All you need to know about your products!
 |
 |
|
3DNews Vendor Reference English Resource - All you need to know about your products! |
|
||||
|
|
|
||||
|
|
|
||||
|
|
|
||||
|
||||||
Gigabyte GV-N595U-GT (NVIDIA FX 5950 Ultra)Date: 24/02/2004
Introduction
I think it won't be an overstatement to say that currently the graphic accelerator market has turned into a relatively quiet spot. ATI and NVIDIA have suddenly ceased their announcements of whatever solutions for all the market sectors. Vendors are gradually selling out their unsold produce and getting morally ready to a disastrous reduction of prices for high-end boards (which is inevitable after announcement of new solutions by the giants of the graphic chip production). Reviewers, for their turn, are also getting bored since they are already tired of examining boards of the same type differing in only the packaging, package bundle and stickers on the reference cooling systems; so every video cards that stands out among the mass in something is taken like sort of a bright event :-). The more interesting it is to review the produce by Gigabyte. Until some time in the past, the company did not produce video cards built on the chips of the two strongest players on the graphic chip market, but recently it again set about production of boards built on the base of NVIDIA graphic chips. The situation with various transitions and returns of old players of the video card market into their original places is very exciting enough. Currently, we can see such metamorphoses on the market of graphic accelerator production which were unthinkable, say, a year ago. And the example with Gigabyte is indicative of that. To date, the company is producing graphic chips based on both NVIDIA and ATI chips. By now we can quite safely claim with confidence that the time when you had to choose between manufacturers of graphic cards has gone, and currently we believe the manufacture of boards based just on the chips of both leading graphic chips manufacturers is the most advantageous option: ATI and NVIDIA. We don't take into account the manufacturers of other chips - their solutions are currently of absolutely no interest for the mainstream market of graphic accelerators. The inarticulate attempts of XGI (among the latest examples) to do something to catch up with today's flagship models of ATI and NVIDIA give us a clear idea - it's still a long way to go to the times when we might see something superior or at least equal at speed to ATI and NVIDIA chips :-). Coming back to the topic of transition of the leading graphic accelerator manufacturers to the production of boards built on chips of both ATI and NVIDIA, we can note that such transition is in fact the only sound decision. It's very good that the rules of the market were broken and manufacturers were able to acquire relative freedom in terms of forming their own product lines. Just think why a notable brand should lose its customers who want a solution, say, by NVIDIA, but the company can only provide solutions built on ATI chips? Now the situation has changed to the better.  Today, we'll be reviewing the fastest of all the Gigabyte product line, namely, the GV-N595U-GT. In the model line of GigaByte, there are two models built on the NVIDIA GeForce FX 5950 Ultra chip at a time - they are GV-N595U-GT and GV-N595U256V. So, what's the difference between the two models? The thing is, the "GT" prefix is used by Gigabyte to label some uniqueness or especially "advanced features" of its products. The same is in the case with GV-N595U-GT - it's not merely an ordinary board built on the base of NVIDIA GeForce FX 5950 Ultra. This modification is again the same NVIDIA chip but which offers essentially increased core clocking relative to the nominal. As is known, the frequencies for NVIDIA GeForce FX 5950 Ultra, as per NVIDIA's specifications, amount to 475 MHz/ 475 MHz (950 MHz). Due to the thorough selection of work chips, Gigabyte was able to produce the GV-N595U-GT model whose core is able running at guaranteed 520 MHz. The gain is substantial enough, and this fact gives us every reason to assert in advance that GV-N595U-GT is the fastest board built on NVIDIA GeForce FX 5950 Ultra, because all the other manufacturers left the frequencies in the products intact. It is understandable that NVIDIA GeForce FX 5950 Ultra is already extremely overclocked NVIDIA GeForce FX 5900 Ultra; and it's an unforgivably costly luxury to select chips that provide guaranteed operation at frequencies considerably exceeding the nominal. But not for Gigabyte.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ATI Radeon 9800 XT | NVIDIA FX 5950 Ultra | |
