New Family of Low-Cost NVIDIA Chips: GeForce FX 5600/5200 Review

by Tim Tscheblockov
04/16/2003 | 11:50 PM

It is no secret that the graphics chip and graphics card manufacturers receive the biggest revenue not from the most expensive and high-performance solutions, but from the Value and mainstream products. This is not at all surprising, I should say: only dedicated enthusiasts would make up their mind to purchase the fastest and most expensive graphics accelerator. All other users will weigh all cons and pros and finally decide on a better value solution, no doubt.

The most powerful today’s chips offered by the industry leaders, ATI and NVIDIA, are R300/R350 (or ATI RADEON 9700 Pro/9800 Pro) and NV30 (or GeForce FX 5800/5800 Ultra). Graphics cards based on these chips are very expensive that is why the majority of users are not very much interested in them.

A little less expensive DirectX9-compliant solutions used to be offered only by ATI for a while: these were the products based on the cut-down version of R300 chips aka RADEON 9500 and 9500 Pro. They didn’t cost as much as the “fully-fledged” RADEON 9700/9700 Pro solutions, but they are still stably above the sacred “$100-bar”.

The recent announcement of the new NVIDIA GeForce FX 5600/5200 family is expected to change this situation drastically. The GeForce FX 5600 Ultra based graphics cards are expected to cost around $199. However, the solutions based on the most inexpensive FX chip, on GeForce FX 5200, should overcome the remarkable $100-limit and settle around $79.

There will be 4 graphics card modifications altogether:

• NVIDIA GeForce FX 5600 Ultra. This solution will be based on NV31 graphics core. The chip and the graphics memory will work at 350MHz and 700MHz (350MHz DDR) respectively. The chip will support 128bit DDR SDRAM memory bus. This is the fastest and most expensive solution of the four newcomers, which is positioned as a replacement to the GeForce4 Titanium family.
• NVIDIA GeForce FX 5600. This product will be built on NV31 core working at a lower frequency than that of the previously described chip. The graphics memory will be also somewhat slower than in the previous case, but the exact values haven’t been disclosed yet. The chip will support 128bit DDR SDRAM memory bus. NVIDIA GeForce FX 5600 are expected to replace NVIDIA GeForce4 Ti4200 and Ti4200-8x solutions.
• NVIDIA GeForce FX 5200 Ultra. The card is built on NV34 graphics core working at 325MHz. The memory will support 650MHz frequency (325MHz DDR) and 128bit DDR SDRAM bus.
• NVIDIA GeForce FX 5200. Here the same NV34 graphics core is used. The working frequencies make 250MHz for the core and 400MHz (200MHz DDR) for the memory. The memory bus is the same 128bit DDR SDRAM. NVIDIA GeForce FX 5200 and GeForce FX 5200 Ultra will replace NVIDIA GeForce4 MX440-8x.

So, the new FX’s from NVIDIA started their invasion into the mass market. But what do the new chips actually look like?

NVIDIA GeForce FX 5600/5200, or NV31/NV34 according to NVIDIA’s internal marking, inherit the NV30 ideology and architecture, but feature lower performance than the predecessor. In other words, they feature fewer functional units or are simply “cut down”.

So, the major remarkable feature of the new better value chips from NVIDIA is the retained functionality of the “fully-fledged” forefather, NV30. In other words, these chips comply with DirectX9 specification, support new fast anisotropic filtering method, support full multisampling providing high performance with enabled full-screen anti-aliasing.

Some of the features of the new NV31/NV34, however, are worth discussing in a bit greater detail (they are marked with comment numbers):

[1]: NV31 and NV34 chips, just like their forefather, NV30, feature very flexible architecture, which is capable of creating different number of pixel pipelines (NV30 architecture was discussed in great detail in our article called “NVIDIA GeForce FX 5800 Ultra Review: New Technologies and Performance”). New NVIDIA chipsets have been simplified compared with NV30. But if the “youngest” of the two newcomers, NV34, doesn’t have anything new about the pixel pipelines, and represents a kind of “half of NV30” from this point of view, then the “elder” NV31 can boast new features. NV31 architecture allows two versions of internal organization: 2 pipelines with 2 texturing units per pipeline (this modification is also available for NV34) and 4 pipelines with 1 texturing unit per pipeline. The latter modification cannot be applied to NV34, because with 4 pixel “pipelines” (just like NV30 with 8 “pipelines”) cannot perform the texturing and write the color value into the frame buffer.

