by Alexey Stepin
11/03/2003 | 10:33 PM
Until quite recently, ATI’s RV350 and NVIDIA’s NV31 were the mainstream graphics processors. The first of them was and is used in the RADEON 9600 (PRO) series. The GeForce FX 5600 (Ultra) series was based on the second one. The RV350 architecture resembles very much the R300 one, but has only four pixel pipelines and uses the 0.13micron manufacturing process (which allows for higher operational frequencies). NVIDIA’s NV31 traces its origin to the notorious NV30. This GPU features four pipelines and is made with 0.13micron technology, too. Following the example of the NV30, the NV31 can configure the pipeline formula to be either 4x1 (4 pipelines with one texture-mapping unit in each) or 2x2.
<%BANNER[article]%>A little while ago, the monsters of the graphics market, ATI Technologies and NVIDIA, both added new mainstream GPUs to their product lines: RV360 (RADEON 9600 XT) and NV36 (GeForce FX 5700 and 5700 Ultra). Both products are heartily welcome for two major reasons:
So, what can we expect from upcoming graphics cards featuring the RADEON 9600 XT or GeForce FX 5700/5700 Ultra? Read our review to learn what the graphics companies want to seduce us with this time.
The specification of the RV360 is looks like “the second edition” of the RV350 specs: their characteristics are really close to being identical. Well, there are differences, of course, but these are slight ones. The new chip works at a higher frequency and supports the dynamic overclocking technology called Overdrive. Even the outside of the chip resembles the looks of its predecessor:
The new manufacturing process uses the improved technology for component insulation aka low-k dielectric. I won’t delve deep into physics, but rather put it in a slightly different way: the new production technology allowed ATI Technologies to boost the GPU frequency by 25%! So, the graphics core of the RADEON 9600 XT (that’s the name of the new mainstream VPUs from ATI) works at 500MHz clock frequency, compared with the 400MHz of the RADEON 9600 PRO. You will see whether this frequency gain leads to a corresponding performance growth a bit later. Now, let’s take a look at one of our today’s testing participants: ATI RADEON 9600 XT.
The déjà vu feeling is hard to shake off: this graphics card resembles both RADEON 9600 PRO and RADEON 9700 PRO at the same time! Just take a look here:
As you see, it is the big square-shaped heatsink that makes the new graphics card look like the ex-flagship of the high-end series from ATI. Here, though, the heatsink is white and is equipped with a different fan. It is natural that we see another, more efficient cooling solution compared to the RADEON 9600 PRO, since the 500MHz frequency requires it. The heatsink is fastened by means of two stiff springy clamps. The ordinary yellow “gum” serves as a thermal interface between the cooler’s foot and the die surface. This might not be the best heat-conducting material in this world, but it does its job quite well and protects the die edges against clipping. The fan is quite noiseless, by the way.
Otherwise, the PCB is nearly identical to the one of the RADEON 9600 PRO, although some components are missing. To my regret, the engineers from ATI Technology once again forgot to mount heatsinks onto the memory chips. On the other hand, the memory frequency remained the same, 600MHz (300MHz DDR), and the memory chips don’t heat up too much. The graphics card uses memory chips from Samsung with an access time of 3.3ns. The access time is rather high; we cannot hope for any rewarding overclocking. There are 128MB of memory with the 128bit memory bus. The card we got for our tests is just a sample, so it doesn’t make much sense to talk about its accessories and package. I would like to say one thing concerning the 2D quality, though: our display had crispy graphics and text in all resolutions up to 1600x1200@85Hz. The picture became a bit fuzzy in 1800x1440, but this was not a standard resolution for our display.
The main idea behind the dynamic overclocking technology from ATI, called Overdrive, is quite simple: R360 and RV360 graphics processors are equipped with an built-in thermal diode to track the core temperature. If the core is cool, the driver can increase its frequency to a definite value. In case of overheating, the frequency is dropped down. As for the RADEON 9600 XT, the core may function at 500, 513 or 527MHz, but the current version of the Catalyst driver supports Overdrive for RADEON 9800 XT cards only. Overdrive for RADEON 9600 XT cards is promised to appear in the next version of the Catalyst.
The NVIDIA NV36 graphics processor is interesting for its ancestry. Unlike the NV31, it is the product of evolution of the NV35 architecture rather than the NV31, as might have been supposed. Thus, the new chip inherited technologies like a full CineFX 2.0 technology support from its elder brother.
NVIDIA claims the GeForce FX 5700 Ultra to process 32-bit pixel shaders at twice the speed of the NV31.
That’s the die of the new graphics processor:
In fact, the NV36 is an NV35, which has somehow lost half of its pipelines and the 256-bit memory bus. In our today’s tests this GPU is represented by the eVGA e-GeForce FX 5700 Ultra graphics card:
The graphics card looks most imposingly due to its huge PCB and massive cooling system. The total weight of this card is about 360g against the RADEON 9600 XT’s 200g. The cooling system accounts for the biggest share in this weight.
The big heatsink is equipped with a quiet fan. It cools both the 475MHz-clocked GPU and the GDDR2 memory chips from Samsung with 2.2ns access time. The memory works at 900MHz (450MHz DDR).
Yes, there is no misprint. NVIDIA returns to the expensive and hot GDDR2 memory as if they have already forgotten about the mishap with the GeForce FX 5800 Ultra. This is quite a surprise. On the other hand, the relatively simple PCB design may make up for the memory price. Anyway, NVIDIA had to roll out the products using GDDR2 sooner or later, even for the sake of getting experienced with production of such devices.
The total memory amount the GeForce FX 5700 Ultra carries onboard is 128MB.
The memory chips on the back side of the PCB are covered with a passive heatsink. There is also an iron plate labeled “NVIDIA” there. It adds robustness to the card and serves as an additional fastening for the heatsink. The heatsink itself is attached by means of spring nuts that don’t allow it to crash the GPU die. Here’s the same yellow “gum” thermal interface between the chip and the heatsink, while the memory chips make contact with the heatsink through a layer of thermal paste deposited on some fiber base. The back part of the PCB with its power connector and voltage regulators resembles a lot the PCB of the GeForce FX 5800 Ultra. Otherwise, the PCB is original. The power connector is robust enough to withstand frequent plugging-in/out of the cable.
The front part of the PCB is more exciting. We can see a landing place for a VIVO chip as well as for an additional TMDS transmitter. Why is it here if the NV36 has an integrated transmitter already? Beside the VGA connector you can see contacts compatible with a DVI-I connector. Thus, installing a second extra transmitter, the manufacturer can produce a graphics card with two digital outputs for work with two LCD panels at a time.
This graphics card also provided an excellent quality of the 2D picture up to 1600x1200@85Hz resolution.
The e-GeForce FX 5700 Ultra graphics card arrived to our test lab in its sparkling brand-new package, painted up in NVIDIA’s traditional black and green.
The competitive edge of the product is emphasized by the pictures and captions scattered all around the box. A distant relative of Balrog from “The Lord of the Rings” postures in the upper left corner, and the proud words “Dominate Your Games” stretch out right below the name of the product. The lower right corner is dominated by the winged elf-lass Dusk, dressed in leather. She somehow reminded me more of the Nazi girls from Wolfenstein rather than of the innocent sylvan creature called Dawn. The traditional slogan, “The way it’s meant to be played”, goes above, next to the assurance that eVGA is an official distributor of solutions based on NVIDIA’s chips.
The backside of the box has a small transparent window and a brief summary of what the eVGA e-GeForce FX 5700 Ultra has and can do. By the way, the size of the package is an exact match of the one the original ATI RADEON 9800 PRO comes in.
Let us open up the package, take out the protective cover and see the following items:
One of the package’s flaps has a curious message suggesting you to contact the tech department of eVGA rather than return the card into the shop, if the card went down.
So, what new things are now available to us with the arrival of the new mainstream graphics processors? You can easily find the answer in the following table:
VGA Adapter | ATI Radeon 9600 PRO | ATI Radeon 9600 XT | NVIDIA GeForce FX 5600 Ultra | NVIDIA GeForce FX 5700 Ultra |
Graphics chip | RV350 | RV360 | NV31 | NV36 |
Production technology | 0.13micron | 0.13micron | 0.13micron | 0.13micron |
Transistors | ~75 mln | ~75 mln | ~80 mln | ~82 mln |
Memory bus | 128bit DDR | 128bit DDR | 128bit DDR | 128bit DDR-II |
Fillrate, GPixel/s | 1.6 | 2 | 1.6 | 1.9 |
Vertex processing speed, mln vertexes/s. | 200 | 250 | 100 | 356 |
AGP modes | 4x/8x | 4x/8x | 4x/8x | 4x/8x |
Graphics chip frequency | 400MHz | 500MHz | 400MHz | 475MHz |
Memory frequency | 600MHz DDR | 600MHz DDR | 800MHz DDR | 900MHz DDR |
Pipelines/TMU | 4x1 | 4x1 | 4x1/2x2 | 4x1/2x2 |
Pixel shaders version / vertex shaders version | 2.0/2.0 | 2.0/2.0 | 2.0+/2.0+ | 2.0+/2.0+ |
Max. FSAA mode Multisampled/Mixed | 6x | 6x | 4x/8x | 4x/8x |
Max. Anisotropic filtering mode | 16x | 16x | 8x | 8x |
Supported technologies | Hyper Z III+, SmartShader 2.0, SmoothVision 2.1 | Hyper Z III+, SmartShader 2.0, SmoothVision 2.1 | IntelliSample, CineFX | IntelliSample, CineFX 2.0 |
Number of RAMDACs/ RAMDAC frequency | 2x400MHz | 2x400MHz | 2x400MHz | 2x400MHz |
TMDS | yes | yes | yes | yes |
TV-Out support | yes | yes | yes | yes |
The NV36 evidently makes a farther step away from its predecessor with its three times higher vertex processing speed and the boosted-up GPU and memory frequencies. Will it be enough to win the race?
As usual, we benchmarked the new GPUs and the respective cards based on them in a variety of synthetic and real tests. The testbed was configured as follows:
The new driver from NVIDIA, ForceWare 52.16, needs a bit of description. This is the first version of the driver in which “optimization” of the shader code is officially implemented. That is, NVIDIA added a translator into the ForceWare, which transforms a portion of DirectX 9 code into a form the GeForce FX processor can easier digest. This is done “on the fly”, of course.
The reason is obvious. NVIDIA had to boost the performance somehow, seeing that many PC users are not satisfied with shader-processing speed of its GPUs. It seems like the GeForce FX architecture demands a thorough re-design to reach the same level of performance the GPUs from ATI Technologies now offer. So far, NVIDIA can only rely on software optimizations. Before proceeding to the benchmarks, I would also like to expose a few ideas concerning benchmarking in general.
So, let me tell you about the benchmarks used in the industry before we pass to some real tests.
Earlier this year both NVIDIA and ATI Technologies were caught on cheating in the 3DMark03 benchmark. The latter explained its position and eventually removed the cheats. The former went on saying that this industry benchmark had been developed with an evil intent – to show NVIDIA GPUs in bad light. This was a bit ridiculous and no one really believed that since 3DMark is co-developed by Microsoft, Dell Computer and a lot of other industry players. Unfortunately for NVIDIA, other cheats in its drivers were later disclosed by the community. Today no one believes the company is fully capable of developing a powerful graphics processor to run games fast enough without software “support”.
It is now perfectly clear that NVIDIA’s latest architecture is not as powerful as the company wants and claims it to be, but is still fast enough to play the vast majority of games available in the market, especially those that make little use of pixel and vertex shaders. The problem lies deeper, though. NVIDIA used to be an extremely successful company in the past and its solutions were tangibly faster and more feature-rich than products from 3dfx or ATI Technologies. Furthermore, this company had been adored by the industry and PC enthusiasts for so long that it simply couldn’t and wouldn’t be without being the number one. So, NVIDIA did not want to lose its King-of-the-Hill position and resorted to cheating. This dethroned the company faster than anything else possibly could, since a lot of enthusiasts feel now pretty suspicious about NVIDIA’s products.
NVIDIA got trapped just like 3dfx did back in 2000. The community of PC enthusiasts relies on benchmarks. When users saw that NVIDIA’s GPUs were slow in a majority of benchmarks used by the press, they started to bash NVIDIA heavily, just like they had criticized 3dfx.
It was largely the tech community and the press who first saluted 3dfx, then pulled NVIDIA on the very top of the 3D Everest, and finally helped ATI Technologies to rise and shine. This same community of PC enthusiasts and this same press dragged 3dfx down and took the crown away from NVIDIA. No doubt, ATI Technologies will be done with the same way, if the company makes a mistake.
All problems obviously come from the fact that advanced end-users sometimes too blindly believe in the benchmarks and believe what the benchmarks show them. Less advanced end-users believe enthusiasts, who believe in benchmarks, and that’s how it is – even if a product is not a competitive one, but scores high in popular benchmarks, the industry eats it. If it is not, it will never be accepted by the industry. If the crowd feels deceived, it totally disregards the product and forgets about all its real advantages. However, there is no use at all trying to convince the public that benchmarks are not to be worshipped, but approached carefully. The benchmark is God!
NVIDIA seems to have realized clearly all the pros and cons of cheating now. We do not expect the company to perform any more of its hardcore optimizations, such as those for 3DMark03 found by FutureMark earlier this year. Instead, NVIDIA now claims that its excellent software optimizes virtually every single shader program to run better on GeForce FX-series GPUs. Since no one can deny NVIDIA’s right to optimize the code run by its hardware, the company will continue the development of its software optimization layer called “Instruction Optimizer”. In case it brings some tangible speed boost, other leading graphics companies, particularly ATI Technologies may also follow the suit. There are a number of implications, however:
The biggest issue with benchmarks for graphics cards testing is the fact that they measure performance in one and only environment. Sometimes, even different demos can reveal totally different and contradicting results in the same game. Moreover, opposite results may emerge even if we test one particular demo in different versions of the game. Needless to say, even gaming benchmarks are not precise performance indicators for games. So, it is rather strange that we expect synthetic benchmarks to show us anything clearly.
Well, let us sum everything up now.
Following the well-trodden path, we start out with fillrate tests using different FSAA modes. Thus, we will get a notion of how the GPU handles the ever-increasing workload.
Well, the standard work mode brings quite standard results. The GeForce FX 5700 Ultra is much faster than its predecessor, GeForce FX 5600. The difference between the new and old RADEON based cards is negligible.
Multi-texturing adds something new to the picture: the high frequency of the graphics core allows the RADEON 9600 XT to show the best performance. The GeForce FX 5700 Ultra starts bravely enough, but soon gives up as the FSAA level grows. The GeForce FX 5600 Ultra behaves much alike, although takes the FSAA easier – its graph is flatter.
As you know, besides 3DMark 2001SE, we have a special test program, which draws a polygon and renders 0 to 4 textures (512x512 pixels resolution) onto it. That is, we see no texture at all in the first case and the pixel’s color is calculated by interpolation of the colors of the polygon’s vertices. The program serves for testing the graphics card in three modes. The first mode: color and Z writes are enabled:
The cards on the NVIDIA GPUs are doing quite fine, but too easily give in in the end: when the number of the textures is high, the GPU switches to the inconvenient 2x2 formula (2 pipelines with 2 TMUs per each). Interestingly, the cards on ATI chips show practically the same performance with no and with one texture.
Now, we enable Z writes:
And see that nothing has changed, really.
So, we turn off color writes, but enable Z writes:
This mode puts no limits onto the GPUs and they are free to show us a fill-rate close to their theoretical maximum. It’s no wonder then that the RADEON 9600 XT wins the test as its core is working at 500MHz frequency. The GeForce FX 5700 Ultra, clocked at 475MHz, grabs at the second place.
This test program supports shaders, too, in all the three modes:
The Megahertz matter! The new GPUs are obviously faster than their predecessors.
No surprises here. The RADEON 9600 XT once again confirms its superiority.
When color writes are disabled, the situation changes dramatically. The NVIDIA cards don’t lose in performance anymore when executing a simple version 2.0 shader. The two new chips are again faster than their predecessors – high frequency is a thing to count for! Note the curious fact that the software optimization helped the GeForce FX 5600 Ultra to catch up with and even outperform the ATI RADEON 9600 PRO somewhere.
3DMark 2001SE and 3DMark03 suites come next:
In the purely geometrical tasks, the processors from NVIDIA have always been strong. They are strong now, too. The high frequency and a much improved vertex processing speed make the GeForce FX 5700 Ultra the winner. Even the abominable RADEON 9800 PRO had to give up in this test in two resolutions out of three.
The GeForce FX 5700 got an even bigger advantage over the competitors when the number of lights is higher.
When rendering relief surfaces with the EMBM method, the NVIDIA GPUs are not at all confident. The RADEON 9600 XT is a definite leader, closely followed by the lower-frequency RADEON 9600 PRO.
Thanks to its fast memory subsystem, the GeForce FX 5700 Ultra wins in two resolutions out of three. Well, the Dot3 bump mapping method has always worked fast on NVIDIA’s GPUs.
The RADEON 9600 XT is best of all at processing vertex shaders, while the GeForce FX 5700 Ultra can only compete with the RADEON 9600 PRO.
Well, pixel shaders are the strongest point of any modern VPU from ATI Technologies. This test confirms the point for a millionth time already. However, the GeForce FX 5700 Ultra shows noticeably higher performance over the GeForce FX 5600 Ultra.
When version 1.4 shaders are in use, the performance gap between the RADEON and GeForce FX is only widening.
The NVIDIA processors make up for their shader failure by handling sprites faster. Note also that the performance here doesn’t practically depend on the GPU model: the pairs of graphics cards are running at the same speed.
The next benchmark, 3DMark03, let’s also pay due attention to the results obtained in it:
NVIDIA did have the reason to claim a three times higher vertex processing speed for its new NV36. The graph above proves the point. The GeForce FX 5700 Ultra easily leaves all its competitors behind to nearly hit the RADEON 9800 PRO in the back.
Version 2.0 pixel shaders. The picture is as gloomy for NVIDIA as it has always been. The “crutches” namely the shader code translator, could hardly hide the impotence of NVIDIA’s GPU in this test. Notwithstanding its higher GPU and memory frequencies, the GeForce FX 5700 Ultra is only once ahead of the RADEON 9600 PRO.
We already said in our reviews that the Ragtroll test is a combo-test, which puts both the GPU’s vertex shader unit and the CPU under a high pressure by rendering the trolls falling from a height. The RADEON 9600 XT wins lower resolutions, but the GeForce FX 5700 Ultra gains the upper hand in higher ones. Its fast memory contributes much to the result.
We cannot pass by the topic of pixel shaders since there appear ever more games using them. To find out which GPU is better at executing shader programs, we used a small but very easy-to-use benchmarking suite, ShaderMark 2.0. It is drawing pixel shaders, differing in beauty and complexity.
Unfortunately, this test doesn’t agree that NVIDIA’s GPUs are fully compatible with the DirectX 9 specification, giving out a message that some shaders could not be executed because of the GPU limitations. We initially thought the error to occur because of some issues with the Unified Compiler implemented into the latest ForceWare 52.16, but the reason appeared to be totally different.
According to Thomas Bruckschlegel from Tommti-Systems – the developers of ShaderMark, NVIDIA’s GeForce FX-series does not support floating point textures and render targets under DirectX 9. Even though one of the company’s Developer Relations officers said that NVIDIA would add support for these capabilities into “future drivers”, floating point textures and render targets still do not function under DirectX 9. Nevertheless, we offer you the full list of results in ShaderMark 2.0:”
You see the winner – ATI Technologies with its RADEON 9600 PRO and XT. There are two things to note about the results. First, the RADEON 9600 XT clocked at 500MHz shows the advantage of its extra frequency in the way of a neat performance gain over the 400MHz RADEON 9600 PRO. Second, the GeForce FX 5700 Ultra is about two times faster in this test than the GeForce FX 5600 Ultra. I guess we should acknowledge the work done by the NVIDIA engineers. However, the inherent drawbacks of the GeForce FX architecture cannot be easily fixed. The new GeForce is still slower than the new RADEON as far as shaders are concerned.
Winding up the theoretical part of this review, with its synthetic benchmarks, I should say that ATI’s RV360 is definitely better so far. It wins in nearly every test, save for the geometry-heavy ones where NVIDIA’s GPUs are traditionally strong. The RADEON 9600 XT is most impressive in pixel shader benchmarks where it simply has no rivals. As for the NV31, it is surely an out-dated product, now that the much faster NV36 is available.
The performance of the NV36 is quite high, especially in geometry-rich tests and in high resolutions (fast memory!). On the other hand, it is still slow at executing pixel shaders, despite of the optimizations in the new ForceWare driver.
Still, I think you don’t care as much about synthetic benchmarks as about real applications – the games you play today. We have benchmarked the new GPUs in several modern computer games.
This time we had a thorough overhaul of our set of benchmarks. All obsolete games were excluded and replaced with new ones that are intended for modern GPUs. Here’s the full list of the applications we are going to use:
We ran our own demos in every game, save for Unreal Tournament 2003 and Splinter Cell, to avoid the “optimizations”. The graphics quality settings were at maximum in every game. There are two work modes: “pure speed” and full-screen anti-aliasing (FSAA) 4x plus anisotropic filtering (AF) 8x. For the sake of comparison, we also include the results of the ATI RADEON 9800 PRO.
This game is a multiplayer 3D-shooter, running on the redesigned engine from the original RTCW, which in its turn is the redesigned engine from Quake 3: Arena. The Enemy Territory add-on provides a higher-quality graphics than the original game, thus demanding more from the graphics card.
Recalling NVIDIA’s GPU were fast in Quake 3: Arena, it is no wonder they show an excellent performance in this test, especially in high resolutions. The RADEON 9600 XT is doing quite fine, too. It only loses 1600x1200 resolution to the competitor, “thanks” to the slower memory.
Star Trek: Elite Force 2 also has the Quake 3 engine deep under its skin.
The RADEON 9600 PRO and 9600 XT are both running close to each other, while the GeForce FX 5700 Ultra is considerably faster than the GeForce FX 5600 Ultra in the FSAA+AF mode. The GeForce GPUs are outperforming the RADEON 9800 PRO even in the “pure speed” mode!
Seems like NVIDIA is all right? Well, don’t forget that the Quake 3: Arena engine has lived its term. It’s not long until it is never used anymore, being replaced with other, more sophisticated engines.
UT 2003 is a more advanced game compared to Enemy Territory and Elite Force 2. Its engine can render complex and beautiful graphics, using all the capabilities of current graphics processors.
The RADEON 9600 XT is close to the GeForce 5700 Ultra in the pure speed mode, while the RADEON 9600 PRO is about as fast as the GeForce FX 5600 Ultra. When we enabled FSAA and AF, the RADEONs went far ahead.
The GeForce FX looks better than the RADEON in the Antalus demo, perhaps we witness the ForceWare 52.16 driver in action? Again, FSAA and AF change the picture: the RADEON 9600 XT is on top, while the GeForce FX 5700 Ultra can only compete with the RADEON 9600 PRO.
HALO is a real masterpiece, at least in graphics. This is truly a representative of the new generation of computer games. After being ported from Microsoft’s Xbox to the PC, the game didn’t lose any of its charm, but gained a lot. Its engine was redesigned to use the DirectX 9.0 API, particularly, version 2.0 pixel shaders. We ran the test in this mode.
Unfortunately, the game refused to work correctly with full-screen anti-aliasing, providing inadequate results with the RADEON 9600 XT. When we tried to run it on the GeForce FX, it just messaged us that the FSAA mode is incompatible with those GPUs. Problems with FSAA may occur because the game was originally developed for XBOX console that cannot use FSAA patterns supported by the GeForce FX or the RADEON graphics processors.
NVIDIA explained that it had asked game developer to forbid FSAA for its hardware, which is why there is no way to enable it.
As you see, the RADEON 9600 XT suits most of all for playing HALO, although the GeForce FX 5700 Ultra is doing well enough thanks to its new drivers.
This game features a heavily enhanced engine, Lithtech, and supports DirectX 9.0 API. If you didn’t play it, here’s the story in brief. The game takes you into the setting of the world’s first computer animation, Tron. The player is to get inside of a computer system and live a full and happy life of a… computer program! The game is rather naïve, but sends you back to those happy times when computers were so much easier. The game world of Tron 2.0 doesn’t use complex textures, but does employ shader effects. Thus, it makes a good modern benchmark.
The game doesn’t support 1280x1024 resolution, offering you the not very common 1280x960 resolution instead.
In the pure speed mode, the RADEON 9600 XT and GeForce FX 5700 Ultra strike up a fight. The RADEON 9600 PRO looks good enough, while the GeForce FX 5600 Ultra falls far behind other participants. FSAA+AF bring what they should: the GPUs from ATI Technologies handle this mode much better. The GeForce FX 5700 Ultra can only reach the RADEON 9600 XT in 1600x1200 resolution, but this victory is practically useless since 17 frames per second are too low anyway. The RADEON 9800 PRO only provides an acceptable gaming performance in this mode.
The following test seems the most interesting, since it is a game of the next generation on a revolutionary engine, fully utilizing the capabilities of the DirectX 9.0 API. We benchmarked the cards using two demos, recorded by ourselves.
Much to our surprise, the GeForce FX 5600 Ultra is running faster than the GeForce FX 5700 Ultra. Is it an error of some kind or something? Well, no. The game simply didn’t run in the DirectX 9.0 mode on the GeForce FX 5600, switching to the DirectX 8.1. The GeForce FX 5700 Ultra managed to run the game in the DirectX 9.0 mode… It should better not do it. The card doesn’t produce enough fps to play the game in the proper way even in 1024x768, not mentioning the higher resolutions. The much-hoped-for ForceWare driver couldn’t help the NVIDIA GeForce FX 5700 Ultra out of the mire. Moreover, it gave out a peculiar artifact: many object shadows looked like black rectangles, which we didn’t see when using the Detonator 45.23. Of course, this bug will surely be corrected in the future version of the ForceWare, but this is the fact: the new driver may bring artifacts into some games. Well, this game is not yet released, so its developers bear a part of the blame.
Meanwhile, the RADEON 9600 XT allows you to play in 1280x1024! However, if you are a hardcore gamer, I would recommend you to pay attention to high-end products like the RADEON 9800 XT and 9800 PRO first.
This game should be well known to you, as we have been using it in our tests for a while already.
Once again, let me remind you that the results shown by NVIDIA’s GPUs in the FSAA mode should not be considered correct. The new driver didn’t eliminate the problems. Take a look at this screenshot – the protagonist and the grid don’t cast the shadow to the wall. Compare this to the screenshot taken on the RADEON 9600 XT, which has it the right way. NVIDIA explained us that the issues with Splinter Cell resemble the problems there are with HALO, so, the FSAA mode cannot be easily enabled on NVIDIA’s GPUs (we had to rename Splinter Cell exe file to turn the FSAA on, otherwise the driver automatically disables the full-scene-antialiasing).
For the sake of truth I have to confess that even though there are a lot less issues with FSAA and ATI Technologies’ VPUs, the RADEON hardware also does not work with the game absolutely flawlessly when FSAA is enabled – we mentioned about this in our ABIT SILURO GeForce FX 5600 Ultra DT review earlier this year.
Alas, I say it once again that the new drivers from NVIDIA bring both higher performance and a number of visual artifacts. Hopefully, the software developers of the company will solve these problems soon.
This game looks much similar to Splinter Cell, but offers somewhat more flexible graphics settings. We set the game up to use 32-bit textures only, and disabled shadows. The shadows use 2.0 shaders and wouldn’t work with the GeForce FX cards.
We have got used to this already. The RADEON 9600 XT puts another victory on record. The GeForce FX 5700 Ultra is even no match to the RADEON 9600 PRO in this test. The GeForce FX 5600 Ultra looks a relic of days long past: its 21fps in 1024x768 is not what we expect from a modern mass graphics card. When FSAA was enabled, the game wouldn’t run on the GeForce FX at all in resolutions higher than 1024x768. There were no error messages: the display just went black to reveal the Desktop in a second. One more minus into the NVIDIA driver’s piggy-bank.
This is one of the most advanced Formula 1simulators. If you are a fan of this sport, you will have a lot of fun playing the game. We are more interested, however, how well this game suits for our testing needs. One guy from our labs, being much into Formula 1, drove a full lap under bad weather conditions. The record was then used for testing with the help of the FRAPS utility.
The game was rather strange when running on the ATI VPUs: the number of fps has never grown above 45, notwithstanding the disabled VSync. This is the reason for the relatively low results shown by the RADEON 9800 PRO. The GeForce FX 5700 Ultra had a glorious performance here. When we enabled FSAA and AF, the situation changed: the cards on the ATI chips left the competitors behind.
X2 is a well-done space simulator from EGOSOFT, upholding the glorious traditions of Elite, X-Beyond the Frontier and X-Tension. The game is now in the development phase. So far, we have only a short demo, which shows the game’s capabilities. Anyway, it suits well for testing graphics cards.
The GeForce FX 5700 Ultra is a little better than the RADEON 9600 XT in the “pure speed” mode, but falls behind when FSAA and AF are enabled.
Although this test seems to be synthetic, it is in fact based on a real DirectX 9.0 game engine, known as “krass”.
The new drivers and high frequencies of the GeForce FX 5700 Ultra allow it to get to the level of the RADEON 9600 XT in the “pure speed” mode. However, the RADEON handles FSAA and AF better and wins the second work mode.
This time, we will run the gaming tests from the Futuremark 3DMark03 suite.
The first test of the suite is quite simple, using only the DirectX 7 API. So, the victory of the GeForce FX 5700 GPU looks logical enough. However, the RADEON 9600 XT is again faster in the FSAA+AF mode.
The second gaming test from the 3DMark03 suite is more sophisticated. The numbers are smaller here, but the overall situation remained the same: the GeForce FX 5700 Ultra wins the “easy” mode, while the RADEON 9600 XT is best in the “hard” one.
The new ones from NVIDIA and ATI are running neck and neck in the third gaming test until we enable FSAA. After that, the ATI VPUs win, although the result of 15 fps looks ridiculous.
The fourth gaming test is a deadly fear for any graphics card. Traditionally, NVIDIA’s GPU were helpless here. This time they are doing better: the new driver improved the situation. Anyway, the RADEON 9600 XT wins in both modes.
So, we have run the benchmarks for you to see the results. The new VPU from ATI Technologies is feeling confident in most currently available games. I can’t say the same about the new GPU from NVIDIA. However, the NV36 has its chance in OpenGL games as well as in games with poor textures but complex geometry. Besides that, this GPU shows good speed when no full-screen anti-aliasing and anisotropic filtering are used, as well as in high resolutions. However, this largely depends on the specific application.
The matter of price can be waved aside – both cards cost about the same amount of money. As for pure usability, the ATI RADEON 9600 XT looks advantageous, too. It takes less space in the system case, produces less heat and noise and requires no additional power. The GeForce FX 5700 Ultra may be interesting to people who are not much into gaming as well as for those who want to have a graphics card with two DVI-I outputs. And of course, all hardcore fanatics of NVIDIA may find the GeForce FX 5700 Ultra an interesting solution for $200. Although our card had only one digital output, there will surely be a model with two of them in the market. The owners of a GeForce FX 5600 Ultra may find themselves behind the times – this GPU may not be able to run the upcoming DirectX 9.0 games properly.
For those of you who are looking forward to the release of such games as Half-Life 2 or S.T.A.L.K.E.R: Oblivion Lost, but who can’t afford a high-end card, the RADEON 9600 XT may suit just fine. The owners of the RADEON 9600 PRO may not bother much about the new VPU. Instead, they can try to overclock the graphics core, because the main difference between the RADEON 9600 XT and the RADEON 9600 PRO is the operational frequency.
ATI RADEON 9600 XT
Highs:
Lows:
eVGA GeForce 5700 Ultra
Highs:
Lows: