950 Ultra (a trait which really surprised us) so we don’t think the fan’s noise will become an issue unless you’re overclocking. We’ll provide more in-depth analysis of the new cooler and overclocking the board in a follow-up article.

[image]
<% print_image("22"); %><% print_image("23"); %>


SIDEBAR: We actually booted our GeForce 6800 Ultra board with one Molex connected on a 350W supply. It booted, but we didn’t playtest it.


Anti-aliasingPage:: ( 7 / 24 )

If you read one of our image quality shootouts from last summer, one of the chief weaknesses that NVIDIA cards have suffered from is anti-aliasing quality. ATI has demonstrated superior AA quality on numerous occasions.

NVIDIA has decided to address this deficiency in its GeForce 6 series by implementing a new rotated-grid anti-aliasing algorithm. Previous architectures relied on a two-by-two grid pattern for each pixel. With the GeForce 6 series, four samples are still taken, but now the pattern has been rotated into a diamond-shaped grid, rather than the square grid used in past architectures. This new grid pattern gives NVIDIA better coverage of the horizontal and vertical dimensions, but how does it look? First we’ll boot up 3DMark 03.

GeForce FX 5950 Ultra:

[image]

<% print_image("24"); %><% print_image("25"); %>

GeForce 6800 Ultra:

[image]
<% print_image("26"); %><% print_image("27"); %>

ATI RADEON 9800 XT:

[image]
<% print_image("28"); %><% print_image("29"); %>


GeForce FX 5950 Ultra


GeForce 6800 Ultra


RADEON 9800 XT


Notes

You can clearly see the difference NVIDIA’s new rotated-grid implementation makes in 3DMark 03, just focus on the leading edge of the wing of the B-17 bomber at the bottom of the scene. Deciding between GeForce 6800 Ultra and RADEON 9800 XT is a tough call however, both cards look very good. The tail on the bottom B-17 looks a little bit smoother on the RADEON 9800 XT, but you have to look very close to see it.

Lock On


Now lets move on to Lock On: Modern Air Combat.

GeForce FX 5950 Ultra:
[image]

<% print_image("30"); %><% print_image("31"); %>

GeForce 6800 Ultra:
[image]
<% print_image("32"); %><% print_image("33"); %>

RADEON 9800 XT:
[image]
<% print_image("34"); %>

Notes

It’s a bit harder to see the differences in this angle, we mainly took it because it was such a good looking screenshot! Let’s zoom in a bit anyway:


GeForce FX 5950 Ultra


GeForce 6800 Ultra


RADEON 9800 XT


Check out the horizontal stabilizer on this F-15C, specifically the left stabilizer, just beneath the “LN” markings on the tail. GeForce 6800 Ultra looks much better than 5950 Ultra here, although in this case we’ve got to give the AA edge to ATI. The RADEON 9800 XT does a better job of removing the jaggies, and doesn’t have the weird swirly marks found on the tail of the F-15C on both NVIDIA cards.

GeForce FX 5950 Ultra:
[image]
<% print_image("35"); %>

GeForce 6800 Ultra:
[image]
<% print_image("36"); %>

ATI RADEON 9800 XT:
[image]
<% print_image("37"); %>

In this batch of screenshots it’s a little easier to see the difference between GeForce FX 5950 Ultra and 6800 Ultra, just focus on the leader in front of the other three F-15Cs. Spotting a difference between the 6800 Ultra and RADEON 9800 XT continues to be a tough battle, but at the same time it’s also pretty obvious that the shadows on the lead F-15C are more intricate on the 9800 XT board than the 6800 Ultra (although this doesn’t deal with AA).



GeForce FX 5950 Ultra


GeForce 6800 Ultra


RADEON 9800 XT




SIDEBAR: AA levels up to 8x are currently available.


Anisotropic filteringPage:: ( 8 / 24 )

We wanted to start off by revisiting the last scene for a second:

[image]

<% print_image("38"); %><% print_image("39"); %>

As you can see in the captions, the first image is taken with anisotropic filtering disabled, while the second has it on. Note the fence behind the F-15s: many titles disable additional features when anisotropic filtering is disabled. This won’t affect game play, but it does ad an additional sense of realism that’s missed when AF is disabled.

Now lets take a look at screenshots from the same mission, only at a different angle:

GeForce FX 5950 Ultra:
[image]
<% print_image("40"); %><% print_image("41"); %>

GeForce 6800 Ultra:
[image]
<% print_image("42"); %><% print_image("43"); %>

RADEON 9800 XT:
[image]
<% print_image("44"); %><% print_image("45"); %>

Notes

Once you focus on the grass to the left of our F-15C in the above screenshot, you can richer textures on GeForce 6800 Ultra than RADEON 9800 XT or GeForce FX 5950 Ultra.



GeForce FX 5950 Ultra


GeForce 6800 Ultra


RADEON 9800 XT


FarCry


GeForce FX 5950 Ultra:

[image]

<% print_image("46"); %><% print_image("47"); %>

GeForce 6800 Ultra:

[image]
<% print_image("48"); %><% print_image("49"); %>

RADEON 9800 XT:

[image]
<% print_image("50"); %><% print_image("51"); %>

Notes

When we took this screenshot, we planned on using the wood to highlight the differences between the various cards with AF. However, we then quickly noticed that the wood, and even the sand behind it looked largely identical across all three cards. Distinctions between cards aren’t apparent until you look at the ridge on the left side of the scene:



GeForce FX 5950 Ultra


GeForce 6800 Ultra


RADEON 9800 XT


Here you can see the limitations of ATI’s AF implementation, at some angles AF isn’t properly applied. If you look at the region to the left of the tree in the above example, you’ll see the textures on the RADEON 9800 XT aren’t as rich as those on GeForce FX 5950 Ultra or GeForce 6800 Ultra.

Unreal Tournament 2004

Unfortunately, NVIDIA’s latest ForceWare driver continues to rely on brilinear filtering in UT 2004, despite the fact that an “application” setting has been available in the control panel for months now. Apparently this is one “feature” NVIDIA users will have to stick with for the immediate future. Anyway, on to the screenshots:

GeForce FX 5950 Ultra:

[image]

<% print_image("52"); %><% print_image("53"); %>

GeForce 6800 Ultra

[image]
<% print_image("54"); %><% print_image("55"); %>

RADEON 9800 XT

[image]
<% print_image("56"); %><% print_image("57"); %>

Notes

Like Far Cry, we decided to take our screenshots from the map we benchmark with, in this case DM 1-on-1 Albatross. As you can see however, GeForce 6800 Ultra doesn’t render the sky correctly – the clouds are completely missing! Hopefully this bug can be fixed in the next driver release.



SIDEBAR: The AF slider maxes out at 16x


Test conditionsPage:: ( 9 / 24 )

System Setup


AMD Athlon 64 3400+

ASUS K8V Deluxe

512MB OCZ EL PC3200 (DDR400) SDRAM

ASUS V9950 Ultra (GeForce FX 5900 Ultra)
eVGA e-GeForce FX 5950 Ultra
eVGA e-GeForce FX 5900 SE (5900 XT)
NVIDIA GeForce 6800 Ultra reference board
NVIDIA GeForce4 Ti 4600 reference board
Driver version ForceWare 60.72 [edit: 4pm, driver version typo]

ATI RADEON 9800 XT
ATI RADEON 9800 PRO 128MB
ATI RADEON 9700 PRO
Driver version CATALYST 4.4

30GB IBM Deskstar DTLA 307030 ATA/100 Hard Drive

Windows XP Professional SP1

DirectX 9.0b

Benchmarks

Lock On: Modern Air Combat (Mig-29 custom demo)
Call of Duty (demo0032 custom demo)
Quake III: Arena version 1.32 (fscrusher demo)
Unreal Tournament 2004 (T3 custom demo)
IL-2 Sturmovik: Forgotten Battles (The Black Death track)
Splinter Cell (FS custom demo)
Tomb Raider: Angel of Darkness (Beyond3D custom demo)
Halo: Combat Evolved (stock benchmark)
Far Cry
3Dmark 03



SIDEBAR: We weren’t given an ETA on when the new driver will be posted on NVIDIA’s site.


3DMark 03Page:: ( 10 / 24 )

3DMark 03 – Direct3D






3Dmark 03 – Game Test Results






Notes

Since we’re testing with unapproved drivers, we decided to limit 3DMark testing to NVIDIA cards only. As you can see, GeForce 6800 Ultra easily surpasses the performance of its predecessors, offering double the performance in shader tests like Battle of Proxycon and Troll’s Lair.



SIDEBAR: 3DMark was patched to 340.


3DMark 03Page:: ( 11 / 24 )

3DMark 03 – Direct3D






Notes

NVIDIA’s GeForce FX line have often been criticized for their PS2.0 performance, but if all indications with 3DMark 03 are accurate, it appears this won’t be a weak spot for GeForce 6800 Ultra. We tried to corroborate these findings with ShaderMark, but unfortunately it wouldn’t run with the 6800 Ultra board.




SIDEBAR: FutureMark is expected to release its follow-up to 3DMark 03 later this year.


Call of DutyPage:: ( 12 / 24 )

Call of Duty – OpenGL






Notes

We omitted testing at 800x600x32, where we’re CPU-limited with 6800 Ultra. In fact, we’re still CPU-bound at 1024x768, and to a lesser extent at 1280x1024 in Call of Duty. By 1600x1200 though the 6800 Ultra takes a commanding lead over the next closest competitor (RADEON 9800 XT) offering over 1.5 times the performance at that resolution.




SIDEBAR: Call of Duty was definitely my favorite game released last year.


IL-2 Sturmovik: Forgotten Battles 4xAAPage:: ( 13 / 24 )

IL-2 Sturmovik: FB - OpenGL






Notes

As you probably know by now if you’ve read a lot of our articles, IL-2 is largely CPU-bound, so we went ahead and ran our tests with AA enabled in order to put more strain on the graphics card. The GeForce 6800 Ultra performs well here, although it doesn’t have the commanding lead seen in previous benchmarks because of this. At 1600x1200, the 6800 Ultra outperforms 5950 Ultra by 28%.



SIDEBAR: Maddox recently announced the follow-up to IL-2, which will take place in the Pacific theater.


IL-2 Sturmovik: Forgotten Battles 4xAA 8xAFPage:: ( 14 / 24 )

IL-2 Sturmovik: FB - OpenGL






Notes

Under the greater demands of 4xAA and 8xAF, the GeForce 6800 Ultra is able to pull away from the other graphics cards a little more, nearly doubling performance of the next closest card, ATI’s RADEON 9800 XT.




SIDEBAR: Is it just me, or is the whole WW2 flight sim genre the slowest its been in awhile?


Quake III 4xAA 8xAFPage:: ( 15 / 24 )

Quake III - OpenGL





Notes

Once again we see GeForce 6800 Ultra delivering nearly twice the performance of today’s fastest video cards, RADEON 9800 XT and GeForce FX 5950 Ultra. Clearly in Quake 3 you can turn it all on with GeForce 6800 Ultra and still expect blistering performance.




SIDEBAR: Geometric detail was set to high in Quake 3.


Splinter CellPage:: ( 16 / 24 )

Splinter Cell – Direct3D






Notes

With its 16 pipeline architecture running at 400MHz, the GeForce 6800 Ultra blazes through Splinter Cell with relative ease. We’re CPU-limited at 800x600, 1024x768, and 1280x1024. Even at 1600x1200 we’re nearly pushing 100 frames per second.


SIDEBAR: Unfortunately it looks like Pandora Tomorrow doesn’t have built-in benchmarking, so we’re still using the original SC.


Tomb RaiderPage:: ( 17 / 24 )

Tomb Raider – Direct3D






Notes

Tomb Raider has been the Achilles Heel of previous GeForce architectures, but clearly that isn’t the case for GeForce 6800 Ultra, as the card was able to nearly deliver double the performance of RADEON 9800 XT.





SIDEBAR: We enabled depth of field for Tomb Raider testing (V49).


Lock On: Modern Air CombatPage:: ( 18 / 24 )

Lock On: Modern Air Combat – Direct3D






Notes

Lock On is a little more demanding than IL-2, so we ran this title without AA or AF enabled (after all, we didn’t want to wait for eternity for the GeForce4 results to finish at high resolutions with AA). Like the other tests we’ve run, GeForce 6800 Ultra drastically outperforms today’s latest cards.



SIDEBAR: If you didn’t check out the Lock On screens from the AA and AF sections, you really should go back and take a look at them.


UT 2004 4xAAPage:: ( 19 / 24 )

Unreal Tournament 2004







Notes

As you can see, we’re CPU-limited with GeForce 6800 Ultra in UT 2004 and 4xAA, the frame rate remains largely the same regardless of screen resolution. Lets see what happens when we crank on the AF though.



SIDEBAR: Map used on our custom UT2004 demo is 1on1 albatross.


UT 2004 4xAA/8xAFPage:: ( 20 / 24 )

Unreal Tournament 2004







Notes

We’re still CPU limited at 800x600 and 1024x768 with 4xAA and 8xAF on GeForce 6800 Ultra. By the time we hit the higher resolutions the 6800 Ultra is nearly offering double the performance of the other DX9 cards.


SIDEBAR: I like UT 2004 a lot more than UT 2003.


HaloPage:: ( 21 / 24 )

Halo







Notes

Since Halo still doesn’t provide custom benchmarking, we’re limiting our tests to NVIDIA cards only. Also keep in mind that the Halo benchmark is based on letterbox cutscenes, so the performance seen here isn’t indicative of real gameplay, the tests are more synthetic in nature.Perhaps it’s because of this that we’re able to see the GeForce 6800 Ultra truly double the performance over GeForce FX 5950 Ultra.




SIDEBAR: I can’t wait for Halo 2 later this year, I’ll finally have an excuse to dust off my Xbox.


FarCryPage:: ( 22 / 24 )

FarCry – Direct 3D







Notes

We created our own custom demo for testing with FarCry, Crytek’s latest shooter that takes advantage of 2.0 pixel and vertex shaders, and with the latest patch, 3.0 shaders as well. We also made sure that all GeForce FX cards were running the same 2.0/1.1 shader mixture as the GeForce 6800 and RADEON cards. All image settings were set at their highest. Since gunfire isn’t accurately reproducible in FarCry’s benchmarking mode, we simply ran through a level and recorded along the way.

Under this environment, the 6800 Ultra is CPU-bound at 800x600 and 1024x768. In fact, it doesn’t really begin to pull away from the other cards until 1600x1200, where its superior memory bandwidth really allows it to shine.





SIDEBAR: We patched Far Cry to version 1.1, which supports GeForce 6800 Ultra.


FarCry 4xAAPage:: ( 23 / 24 )

FarCry – Direct 3D










SIDEBAR: Since we don’t have DX9.0c, the GeForce 6800 Ultra used a mixture on 1.1 and 2.0 shaders in our testing.


ConclusionPage:: ( 24 / 24 )

So far, NVIDIA’s GeForce 6800 Ultra is shaping up to be a real winner. NVIDIA is the first to adopt shader model 3.0 with GeForce 6800 Ultra, and have already lined up the following titles:

Lord of the Rings, Battle For Middle-earth
STALKER: Shadows of Chernobyl
Vampire: Bloodlines
Splinter Cell X
Tiger Woods 2005
Madden 2005
Driver 3
Grafan
Painkiller
FarCry

It remains to be seen how far some of the developers on these titles will go with their 3.0 implementations, but it’s an interesting development nonetheless, as the first shader model 3.0 titles will hit the market much faster than its predecessors. Technically, the first title, FarCry, is already shipping (although Microsoft hasn’t released DX9.0c yet).

NVIDIA has also addressed their anti-aliasing quality by adding a new rotated-grid anti-aliasing sampling pattern. We definitely saw the IQ improvements in our AA testing.

Clearly when it comes to performance GeForce 6800 Ultra is in a class of its own. In many cases GeForce 6800 Ultra outperforms its nearest competitor by a factor of nearly two, and keep in mind that this is with early drivers, performance will only get better from here.

There is, however, reason to reserve judgment. ATI has yet to step up to the plate with R420. Given their rather remarkable silence over the past few months, a silence not unlike the one preceding the Radeon 9700 line, it may be that they can surprise us. Perhaps they're holding off on final specifications until they see these GeForce 6800 Ultra numbers and know what they have to beat, or perhaps they're not sure if R420 can deliver. We'll soon find out, maybe even as early as today, which marks ATI's second Shader Day event.

Another factor that may play against GeForce 6800 Ultra is its extraordinary power requirements. With its two Molex power connector requirement, and NVIDIA’s recommendation of a 480-watt power supply, many enthusiasts will have to upgrade their power supply unit in order to run GeForce 6800 Ultra. This could make a whole lot of money for PSU manufacturers like Antec and Enermax, or it could lead enthusiasts to settle for GeForce 6800. We’ll just have to wait and see how that part plays out.

You won’t have to wait much longer, fortunately. The first GeForce 6800 and 6800 Ultra boards should be hitting retail a little over a month from now. Expect to see us cover the first wave of GeForce 6800 boards as soon as they’re available!

Discuss the 6800 Ultra, ATI's possible responses and anything else about the article at our forums!




SIDEBAR: Are you concerned about GeForce 6800 Ultra’s power requirements, or does the $500 price tag more of an issue? What do you think of NVIDIA’s new AA? Speak up in the news comments!

© Copyright 2003 FS Media, Inc.
[ Print Article! | Close Window ]