NV31 with its progressive architecture will evidently boast an advantage over NV34, because it will be able to select the most suitable internal organization for each particular case. For example, in Quake3 Arena the use of two pixel “pipelines” will be more helpful for most surfaces featuring two textures (base texture + lightmap), while all sorts of explosions or rocket tracks will be drawn more efficiently with 4 pipelines involved.

It looks as if this particular internal architecture had been initially implied for NV30, but for some reason they failed to implement it completely, and only in NV31 NVIDIA managed to put this idea fully into life.

The results comparison for NV31 and NV34 in real applications will show, how efficient is this way of building the chip architecture.

[2]: Pixel and Vertex pipelines of NV34 boast limited features compared with NV31 and NV30. Nevertheless, NV34 meets the minimal Microsoft DirectX9 requirements on the hardware level.

[3]: Unlike NV31 and NV30, NV34 doesn’t support frame buffer compression. The absence of frame buffer support in NVIDIA GeForce FX 5200/5200 Ultra may seriously affect the performance of this solution with enabled full-screen anti-aliasing.

Frame buffer compression is used during full-screen anti-aliasing by means of multisampling (you can find more details on multisampling and supersampling in our article called “On the Way to Ideal Picture: Anti-Aliasing by Contemporary Graphics Cards”). Bearing in mind that during multisampling all pixels located not on the polygon edges feature subpixels of the same color, the chips can write into the frame buffer only one color value. Unfortunately, this technique cannot be applied to pixels located on the polygon edges, because in this case each subpixel can be of a different color. However, these pixels are not very numerous, so that the overall data volume still gets considerably reduced. For instance, in case of 4x multisampling the data transferred to the frame buffer can get down to ¼ of the initial amount.

Well, it’s high time we moved over to the real products and real tests.

In this review we are going to take a closer look at the reference graphics cards from NVIDIA based on GeForce FX 5600 Ultra and GeForce FX 5200 Ultra, and one of the first mass graphics cards built on NVIDIA GeForce FX 5200. The today’s heroes will be compared with their predecessors, namely NVIDIA GeForce4 Ti4200-8x and MX440-8x. Also we will take a few solutions from NVIDIA’s direct rival, ATI: RADEON 9500 Pro as a competitor to NVIDIA GeForce FX 5600 Ultra and RADEON 9000 Pro as a competitor to NVIDIA GeForce FX 5200/5200 Ultra.

Closer Look: NVIDIA GeForce FX 5600 Ultra and GeForce FX 5200 Ultra

Two reference graphics cards based on NVIDIA GeForce FX 5600 Ultra and GeForce FX 5200 Ultra feature absolutely identical PCB design, so that it is absolutely impossible to distinguish between them at first glance. Therefore, I will offer you only one pair of pictures: NVIDIA GeForce FX 5600 Ultra:

Just like NVIDIA GeForce FX 5800 Ultra, the new cards feature a special power supply connector, even though the new chips consume less power:

Both graphics cards feature 128MB of DDR SDRAM memory in BGA-chips from Hynix with 2.5ns clock time:

The only thing different about the cards design is the graphics chips: NVIDIA GeForce FX 5600 Ultra features a chip marked as NV31, and GeForce FX 5200 Ultra – as NV34:

The clock frequencies of the NVIDIA GeForce FX 5600 Ultra are 350MHz for the chip and 700MHz (350MHz DDR) for the graphics memory. NVIDIA GeForce FX 5200 Ultra works at slightly lower frequencies: 325MHz chip and 650MHz (325MHz DDR) memory.

It looks as if the PCB design of the mass graphics cards based on NVIDIA GeForce FX 5600 Ultra and GeForce FX 5200 Ultra will be different from the reference, because it doesn’t make much sense to use the same design for solutions working at different frequencies and selling at pretty different prices. And the first mass graphics card from Albatron, which we had at our disposal, is clear evidence to that.

Closer Look: Albatron GeForce FX 5200

The mass graphics card from Albatron doesn’t have anything in common with NVIDIA’s reference design. It looks more like one of the numerous GeForce4 MX440 based solutions. Take a look yourselves:

Albatron graphics card is equipped with 128MB of 128bit DDR SDRAM memory in Samsung chips with 5ns clock time:

The working frequencies of this solution are exactly as recommended by NVIDIA: 250MHz for the chip and 400MHz (200MHz DDR) for the memory, that is considerably lower than those of an Ultra version of the card. This significant reduction of working frequencies resulted into a much lower power consumption. So, the chip can be cooled with a pretty small (according to today’s parameters) cooler and there is no additional power supply connector on the PCB.

Drivers

The new NVIDIA based graphics cards appeared in our test laboratory almost simultaneously with the new 43.45 driver. The new driver version differs a little bit from 42.68 driver version, which we used to test NVIDIA GeForce FX 5800 Ultra (see our NVIDIA GeForce FX 5800 Ultra Review).

First of all, there appeared new full-screen anti-aliasing modes for NVIDIA GeForce FX 5800 Ultra: 8x and 16x. These modes using supersampling are also available for NVIDIA GeForce FX 5600 Ultra:

The second interesting innovation in the new 43.45 driver, which you can see on the screenshot above is the “new” Quality mode, which used to be called Balanced. Not bad, don’t you think so?

You know that the best image quality can be obtained on NVIDIA GeForce FX chips family in Application mode, where all the optimizations are disabled completely. Then comes a faster Balanced mode with slightly worse image quality (now it is called Quality). The last mode, which is the fastest, but provides the lowest quality of all three is Aggressive mode (now it is called Performance).

In other words, if we call these modes according to the speed and image quality provided, it would be more logical to call them Quality, Balanced and Performance, accordingly.

This is no secret to us why NVIDIA had to rename these work modes, giving up the common sense and logic. When you see the Quality mode in the control panel (which is used by default, by the way), the users and reviewers will be sure that this mode provides the best quality. In reality the cards will provide not the best image quality, but will get a performance boost compared with the performance in Application mode instead. As a result, they will look better when compared to competing solutions.

Well, I don’t like to be cheated on, do you?

But these are not all the surprises prepared by the “new” Quality mode. We will continue our investigation a bit later, when we will dwell on full-screen anti-aliasing quality provided by the new graphics solutions.

Testbed and Methods

We used the following testbed for our test session:

• Intel Pentium 4 2800MHz CPU;
• ASUS P4S8X (SiS648) mainboard;
• 512MB DDR400 SDRAM;
• IBM DTLA 305030 30GB HDD.

The software set included:

• Microsoft Windows XP Professional;
• Microsoft DirectX 9.0;
• SiS AGP v. 114 driver;
• Detonator 43.45 for NVIDIA based graphics cards;
• Catalyst 3.1 for ATI based graphics cards.

Our today’s heroes will be tested against the following rivals:

• NVIDIA GeForce4 Ti4200-8x 128MB, 250MHz chip / 512MHz (256MHz DDR) memory;
• NVIDIA GeForce4 MX440-8x 128MB, 275Mhz chip / 512MHz (256MHz DDR) memory;
• ATI RADEON 9500 Pro 128MB, 275MHz chip / 540MHz (270MHz DDR) memory;
• ATI RADEON 9000 Pro 128MB, 275MHz chip/ 500MHz (250MHz DDR) memory.

Performance: Synthetic Benchmarks

Fillrate

At first we will take a look at the Polygon Fillrate test from 3DMark 2001 SE test set. To evaluate the efficiency of technologies intended to optimize the use of memory bus bandwidth, we ran this test with enabled FSAA.

The results shown by NVIDIA GeForce FX 5600 Ultra and ATI RADEON 9500 Pro with enabled FSAA do not drop down as greatly as the results of other testing participants. This can be explained by the fact that only NVIDIA GeForce FX 5600 Ultra and ATI RADEON 9500 Pro support frame buffer compression.

With enabled multi-texturing the NVIDIA GeForce FX solutions and ATI RADEON 9500 Pro lay 8 textures per pass, unlike NVIDIA GeForce4 Ti4200-8x, MX440-8x and ATI RADEON 9000 Pro. As a result, they have to address the Z-buffer and frame buffer more rarely, so that enabled full-screen anti-aliasing hardly tells on their performance.

The fillrate tests from 3DMark 2001 can hardly allow us to draw final conclusions about the advantages of the NVIDIA GeForce FX internal architecture. We decided to make up for the missing information about the way NVIDIA GeForce FX organizes its architecture with the help of a small program which draws a surface with 0-4 textures laid upon it. To make the picture complete, I also added the results for NVIDIA GeForce FX 5800 Ultra (NV30):

NVIDIA GeForce FX 5800 Ultra and GeForce FX 5200 Ultra act very similarly, which proves the supposition about their identical internal architecture, with that only difference that GeForce FX 5200 Ultra feature only 2 pixel “pipelines” with two TMUs on each.

However, GeForce FX 5600 Ultra (NV31) stands out here: when there is only one texture laid, this chip shows better results than the theoretical maximum for configurations with two pixel pipelines. Of course, with only one texture processed NV31 uses the mode with 4 pixel “pipelines” and 1 TMU per pipeline. In case there are two or more textures to be processed NV31 resorts to a different internal organization scheme: 2 pixel “pipelines” with two TMUs per pipeline.

This flexibility should offer GeForce FX 5600 Ultra a significant advantage over NVIDIA GeForce FX 5200 Ultra even if they work at equal clock frequencies. In most today’s games there surfaces with a single texturing layer as well as surfaces with two and more textures. Note that the share of single-texture surfaces is not that small at all. In those cases when there is only one texture used on a certain surface, such as grass, tree leaves, various explosions, flames, smoke and clouds, NVIDIA GeForce FX 5600 Ultra will work twice as fast.

Pixel Shaders

To test the pixel shaders performance we resorted to benchmarks from 3DMark2001 and 3DMArk2003 test sets:

The results of NVIDIA GeForce FX 5200/5200 Ultra are not so optimistic: GeForce4 Ti4200-8x and ATI RADEON 9500 Pro managed to outperform them. It looks as if the pixel shaders implementation and integer calculations by NV34 remained absolutely the same as by NV30.

However, the “eldest” newcomer, NVIDIA GeForce FX 5600 Ultra proved impressively fast here, having left NVIDIA GeForce FX 5200 Ultra far behind and almost caught up with the 8-pipeline ATI RADEON 9500 Pro. Well, the integer part of the NV31 pixel shaders seems to have got significantly improved since the times of NV30.

In case of a more complex pixel shader, new NVIDIA chips give in: GeForce4 Ti4200-8x and ATI RADEON 9000 Pro surpass them quite significantly. Not to mention ATI RADEON 9500 Pro...

In case of floating point DirectX9 pixel shader ATI RADEON 9500 Pro appears far ahead of the new GeForce FX solutions. Nevertheless, NVIDIA GeForce FX 5600 Ultra proves almost twice as fast as GeForce FX 5200 Ultra.

T&L and Vertex Shaders

At first let’s check the high Polygon Count test from 3DMark2001 SE package:

“cut-down” vertex shaders of the new NVIDIA chips show much lower results than NVIDIA GeForce FX 5800 Ultra. So, despite higher working frequencies NV31 and NV34 get completely defeated by GeForce4 Ti4200-8x and ATI RADEON 9500 Pro even during Fixed Function T&L processing.

With DirectX8 vertex shaders the situation is just the same: this time the newcomers were beaten even by ATI RADEON 9000 Pro.

The results remained the same even with DirectX9 vertex shaders.

So, let’s try to draw some preliminary conclusions basing on the results in synthetic benchmarks.

First, it is evident that the difference between NV31 and NV34 is not only in clock frequencies and size of texturing, vertex and other caches. GeForce FX 5600/5600 Ultra, just like GeForce FX 5200/5200 Ultra inherited NV30 architecture, but at the same time acquired a few important improvements implying faster pixel shaders processing as well as fully-fledged flexible support of the work modes with two or four pixel “pipelines”.

Besides that, NV31 is the only one of the two NV30 successors that retained the frame buffer compression during FSAA, while NV34 received the older implementation of anti-aliasing used in NVIDIA GeForce4 Ti family.

And finally, the vertex shaders performance of NVIDIA GeForce FX 5600/5600 Ultra and 5200/5200 Ultra appeared much lower than by NVIDIA GeForce FX 5800 Ultra.

By the way, when ATI Company designed their “cut-down” RADEON 9500/9500 pro chips, they simply reduced the number of pixel pipelines, having left the number of vertex processors unchanged.

Well, the most exciting part of our review comes now. We will be evaluating the performance and image quality provided by the new NVIDIA solutions with full-screen anti-aliasing and in games.

Performance: Anisotropic Filtering and Full-Screen Anti-Aliasing: Speed and Quality

As for anisotropic filtering, NV31 and NV34 perform absolutely the same as their predecessor, NV30. that is why I will not provide any screenshots and performance drop measurements here. If you are looking for more information on that, please, see our NVIDIA GeForce FX 5800 Ultra Review.

The FSAA modes by NVIDIA GeForce FX 5600 Ultra are the same as those of GeForce FX 5800 Ultra, so there are no surprises here, too.

The most exciting things start happening when we turn to NVIDIA GeForce FX 5200/5200 Ultra. Just like GeForce4 chips, they support FSAA 2x, Quincunx and 4x with the help of multisampling. Besides that, you can also get 4xS in Direct3D, which is a combination of supersampling and multisampling. The chips ensure adequate anti-aliasing quality in Application mode, but as soon as you shift to Quality or Performance mode, the picture quality in OpenGL games gets significantly worse. The worsening is represented by slight trembling of the polygon edges if the polygons are close to horizontal. This trembling effect becomes even more noticeably when the resolution gets lower: 1024x768, 800x600, 640x480.

Having analyzed the image quality, we can conclude that in OpenGL with 4s FSAA enabled in Quality and Performance modes NVIDIA GeForce FX 5200/5200 Ultra uses only two subpixels instead of four to calculate the pixel color. Moreover, in case of an even frame two upper subpixels are taken from the 2x2 block, while in case of odd frames – two lower subpixels. This is exactly where this unpleasant lines trembling effect comes from. However in high resolutions with high refresh rate or in case of an LCD monitor with high inertia, this trembling is hardly noticeable. So, even though 2s anti-aliasing is involved, you get the impression that there is fully-fledged anti-aliasing based on 4 subpixels.

It is remarkable that when you run the applications in windows, the real 4x method is applied. But when you take a screenshot the picture again looks as if the true 4x anti-aliasing were used. That is why I failed to take any screenshots proving my point, so I suggest that you take a quick look at a self-made reconstruction with the help of an animated picture:

And below you can see a picture with real 4x anti-aliasing:

Of all graphics cards based on new NVIDIA chips only GeForce FX 5600 Ultra doesn’t have this 4x anti-aliasing optimization. All other graphics cards including NVIDIA GeForce FX 5200/5200 Ultra, GeForce4 Ti4200-8x, GeForce4 MX440-8x demonstrated the work of an “optimized” 4x anti-aliasing algorithm with 43.45 driver. As I have already said, this annoying trembling can be best seen in low resolutions. Moreover, by GeForce4 MX440-8x these are not only the polygon edges that tremble, but the entire image.

This optimization has the right to exist and is absolutely justified: during our test session we discovered that it worked on those cards, which didn’t support frame buffer compression during full-screen anti-aliasing. But why did they call the mode when this optimization gets activated, “Quality”? it would be much more logical to call Quality the current Application mode, where the cards do honestly perform 4x full-screen anti-aliasing.

Well, let’s now estimate the performance difference shown by graphics cards in different quality modes. We will use Quake3 Arena game:

Without FSAA and anisotropic filtering, the performance difference in various quality modes is not that big...

But as soon as we enable full-screen anti-aliasing, the performance difference starts growing. You can clearly see how greatly the performance dropped in Application mode enabled by NVIDIA GeForce FX 5200 Ultra: the chip has to perform true 4x full-screen anti-aliasing.

With anisotropic filtering enabled, the performance difference also grows bigger. The Application mode appears the hardest for all chips, because there they have to perform true tri-linear filtering instead of a mixture with bi-linear filtering.

In the hardest testing conditions with enabled anisotropic filtering and full-screen anti-aliasing, the graphics cards in Application mode appear up to 30-40% slower than they were in Performance mode.

So, if we take into account the quality and fastness of anisotropic filtering and full-screen anti-aliasing, we can say that new NVIDIA chips show practically the same results as NVIDIA GeForce FX 5800 Ultra.

We can once again praise the new NV31/NV34 chips for the same anisotropic filtering method as by NV30, which ensure high image quality at the expense of pretty low performance losses.

As for FSAA, things are not so smooth here. The brand new “optimization “ of the 4x method may discourage you so greatly in OpenGL applications that you will never use the Quality mode again. Especially, since in the Application mode, the cards still run acceptably fast due to well-implemented multisampling support.

Performance: Gaming Tests

Quake3 Arena

As usual, let’s start with Quake3 Arena. Settings look as follows: 32-bit texture color and frame-buffer depths, maximum amount of textures and objects, tri-linear filtering is on, texture compression is off.

All graphics cards were tested in default modes. In other words, NVIDIA based graphics cards were tested in Quality mode (former Balanced mode), and ATI based cards – in Balanced mode.

NVIDIA GeForce FX 5600 Ultra proved just excellent here, having outperformed GeForce4 Ti4200-8x, and in 800x600 and 1024x768 it even surpassed ATI RADEON 9500 Pro.

NVIDIA GeForce FX 5200 Ultra reference graphics card fell behind GeForce FX 5600 Ultra more than we had expected, even taking into account the difference in the cards working frequencies. The reason for this severe lag is the fact that NV34 doesn’t support the internal organization with 4 pixel pipelines and 1 TMU per pipeline.

Nevertheless, GeForce FX 5200 Ultra managed to easily outpace GeForce4 MX440-8x as well as the ATI RADEON 9000 Pro competitor. A slower version of NVIDIA NV34 based solution from Albatron has no evident advantage over these solutions and performs almost as fast as GeForce4 MX440-8x and RADEON 9000 Pro.

Maximum resolution supported by ATI RADEON 9000 Pro in SMOOTHVISION 4x Quality mode (which is exactly the mode we tested this graphics card in) makes 1024x768. That is why we do not see any results of ATI RADEON 9000 pro in higher resolutions. Anyway, even in available modes the card failed to compete on equal terms with the other testing participants: unlike other chips, RADEON 9000 Pro supports only slow outdated supersampling algorithm.

NVIDIA GeForce4 Ti4200-8x and GeForce4 MX440-8x graphics cards with 43.45 driver version displayed just an empty screen when we enabled full-screen anti-aliasing in 1600x1200. That is why these results are also not on the diagram.

As for everything else, there were no problems, and new NVIDIA chips performed excellently well. The fastest NVIDIA GeForce FX 5600 Ultra, easily defeated GeForce4 Ti4200-8x and fell just a tiny bit behind ATI RADEON 9500 Pro, while the youngest member of the family, GeForce FX 5200 Ultra, got ahead of GeForce4 MX440-8x in no time.

But you shouldn’t forget that only GeForce FX 5600 Ultra used “true” 4x anti-aliasing. All other NVIDIA chips used an optimization with trembling polygon edges, described above.

For ATI RADEON 9500 Pro we used Performance anisotropic filtering mode, and for NVIDIA chips – Quality mode (formerly known as Balanced, so there are no discrepancies in the image quality). With these particular settings NVIDIA GeForce FX chips and ATI RADEON 9500 Pro do not use true tri-linear filtering together with anisotropic filtering.

As a result, new GeForce FX chips managed to speed up quite a bit. And though NVIDIA GeForce FX 5200/5200 Ultra didn’t beat GeForce4 Ti4200-8x, then GeForce FX 5600 Ultra left the latter quite far behind.

The leader here remains ATI RADEON 9500 Pro, which owes its victory to very fast anisotropic filtering algorithms in Performance mode.

NVIDIA GeForce4 MX440-8x solution didn’t participate in this benchmark: the maximum anisotropy level supported by GeForce4 MX440-8x chip is equal to 2.

In the heaviest mode the leadership of ATI RADEON 9500 pro remains unchanged.

The new NVIDIA GeForce FX chips perform quite well here, and if in 800x600 resolution the fastest newcomer, NVIDIA GeForce FX 5600 Ultra, is just a little ahead of GeForce4 Ti4200-8x, then in 1280x1024 even GeForce FX 5200 has almost managed to beat the latter.

Serious Sam: The Second Encounter

The next game using OpenGL is Serious Sam: The Second Encounter. We tested with 32bit color and also ran a standard “GFX:Extreme Quality” add-on enabling maximum image quality settings, including maximum supported anisotropic filtering. Just like in Quake3 Arena, we used Performance mode for the anisotropic filtering by ATI RADEON 9500 Pro, and Quality mode for anisotropic filtering by NVIDIA chips.

Well, here the new NVIDIA solutions do not look in the best way, I assume. However, let’s dwell on the results and discuss them in greater detail. First, NVIDIA GeForce4 MX440-8x can be disregarded from the very beginning, as this chip supports only up to anisotropy level=2 and ensures a considerably lower texture filtering quality than the other testing participants. Second, we can disregard ATI RADEON 9000 Pro results. This chip supports up to 16x anisotropic filtering, but the filtering quality provided is hardly comparable with what we see by GeForce FX solutions. So, what do we have left? ATI RADEON 9500 Prom which boasts a much better anisotropic filtering quality even in Performance mode, NVIDIA GeForce4 Ti 4200-8x and GeForce FX based solutions.

No wonder that ATI RADEON 9500 Pro appears faster than the new NVIDIA GeForce FX chips. However, the fastest of the new chips, GeForce FX 5600 Ultra, which is expected to replace GeForce4 Ti, shows much better results than GeForce4 Ti4200-8x solution.

With enabled full-screen anti-aliasing the situation changes: GeForce FX 5600 Ultra easily beats GeForce4 Ti4200-8x, while GeForce FX 5200 Ultra and GeForce FX 5200 turn out faster than ATI RADEON 9000 Pro an d GeForce4 MX440-8x.

ATI RADEON 9500 Pro remained too far ahead. The new NVIDIA GeForce FX couldn’t even get close to it.

Unreal Tournament 2003

We tested with the following game settings: Texture Detail: Highest, World Detail: Highest, Character Detail: Highest, Physics Detail: Normal, Character Shadows: ON, Dynamic Lighting: ON, Detail Textures: ON, Projectors: ON, Decals: ON, Coronas: ON, Decal Stay: Normal, Foliage: ON, Tri-linear Filtering: ON. We tested in Flyby-scene from Antalus.

The graphics cards were tested in default modes.

In Unreal Tournament 2003 the new NVIDIA GeForce FX chips performed slightly worse than in OpenGL games. The eldest solution of the newcomers, GeForce FX 5600 Ultra, only in high resolutions managed to get a little ahead of GeForce4 Ti4200-8x. The slowest of the new cards, GeForce FX 5200, in its turn appeared a little faster than GeForce4 MX440-8x.

It is remarkable that in Unreal Tournament 2003, just like in OpenGL games, a more flexible GeForce FX 5600 Ultra proved much faster than GeForce FX 5200 Ultra, with only 10% higher working frequency.

If we compare the newcomers’ performance with the competing solutions from ATI, new NVIDIA chips will look twofold: on the one hand, ATI RADEON 9500 Pro outperforms GeForce FX 5600 Ultra, but on the other, ATI RADEON 9000 Pro is unable to compete even with GeForce FX 5200.

With enabled full-screen anti-aliasing NVIDIA GEForce4 Ti4200-8x without frame buffer compression support, and ATI RADEON 9000 Pro and GeForce4 MX440-8x using supersampling method, draw back quite s lot, which allows even NVIDIA GeForce FX 5200 Ultra to almost catch up with GeForce4 Ti4200-8x.

The slowest of the newcomers, NVIDIA GeForce FX 5200 from Albatron, turns over 1.5 times faster than the most probably rivals, NVIDIA GeForce4 MX440-8x and ATI RADEON 9000 Pro, due to multisampling support.

ATI RADEON 9500 Pro is again not even within reach for the new NVIDIA solutions.

When we enabled anisotropic filtering, NVIDIA GeForce MX440-8x leaves the race, because it doesn’t support 8x anisotropy.

Among those graphics cards left, the evident victory was won by ATI RADEON 9500 Pro in Performance anisotropic filtering mode, and RADEON 9000 Pro completely defeated NVIDIA GeForce FX 5200/5200 Ultra as well as GeForce4 Ti4200-8x. However, these results of RADEON 9000 Pro appeared incorrect: the numbers indicate that forced anisotropic filtering simply didn’t work in Unreal Tournament 2003.

NVIDIA GeForce4 Ti4200-8x, which doesn’t support the new “fast” anisotropic filtering algorithm, draws back in this test quite noticeably, so that in 1600x1200 even GeForce FX 5200 managed to beat it.

In the heaviest mode with enabled full-scene anti-aliasing and anisotropic filtering at the same time, the results remain almost the same, with that only exception that ATI RADEON 9000 Pro using slower supersampling rolls down to the last rating position.

Codecult Codecreatures

Here the new NVIDIA GeForce FX based graphics cards perform well enough against the background of their predecessors and ATI RADEON 9000 Pro. But they still can’t really compete with ATI RADEON 9500 Pro with its 8 pixel pipelines and much faster vertex shaders unit.

With enabled FSAA and anisotropic filtering the gap between ATI RADEON 9500 Pro and NVIDIA GeForce FX 5600 Ultra gets somewhat smaller, but at the same time NVIDIA GeForce FX 5200/5200 Ultra slow down quite significantly, because they do not support frame buffer compression. This way GeForce4 Ti4200-8x manages to outrun them.

3DMark2001 SE Gaming Benchmarks

3DMArk 2001 gaming benchmarks were run in Low Detail mode, because in High Detail mode the results are very often limited by the CPU performance, which makes the further graphics card comparison an absolutely useless thing.

The tests were run with 32bit texture color and frame-buffer depths; Z-buffer color depth equaled 24 bit. The settings of the graphics cards were left by default.

The results shown by the new NVIDIA GeForce FX based graphics cards are all in all the same as in real games.

Conclusion

So, what could we say about the new NVIDIA GeForce FX graphics chips family?

At first we would like to make a few traditionally pessimistic statements about DirectX9.

Of course, the appearance of low-cost graphics cards supporting DirectX9 should speed up the coming of contemporary 3D technologies into masses. However, most today’s graphics cards used in home systems hardly have even DirectX8 support. And it means that new products with fully-fledged DirectX9 support will come not that soon yet: will you profit a lot from selling a game, which doesn’t run on most of the today’s systems?

However, new GeForce FX family is intended to improve the situation.

First, the performance of graphics cards based on new NVIDIA GeForce FX chips is generally higher compared with the performance of the previous generation graphics solutions. NVIDIA GeForce FX 5600 Ultra is faster than GeForce4 Ti4200-8x, and GeForce FX 5200 Ultra (even GeForce FX 5200) are faster than GeForce4 MX440-8x.

Second, all new chips inherited optimized anisotropic filtering algorithm from NV30, which allows them to achieve much better  results in those applications where anisotropic filtering is used. Definitely, this is an advantage over their predecessors.

Third, GeForce FX 5600/5600 Ultra chips intended to replace GeForce4 Ti family, support frame buffer compression and better-quality full-screen anti-aliasing methods. GeForce FX 5200/5200 Ultra expected to replace GeForce4 MX family can now support fully-fledged multisampling. These improvements allow the new graphics chips to show their best during full-screen anti-aliasing.

At last, the prices of the new graphics solutions based on the freshly announced NV31 and NV34 chips promise to be very attractive. If GeForce FX 5200/5200 Ultra based graphics cards will sell at about the same or a little higher price than the today’s GeForce4 MX440-8x, then I don’t think you will find it hard to make a decision about the graphics solution. Especially since the competing solution from ATI, RADEON 9000 Pro, is slower than GeForce FX 5200/5200 Ultra in many tests.

No wonder that graphics card makers are especially excited about the low-cost GeForce FX 5200 graphics chips. One of the first mass graphics cards based on NVIDIA GeForce 5200 chip, which we managed to test, a solution from Albatron, is a real proof to the point. This graphics card combines not very high price with very high quality, good performance and contemporary 3D technologies support. Of course, GeForce FX 5200 based graphics cards, just like Albatron’s solution, will feature the same power consumption and heat dissipation, alongside with low noise level. These qualities will make graphics cards like Albatron GeForce FX 5200 an excellent choice for low-cost home gaming systems and as a worthy replacement to morally outdated ATI RADEON 9000 Pro and NVIDIA GeForce4 MX440.

As for the more expensive NVIDIA GeForce FX 5600/5600 Ultra chips, we can’t draw any final conclusions yet. On the one hand, GeForce FX 5600/5600 Ultra feature a lot of advantages over GeForce4 Titanium family, such as faster anti-aliasing, faster anisotropic filtering, DirectX9 support. On the other hand, ATI RADEON 9500 pro managed to beat these solutions in performance in most cases. So here the attractiveness and popularity of the graphics cards will be mostly determined by their price compared with rivalry solutions on ATI chips. In fact, NVIDIA has a lot of room for price reduction, so I am pretty sure that NVIDIA GeForce FX 5600/5600 Ultra based graphics cards will cost little enough to be an attractive buy.

Well, and in conclusion a few words about the drivers: don’t you think they should finally bring the names of different “quality-performance” modes in accordance with common sense?