| Code name | R360 | NV38 |
| Chip technology | 256 bit | 256 bit |
| Process technology | 0.15 mk | 0.13 mk |
| Q-ty of transistors | ~107 mln (?) | 130 mln |
| Memory bus | 256-bit DDR/DDR2 | 256 bit DDR |
| Memory bandwidth | 23.4 GB/s | 30.4 GB/s |
| Pixel fillrate | 3.04 Gpixel/s | - Texelrate (Color+Z): 1900 texel/s - Z-rate (Z): 3800/s - Stencilrate: 3800/s - Texturerate: 3800/s |
| Maximum antialiasing mode | 6x | 8x |
| Triangle transformation speed | 412 Mtriangles/s | 356 Mtriangles/s |
| AGP bus | 2x/4x/8x | 2x/4x/8x |
| Memory | 256 MB | 128/256 MB |
| Core speed | 412 MHz | 475 MHz |
| Memory speed | 365 MHz (730 DDR) | 475 MHz (950 DDR) |
| Q-ty of vertex shader blocks | 4 | FP array |
| Pixel pipelines | 8 | 4 (8) |
| Texture blocks per pipeline | 1 | 2 (1) |
| Textures per texture unit | 8 | 16 |
| Vertex shader version | 2.0 | 2.0+ |
| Pixel shader version | 2.0 | 2.0+ |
| DirectX version | 9.0 | 9.0 |
| Antialiasing modes | Multisampling | Multisampling |
| Memory optimization | Hyper Z III+ | Optimized LMA II Color compression |
| Optimizations | SmartShader 2.1 SmoothVision 2.1 |
IntelliSample HCT |
| Q-ty of monitor outputs | 2 | 2 |
| Integrated RAMDAC | 2 x 400 MHz | 2 x 400 MHz |
| Bits per color channel | 10 | 10 |
| Special features | Integrated TV coder; FullStream Adaptive filtering F-Buffer |
Integrated TV coder; Extended Programmability Adaptive filtering UltraShadow |
Inside the nice-looking glossy Gigabyte-styled box made specially for NVIDIA produce, we found the following contents:
The package bundle is up to the mark, however, what's missing is some sort of a "spice" that every manufacturer tries to put inside not only Hi-end product boxes, but to the boxes of regular Middle-end products shipped to the retail. What's a bit disappointing is that Gigabyte offers quite a standard package bundle that doesn't stand out among the others. They might have added some more bonus items for a card of such price tag!
Design and layout
The first glance at Gigabyte GV-N595U-GT gives us a clear idea that we deal with a complete replica of NVIDIA's reference design. This is not only about the PCB of the video card, but the cooling system as well which was completely copied from NVIDIA. There aren't any differences in either the component layout nor even in the positioning of capacitors. The cooling system, as we already noted above, is also reference-based; only the engraved GT logo makes us aware that it's not just a reference board by NVIDIA.
The board itself proved to be quite massive. This is certainly because of the very heavy cooling system (for more details, read below). This makes you recall the NV30 with its cooling system, however, the anxiety is unfounded - NV38 uses a different operating principle for the reference cooling system.
The board takes up not only the AGP slot, but the adjoining PCI as well.
The video card is made in Gigabyte's characteristic blue PCB, offers 256 MB DDR onboard with a 256-bit data transmission bus (eight 32-bit chips positioned over the front and rear sides of the PCB). The video card offers the AGP 2x/4x/8x interface and a standard set of outputs: one DVI-I, one analogous, and one VIVO.
The signal for digital monitors is formed by the Sil164CT64 TMDS-transmitter made by Silicon Image.
The board also offers a SAA7108AE VIVO-chip made by Philips.
Besides, on the front side of the PCB you can see a connector for additional power necessary for video cards built on NVIDIA GeForce FX 5950. Additional power doesn't have to be applied, but in this case the card would run at reduced speeds, of which it will earnestly report upon the very first boot-up of Windows. The shortcoming of the additional power connector on Gigabyte GV-N595U-GT, like that on MSI GeForce FX 5950 Ultra, is its vertical positioning. First, it's quite difficult to apply power with an AGP video card already fitted in place. Secondly, the fastening of the connector leaves much to be desired.
There is an advanced BGA-packaging for memory chips made by Hynix (HY5DU283222). The access time of memory chips is 2 ns, which is equivalent to 500 MHz (1000 MHz), but the memory runs at the specified 475 MHz (950) which is prescribed by NVIDIA for GeForce FX 5950 Ultra. In other words, the card offers a small margin of memory overclocking. The GPU operating frequency has been overstated and amounts to 520(!) MHz.
If we once dismantled the whole cooling system of the board, it would be unfair not to look inside NV38 itself. The chip was manufactured on the 38th week of the year 2003 and has revision A1.
Of course we can't pass over the cooling system of the Gigabyte GV-N595U-GT in silence. At first glance, it may seem we have come back to the FlowFX again. In fact, it's not the way it seems. Yes, the system looks clumsy and awkward, but the operating principle is already different. The air for cooling the radiator that covers the NV38 graphic chip is sucked from outside through the grid in the fastening plane of the video card.
The air entrapped from outside is directed to the massive radiator that tightly covers the graphic chip. After the radiator is cooled by the air applied from outside falls inside the housing (already heated from the radiators). Of course this will not effect positively on the overall temperature inside the CPU unit, but the video card will always be blown about by the external cool air. The RPM rate is quite high, which makes much noise, but the noise level is much lower than that of FlowFX, so you can easily put up with it.
As the thermal interface, the regular so-called "chewing-gum" is used.
Software
Of all the bundled software, we'd like to note Gigabyte's proprietary utility for hardware monitoring and overclocking, the Vtuner.
Along with the overclocking, the utility also does hardware monitoring of the board's condition: the GPU temperature, the ambient temperature, and the fan's rotational speed.
After the installation, the program is hidden into the tray thus giving the user an opportunity to change a number of parameters which were possible to reach only through NVIDIA's control panel.
To compare the performance of Gigabyte GV-N595U-GT, we used the following video cards:
We have already tested the first board. It was chosen as an ordinary board built on the NVIDIA GeForce FX 5950 Ultra chip with default frequencies to estimate the performance of Gigabyte GV-N595U-GT relative to a regular board based on NVIDIA GeForce FX 5950 Ultra.
Sapphire Radeon 9800XT was selected as a direct competitor to NVIDIA GeForce FX 5950 Ultra. So, in today's review we'll be comparing ATI's and NVIDIA's flagship solutions, as well as, in view of the release of the new Catalyst 4.2, estimating its performance versus Catalyst 4.1. Among the essential in Catalyst 4.2, the following is worth mentioning:
Versions of the components:
You can view a complete list of improvements, amendments and limitations of the new driver available from here.
The increase of performance hasn't concerned DirectX 9.0 cards, so tests of the new driver version actually promises nothing interesting. But that the screen doesn't die away any more when changing the D3D settings - that's a really useful improvement, and it works excellent, I must admit.
We have finally upgraded our AMD-based test configuration. Now it is a quite powerful system based on AMD Athlon XP 3200+ Barton (200x11 = 2200 MHz).
| CPU | AMD Athlon XP 3200+ (Barton (200x11 = 2200 MHz)) |
| MoBo | MicroStar K7N2 Delta (nForce 2 Ultra 400) |
| Memory | 2x512MB DDR PC 3200 Kingston (Dual-channel mode, latency timings 5:3:3:2) |
| Video cards | • Gigabyte GV-N595U-GT (NVIDIA GeForce FX 5950 Ultra); • MSI FX5950 Ultra VTD256 (NVIDIA GeForce FX 5950 Ultra); • Sapphire Radeon 9800XT (ATI Radeon 9800XT) |
| OS | WinXP + SP1 + DirectX 9.0b |
| Drivers | Detonator 53.04 Catalyst 4.1, Catalyst 4.2 |
We remove all the decorative "niceties" out of the Windows GUI and set the operating system to maximum performance.
Disable the Vsync forcedly via the drivers both in OpenGL and in Direct3D applications. The S3TC texture compression was also disabled.
Test software:
As you can see from the list of our benchmarks used, we have changed their qualitative selection as well as upgraded some benchmarks to the most recent versions. Unreal II has also been upgraded to version 1403: The Awakening in which along with the purely gameplay-related corrections and bugfixes, a performance boost was attained due to solving artless issues of illumination and other fine revisions to the engine.
HALO: Combat Evolved was upgraded to version 1.3. As per the information in the readme-files to the patch, the game does not offer any amendments to the performance boost, nevertheless, the tests will be run with the latest version of the game.
We also recorded our own demos for Half-Life 2. Yes, that's an alpha version that leaked to the Internet, it's rather raw, and it's still premature to make conclusions on the base of this version of the gaming engine, but as the practice shows, such tests are of really keen interest.
Among the new benchmarks, there is Tomb Raider: Angel of Darkness, a shader game of the DirectX 9.0 generation that uses version 2.0 pixel shaders to the full. We'll tell more details during the tests of the boards.
Besides, we switched to testing in X2: The Threat from the demo to a full version of the game.
Also, in today's review that will be the first time we try a demo version of Unreal Tournament 2004. The demo appeared in the Internet not so long ago, but in the opinion of gamers is of exciting interest to the gaming community. That is, the transition from Unreal Tournament 2003 to Unreal Tournament 2004 is merely a matter of time, so it would be interesting for us to find out about the alignment of forces among the top-end ATI's and NVIDIA's accelerators in this game. We'll give more details about the game and the amendments to the graphics component during the test itself.
The video card was overclocked with the RivaTuner utility. The operational stability during overclocking was measured with Future Mark's 3DMark2003 v330 and MadOnion's 3DMark2001SE benchmarks. In the first of them, the third scene from Trolls' Lair in the "cycled" mode was played to reveal the overclocking potentials of the boards. On finding the critical frequencies we ran 3DMark2001SE benchmark three times. Once distortions appeared on the screen ("broken" pixels, stripes, lines etc.), then we pushed the memory frequency down by a few megahertz. If the computer hanged, we reduced the clock speed of the graphic chip. If the tests ran successfully, the overclocking was regarded as successful. We did not apply additional cooling.
Therefore, we were able to overclock the Gigabyte GV-N595U-GT video card to the following clock speeds:
| Video card | Core/memory Max |
| Gigabyte GV-N595U-GT | 560 MHz / 1035 MHz |
| MSI FX5950 Ultra VTD256 | 535 MHz / 1060 MHz |
To compare the overclocking potentials of boards on the base of NVIDIA GeForce FX 5950 Ultra, we're bringing in the overclocking results for the already tested MSI FX5950 Ultra VTD256. As we see, the overclocking potential of the chip in Gigabyte GV-N595U-GT is at a much higher level than in MSI FX5950 Ultra VTD256. This is explained easily enough: the chips that come to the manufacture of Gigabyte's GT-models undergo a thorough quality assurance when run at increased speeds, so it's no wonder that Gigabyte GV-N595U-GT offered much higher chip overclock than the MSI's board. But the memory overclock is not so high relative to the MSI's board.
Following the tradition, we start with our radically amended benchmarking package.
The handling of version 2.0 pixel shaders is just a bottleneck for NVIDIA FX video cards. As compared to Sapphire Radeon 9800XT, even the overclocked version of NVIDIA GeForce FX 5950 Ultra by Gigabyte - Gigabyte GV-N595U-GT being reviewed today cannot compete on par with Sapphire Radeon 9800XT and loses to the latter at handling some shaders in multiple times!
This test allows assessing the speed at which the geometry is processed by the accelerator. We used the most advanced mode with three diffuse-specular light sources in combination with three different operating modes: the traditional TCL (Fixed-Function Pipeline), version 1.1 vertex shaders and version 1.1 pixel shaders, version 2.0 vertex shaders and pixel shaders, as well as the simplest mode of ambient illumination is also in combination with the three operating modes: the traditional TCL (Fixed-Function Pipeline), vertex shaders 1.1 and pixel shaders 1.1, vertex shaders 2.0 and pixel shaders 2.0. In the case of Ambient illumination (the simplest constant illumination) and transformation we reach the practical limit for the card's bandwidth at processing the triangles.
Let's look at the test results. As we can see, the NVIDIA chip does a better job at the fixed T&L than ATI Radeon 9800XT due to known reasons. The latter lacks hardware support for the T&L emulation. And 148+ million triangles per second in Gigabyte GV-N595U-GT makes it stand out (the higher frequency of the graphic chip makes itself felt).
In ATI Radeon 9800XT, there is no dependency on the version of the shader used, the chip shows absolutely identical results of peak performance at triangles in all the modes. Albeit insignificantly, NVIDIA GeForce FX 5950 Ultra changes the values of its peak performance.
Let's see how things are with the most complex illumination model: in it there are three diffuse-specular light sources. In the case of the emulation of traditionally old T&L, NVIDIA chips lead as usual, but when migrating to DirectX 8.0 generation shaders the chip become more or less equal. Finally, in using version 2.0 pixel and vertex shaders ATI makes a sharp leap forward. Well, NVIDIA's bottleneck has been known for a long time, and we can only recall the results for NVIDIA boards in this test with old (nor ForceWare) drivers and think that things are not that bad =).
This test performs a number of various tasks, but we were mostly interested in the possibility of measuring the performance of frame buffer filling. We used two schemes in this test: both with a 256x256 texture, and without such.
The values of frame-buffer fill-rate produced in this test with the ATI chip do not coincide a bit with the theoretical maximum announced by ATI itself, but with NVIDIA everything is correct for the case of the Texelrate (Color+Z) mode.
However, the frame-buffer filling anyway occurs in ATI chip faster than in the NVIDIA chip.
In the case of simultaneous texturing, the alignment of forces doesn't change.
This test in the D3D RightMark benchmarking package allows to estimate the performance of executing various pixel shaders of the second version. In this test, the geometry has been substantially simplified to minimize the dependence of results of the test on the geometric performance of the chip and verify the operation of pixel pipelines only. We've brought in the operational modes for both 16-bit and 32-bit floating-point precision (switching between the precision modes is topical for NVIDIA chips only).
NVIDIA chips proved shattered at performing version 2.0 shaders under 32-bit floating-point precision. Yes, we can make reservation for that the ATI chip was running that time at its customary 24-bit precision, but let's look at the fps values produced with 16-bit floating-point precision. In this case, the NVIDIA chip is in a vantage point, since it uses less precision rate than the ATI chip (whose absolute values, as expected, do not change - it runs at its invariable 24-bit).
And again a fiasco. At the results of the test, NVIDIA chip executes pixel programs twice slower than ATI chips do.
This test is aimed at revealing the accelerator speed at displaying point sprites. In the test settings, we used 2 diffuse light sources.
In the most demanding mode with two diffuse-specular light sources, the ATI chip again beats NVIDIA. Note that the version of the executable shader does not play any part for ATI Radeon 9800XT, while NVIDIA GeForce 5950 Ultra starts lagging behind if version 2.0 vertex shader is used.
This test allows to estimate the efficiency of removal of hidden points and primitives by the accelerator.
At that, NVIDIA chip does much better.
Despite the poor results at handling pixel shaders in NVIDIA GeForce FX 5950 Ultra, we see a leadership of NVIDIA boards. What to say to this? NVIDIA optimizations are used to the full :-).
In the antialiasing mode, the overall picture does not change.
To date, at 3DMark 2003 among the top-end accelerators there is more or less parity, with the reservation that in 1024x768x32 NVIDIA wins due to using the Gigabyte GV-N595U-GT overclocked by the manufacturer. If impartially compared to the NVIDIA reference, then GeForce FX 5950 Ultra made by MSI loses.
There is no difference between Catalyst 4.1 and Catalyst 4.2.
From synthetic applications, we are now moving on to analyzing the performance of the graphic boards in real gaming applications.
In the most comfortable mode (no antialiasing and anisotropy enabled), we can see a customary leadership of NVIDIA boards at this test. However, like in 3DMark 2003 we can see some strange anomalies at 1024x768x32. And only Gigabyte GV-N595U-GT again brings NVIDIA to leadership thus justifying the name of the fastest board of the NVIDIA GeForce FX 5950 Ultra class.
The antialiasing modes do not bring us anything new in the alignment of forces except the more realistic situation at 1024x768x32.
In the AF mode, NVIDIA again loses. The analysis of AF quality is beyond the scopes of this article, so we have to be content with the figures produced with these accelerators.
Finally, the most demanding mode (AF+AA) shows a leadership of the board based on ATI Radeon 9800XT. The faster anisotropy in ATI does add much to the results.
ATI boards demonstrate a traditional leadership at this benchmark. NVIDIA's boards on GeForce FX 5950 Ultra lag behind quite substantially.
In this latest OpenGL game we can see a sure leadership of ATI boards. It's quite strange considering the traditional commitment of NVIDIA boards to OpenGL products.
As we expected, the modes of combined AF+AA do not bring anything new.
The "Ultra Shadow" technology implemented in NVIDIA chips allow them easily bypassing ATI boards in this test. Due to this technology, stencil shades whose rendering technology is widely used in the game are built faster. Similar effects will be used in future DooM II, so some assumptions of the alignment of forces among the graphic accelerators in Carmack's creature can already be made right now.
Unfortunately, we haven't got any information on the engine of this benchmark, so we can only assert the leadership ATI boards at it.
At that, we see a slight superiority of Sapphire Radeon 9800XT board relative to NVIDIA GeForce FX 5950 Ultra. And that is under unforgivably low operation of version 2.0 pixel shaders in synthetic tests for NVIDIA boards...
At that, NVIDIA boards feel comfortably enough, despite the unforgivably low operation with DirectX 9.0 shaders which play not the last part here albeit not used to the full.
In a real-world gaming application that uses version 2.0 shaders (we invoked their forced use via the game's command line), NVIDIA boards lose. But as the resolutions go up, the difference is graded.
The game appeared in sales quite a long time ago, but are able to test it as part of our testing toolkit only now.
For tests, we used 3 recorded demo benchmarks:
• Paris5_4;
• Paris1c;
• Paris2g.
In our tests, we used the defaults settings for boards built on both ATI and NVIDIA chips.
Tomb Raider engine: Angel of Darkness is a technically advanced DirectX 9.0 solution. The game uses version 2.0 pixel and vertex shaders, which, considering the results of synthetic tests, is not going to bring anything good for NVIDIA boards. Nevertheless, let's look at the figures.
Yes, it seems like NVIDIA boards do quite a poor job handling vertex 2.0 shaders in real-world applications, but not in overly optimized benchmarks. This test is remarkable in that the benchmark is saturated with effects based on version 2.0 pixel programs. This is the reason why NVIDIA boards show low frame-rate. Also note that the results for NVIDIA boards can't be regarded valid in this test because of the problems with the rendering. In particular, there are problems with shadows display at the start of the test scene. This bug was fixed in version 52 patch, but the benchmark functionality was excluded from there, so we have to take readings using version 49.
This test is abundant with shaders in the less extent than in the previous, nevertheless this does not allow NVIDIA boards to take a lead.
The release of games like Unreal Tournament 2004 is always anticipated with impatiently. Of course, there is not so much hype like it is about Half-Life 2 or DooM III, but the key points of this game are highlighted in a somehow different way. Originally, Unreal Tournament 2004 (like Unreal Tournament 2003) was targeted at wide gaming community with the emphasis made on that the game be played for many years. The notion "cyber-sport" (or "e-sport") was tightly associated with Unreal Tournament (in the year 2003 edition, and now in year 2004).
It's just this fact that attracted undivided attention from graphic chips manufacturers (and, above all, their driver-writers :), because hardly ever one wants to lose potential customers who find it important that the operation of the board in this very game be fast enough, and the gaming community is more than honorable.
The graphic engine of Unreal Tournament 2004 does not bring anything new. The Unreal Engine does not use great number of pixel and vertex shaders in the game, so the board's performance at Pixel Shaders 2.0 is the cornerstone in reviewing the performance of boards in this application.
Originally, the demo version of Unreal Tournament 2004 did not provide benchmarking functionality. Although the "Benchmark" folder existed, it was empty. That is, it seemed like one had to forget about the benchmark (especially that the game's option of creating your own demos through the "/record" command was also disabled), but the solution to the problem turned out to be simple enough. As it turned out, there are demos for the benchmark - but their use was simply not enabled, and to make them run you would have to change the paths and map names in the *.bat files as per the resources provided in the demo version of Unreal Tournament 2004. In the branch of the same forum you can find a link to ready *.bat files for the benchmark. However, note that the thus enabled benchmark works incorrectly with flyby demos. On starting the above, simple the level with the only fixed item is shown, but no fly-around is performed. Therefore, we had to use only the "botmatch-tests" recorded in all possible modes of the future game of Epic's. Also, before we proceed directly to tests in the game, note that the quality of textures in the demo version matches that for the "Normal" level. That is, tests in the demo version of Unreal Tournament 2004 will not give us a vivid idea of the absolute fps values, nevertheless it places all the tested boards in similar conditions.
For tests, we traditionally selected the maximum possible quality settings from the menu. The antialiasing and anisotropic filtering were forced from the video card drivers.
In most of the demos presented in the demo version, we can't establish a one-one leader. Gigabyte GV-N595U-GT, due to its increased core clocking takes a small lead at all of the tests. As regards the fair comparison of ATI Radeon 9800XT vs NVIDIA GeForce FX 5950 Ultra, we can only establish a complete parity except in the Botmatch Colossus test at which the board based on ATI chip leaps forward leaving its rivals well behind.
And again we get an approximate parity of all the cards.
In the 1600x1200x32 resolution, when the fps in the game depends more on the operation speed of the graphic accelerator and not on the CPU speed, we can get a better idea of the alignment of forces on the graphics front. The leadership of ATI board in the "Botmatch Colossus" test enhanced with the increase of resolution, but in the "Botmatch Bridgeoffate" test, on the contrary, the board based on NVIDIA GeForce FX 5950 Ultra proved to be a sure leader.
Such vague and queer, at first glance, ratios of performances for the top-end accelerators of ATI and NVIDIA in different demos can be explained by the difference in the scene geometry, its size (e.g. narrow corridors of DeatchMatch levels versus vast spaces in Assault), which in the end gives a leadership for ATI boards in one test as opposed to NVIDIA'a leadership in other tests. In the total, we get an approximate parity. However, let's see how the alignment of forces will be affected if we enable the image improvement techniques
Once the FSAA is enabled, we see quite uniform results for all the cards without exception. Only Gigabyte is traditionally a bit faster because of the 520 MHz graphics core.
With the forced AF, the situation dramatically changes to the better in favor of NVIDIA. All this happens despite that NVIDIA's AF results are usually worse because of the specific image processing algorithms... The AF modes grab the leadership crown away from NVIDIA's products in most applications. This fact will be studied in more detail in further materials.
And finally, the most demanding modes combining the AF+AA bring victory to NVIDIA - the faster anisotropy in this benchmark makes itself felt.
As we have already mentioned in the explanatory part of the material, we recorded our own demos to this alpha version of Half-life 2 that leaked to the Internet.
The first demo recording (3DNews001) was made on the same level which was demonstrated at the past E3. The scene is abundant with pixel shaders (in particular, the water is made in a most effective way), so it puts quite a lot of load on the accelerators.
The second demo recording (3Dnews005) is a typical gaming situation at one of the game levels. It is less abundant with shaders, but due to the fast dynamic action the frame-rate in the demo can't be regarded as high.
This demo is closest to the real gaming situation, so it should be of most interest to potential gamers. But, as we have repeatedly mentioned in our materials, it is premature to make final conclusions on the performance of an accelerator basing on a raw and leaked version, since in the final release many things might change (we are more than confident that it will be the way).
In one of the most eye-catching demos that makes active use of shaders we see an evident failure for the NVIDIA cards. The pure HLSV that the Half-Life 2 engine is currently cannot be handled properly on NVIDIA family video cards due to the architectural specifics of NVIDIA cards. The increase of core clocking in Gigabyte GV-N595U-GT gives its advantages, but they don't solve the global architectural problem on the whole.
The gap between Sapphire Radeon 9800XT and video cards on the base of NVIDIA GeForce FX 5950 Ultra chip in this demo is no longer a few percent, it differs now in multiple times. Note that the gap of ATI chip from NVIDIA boards increases with the rise in resolution, that is as processor dependence of the test decreases.
In our previous material(" ASUS Radeon 9600XT Video Card Review ") we dwelled enough on this benchmark. As we remember, Middle-End chips of NVIDIA were not superior in that test despite the flexible system of scene processing by the game engine owing to the scene rendering technique most optimum for the accelerator. Let's look at the Hi-end.
As we see, in the least demanding mode without AA and AF the NVIDIA GeForce FX 5950 Ultra lags well behind ATI Radeon 9800XT.
The picture is the same. As the resolutions go up, the gap gets wider.
The most demanding mode with the AF+AA combined gives the same alignment of forces. NVIDIA GeForce FX 5950 Ultra can't gain revenge in this mode.
Running tests with new games appearing on the market is always exciting. First, it's always interesting (and necessary) to estimate the speed of graphic boards against the latest developments of the industry; secondly, it's interesting to see why you should fork out your 400 "evergreen" bucks for a latest Hi-end board if nothing else can be run with it except 3DMark 2003 to feast your eyes upon shader technologies :).
The game FarCry is a bright example of what we can expect from Hi-End boards with support for DirectX 9.0. The beauty demonstrated in this albeit raw demo version (even though released officially) gives cause for thinking about the necessity of buying a hi-end graphic accelerator, because you can see the real prospects of using such products in practice already now.
We ran tests with one of our own demos (3Dnews004 Demo) which is a fly-about above a part of the island where all the action in the FarCry demo takes place. We had to give up testing with the demo recorded by us which imitated the gaming process in FarCry Demo because such demos are incorrectly played. They suffer from numerous bugs like disappearing models of the enemies, no shooting and other annoying things which prevent from taking 100% identical readings from all the boards. However, you can also get a good idea of the performance level of the game with the demo 3Dnews004, as well as the inconsistency of performance results among the top-end boards by ATI and NVIDIA.
In this demo, we used a special patch dubbed "WaterFix" to solve the problem of reflection of various objects in the water, because this option remains disabled in the demo version of FarCry. On the screenshots above you can clearly see the difference between the "unpatched" version of FarCry Demo (left) and the "patched" one (right). Of course, this innovation will affect the performance, however, we have a right to expect the best image quality from Hi-end boards with their fabulous price.
Traditionally, all the settings are set to the maximum possible level ("Very High") except the texture quality settings - the demo version has a restriction on their quality to the "Medium" level, and it's impossible to enable the higher detail level. Most likely, it is caused by the developers' intention to reduce the size of the distribution package, so the maximum-quality textures were not included in the demo version but for the medium-quality textures.
We were unable to run tests in the antialiasing and anisotropic filtering modes, since the game does not respond to forcing any image quality improvement techniques manually out of the driver, but the use of special console commands ("d3d9_Texture_Filter_Anisotropic 1" and "r_Texture_Anisotropic_Level X", where X - the number of anisotropy samples, and "r_FSAA 1" + " r_FSAA_samples X", where X - the number of antialiasing samples ) to enable, say, the antialiasing, produces a great number of artifacts, which makes it impossible not only to run the tests, but the gaming process as well. Let's wait for the official version of the game.
Anyway, let's look at the test results. So, what can we see now? Frankly, we expected a more pessimistic scenario of behavior for NVIDIA boards in this test. Yes, - ATI Radeon 9800XT is an undisputable leader of this test. Tracing the dynamics of resolution rise, we see that as the resolution goes up, the gap also widens (we were able to see that many times in all the tests most demanding to the graphic subsystem of the computer).
On the other hand,in the most playable resolution 1024x768x32 NVIDIA and ATI boards go abreast. And Gigabyte GV-N595U-GT even leaps forward thus justifying the name of the fastest board built on the NVIDIA GeForce FX 5950 Ultra chip.
As we see, shaders are again still a hard nut to crack to NVIDIA chips, but anyway the progress is there. What is it about? Of course, the architectural problems of the GeForce FX haven't gone anywhere. Most likely, the game developers did quite a global work on optimizing the game engine for the architectural specifics of NVIDIA boards, since FarCry is a participant of NVIDIA's "The way it`s meant to be played" initiative.
Well, it's high time we summed it all up. Today, we've had an unusual testing session in many ways. A number of new tests, the change of the test configuration, as well as reorganization of many tests... By the totals of comparison of the latest solutions by NVIDIA and ATI, we still can't name a one-one leader in the Hi-End niche. We can safely assert an approximate parity (by the total number of tests passed) in old applications which do not use shaders of the DirectX 9.0 generation. At that, the boards show approximate parity in the total. In shader games of the latest generation, all is not so straightforward. If the game developers worked quite closely on the adaptation of the game for the architectural specifics of NVIDIA boards, then things with speed for NVIDIA are more or less good. If such approach was not practiced, then purely shader games will run faster with boards built on ATI Radeon chips. Proceeding from this reasoning, as well as considering the prices for the products, the buyer would have to choose between Hi-end boards.
The current alignment of forces between ATI Radeon 9800XT and NVIDIA GeForce FX 5950 Ultra presented in this material should preserve until the release of new solutions by ATI and NVIDIA. Sure, there are persistent rumors about ForceWare2, but these are still rumors =).
Regarding the Gigabyte GV-N595U-GT Video Card. Leaving out the quite costly solution of Gainward's offering water cooling (i.e. rather nonstandard and unusual for the mainstream market), in today's review we have tested the currently fastest board built on NVIDIA GeForce FX 5950 Ultra whose graphic chips is able running safely at 520 MHz versus standard 475 MHz in all the GeForce FX 5950 Ultra boards. Basing on the test results, this of course brings advantages, but they will make sense provided the prices for boards are reasonable and sane. In all the other respects, the board demonstrated the best of its showing. The package bundle was also up to the mark. So, if you are looking for a board based on NVIDIA GeForce FX 5950 Ultra that is able running safely at higher clock speeds of the core, and if you are not confused by the higher price, then Gigabyte GV-N595U-GT is the right board for you.
The Gigabyte GV-N595U-GT video card for tests was granted by Gigabyte
Sapphire Radeon 9800XT was presented by Sapphire Technology
NVIDIA VGA Roundup `2003ATI VGA Roundup `2003GeXcube Radeon 9600XT ExtremeASUS Radeon 9600XTNVIDIA FX5700 UltraMSI GeForce FX 5950 Ultra3DMark2003 build 320 vs 340: A rare moment of truth?Tests of ForceWare 52.16: FX 5900 versus Radeon 9800ProASUS RADEON 9800 XT: a turning pointTests of ATI Radeon 9800 PROFX 5900 versus Radeon 9800ProFX 5600Ultra versus Radeon 9600ProFX 5200 versus Radeon 9200
|
|
| |||||
By: