Introduction

On 5th October this year the Canadian chip manufacturers ATi presented the latest generation of graphics card in the form of the Radeon X 1800, X 1600 and X 1300 and the need for discussion in various forums was, as usual, extremely high. So, the R5x0 chip provides among other things, a famous quality of the anisotropic filter – as long as you on previous ATi products, as well as on the NV4x and the competition from nVidia G70 refers this statement – and knows how to convince also the feature set. So, support the shader model 3.0 finally accordingly equipped video cards and are also able to use this in combination with multi-sampling-anti-aliasing even high dynamic range rendering.

Therefore it has the GeForce 7800 series even a feature ahead, which is why the latest addition to ATi was justly highly praised. Only on the performance, it kränkelte something. This was although overall at a very high level, the flagship of nVidia, which could GeForce 7800 GTX, but are only just beaten in high-quality settings. Without anti-aliasing and anisotropic filter – here, but must say that in these settings probably few with a high Accelerator play – lost the Radeon X 1800 XT even against the fastest product of G70.

Now the presentation of the Radeon X1k series is already a few weeks and within this short period of time much in the driver development has been done already. As ATi could accelerate 5.10a significantly the OpenGL performance with the catalyst and the performance-eating game “Fear” is running on the new graphics cards now significantly faster than later in the month of October. In addition, we had launch driver still with some partly to fight disturbing errors and want to control, whether these were now cut off. But of course also the competition does not sleep and released eager new ForceWare driver, we are also control with regard to the performance.

Still, the article will deal also with the new dual-core CPU of AMD, the Athlon 64 X 2. This has made its way into our test lab, and will need to show there, what get the dual-core optimizations from the ForceWare drivers for nVidia graphics cards.

As additional specialities, we can also present a first beta driver from the House of ATi, which should benefit from the new CPUs. But that still not enough: In addition we have taken the image quality and performance in HDR rendering on a Radeon X 1800 under the microscope. This ATi has can make kindly an early version of the beta patch for far cry us available, which makes it possible not only HDR on a R5x0 in the development of CryTek, but also in addition can be anti-aliasing.

Furthermore, this article contains also a comparison between a 256-MB and a 512 MB card made by nVidia and ATi. Already today is worth the extra charge for the larger VRAM or you can save some easy euro with the 256 MB version? We would like to measure not only performance, but want to measure memory usage in current 3D games on the basis of a tool. To start the test!

Test system

Test system:

• Processor
  • AMD Athlon 64 4000 + (San Diego core, 90 nm, SSE3, 1024 kB level 2 cache)
  • AMD Athlon 64 X 2 3800 + (Manchester core)
  • AMD Athlon 64 3800 + (NewCastle core)
• Motherboard
  • Gigabyte GA-K8NXP-SLI (nVidia nForce 4 SLI, Socket 939)
• Memory
  • 2 x 512 MB crucial BallistiX DDR400 (2-2-2-5)
• Graphics cards
  • ATi Radeon X 1800 XT (625 / 750)
  • ATi Radeon X 1800 XT (500 / 500)
  • ATi Radeon X 1800 XL (500 / 500)
  • ATi Radeon X 1600 XT (590 / 690)
  • ATi Radeon X 1300 Pro (600 / 400)
  • nVidia GeForce 7800 GTX 512 (430 / 600)
  • nVidia GeForce 7800 GTX (430 / 600)
  • nVidia GeForce 7800 GT (400 / 500)
  • nVidia GeForce 6600 GT (500 / 500)
• Peripherals
  • AOpen AAP-1648Pro-DVD drive
  • Samsung S-ATA 2 HDD 200 GB disk space (NCQ enabled)
• Driver versions
  • nVidia ForceWare 78.03
  • nVidia ForceWare 81.87
  • nVidia ForceWare 81.94
  • ATi 8-173-1-050921a-026915E (beta driver for Radeon 1 x 00)
  • ATi Sample_8-201-051122a-028804E (beta-version of the dual-core driver for ATi graphics cards)
  • ATi catalyst 5.10a
  • ATi catalyst 5.11
• Software
  • Microsoft Windows XP Professional SP2
  • “Video memory Watcher” by nonatainment
  • Microsoft DirectX 9.0 c

Benchmarks

The following benchmarks were used during our test:

• Synthetic benchmarks:
  • 3DMark 01 SE version 330
  • 3DMark 05 version 1.2.0
  • Aquamark 3
• Game benchmarks:
  • Age of Empires 3 demo
  • Earth 2160
  • Far cry version 1.3
  • Splinter Cell: Chaos theory
  • Half-life 2
  • Fear multiplayer demo
  • Serious Sam 2 demo
  • Doom 3
  • The Chronicles of Riddick

All benchmarks run with maximum details, so that the graphics card will be charged as high. As settings we chose this 1280 x 1024. In addition to the pure resolutions we can go through anisotropic filter the benchmark course with two anti-aliasing as well as 16-fachem, because these are often occurring quality settings. The SLI benchmarks and tests for memory usage go we see 1600 x 1200, however, as well as 4 x AA and 16 x AF.

Attention:After careful consideration and multiple analysis of even recorded game sequences, we have come to the conclusion in the ForceWare driver for nVidia cards increase the quality settings on high quality, because we can effectively combat the texture flickering only with this setting. In addition, this mode is similar to the setting “catalyst a.i. standard” on the ATi counterparts, resulting in largely a tie in image quality.

Driver settings: nVidia video cards

• Performance: High quality
• Vertical sync: off
• Force MipMaps: no
• Trilinear optimization: from
• Anisotropic mip filter optimization: from
• Optimization of anisotropic pattern: from
• Negative LOD bias: clamp
• Gamma adjusted AA (G70): A
• Transparency AA (G70): Off

Driver settings: ATi video cards

• Catalyst a.i.: Standard
• Mipmap detail level: high quality
• Wait for vertical refresh: always off
• Adaptive Anit-aliasing: off
• High quality AF: off
• Truform: Always off

Pixel shader performance

While nVidia for the NV40 chip in the form of the GeForce 6800 series, the shader model dominated 3 and hence the pixel shader version 3.0, the performance of ATi took significantly longer. With the launch of the Radeon-X 1000 series the Canadians have now ventured same step and also offer the shader version. Both the implementation of the pixel shader 3.0 is identical to the popular nVidia-as well as on the ATi cards. Because unfortunately cannot easily it is possible to measure only the pixel or vertex-shader performance, with special programs we have limited ourselves to four current 3D applications, where between the shader model 3, the predecessor to the version of SM2 can be switched. All measurements were made in 1600 x 1200 without anti-aliasing and anisotropic filter.

The 3DMark05 the GeForce is 7800 GTX before the Radeon X 1800 XT, what you can tell by the high pixel shader performance due to the two performance shader units per pixel pipeline. Both cards can benefit in the synthetic benchmark quite clearly from the more modern shader instructions. A specific pixel shader-test, which is found in the game earth 2160, shows the GeForce 7800 GTX of a Radeon X 1800 XT-mile, consider what you can isolate but probably to a driver error. Thus, the results are not comparable.

In Splinter Cell 3 Radeon X is generally superior to the nVidia equivalent 1800 XT. Nevertheless, both maps of the shader model 3 can score low compared to the shader of version 1.1. In addition it is possible, to enable the shader model 2.0, ATi card which shows a slightly better performance, the game section, but has a worse image quality. The “Pier”level in the shooter far cry can also easily benefit from the shader model 3, and also again the ATi adaptation front lies. The distance turns out but less than in Splinter Cell 3.

Vertex-shader performance

The vertex-shader units of R5x0 are compatible with the shader 3.0 specifications (such as those of the G70 chip by nVidia). But ATi has exploited a small gap in the specifications, which the Radeon X1k chip does not have a feature: “Vertex Texture Fetch”. Vertex texture fetch allows that the vertex shader can read textures directly from the texture memory without unnecessary detours. The R520 not dominated this feature however, as ATi had to choose according to their statements between the multiple benefits by VTF and needed transistor usage. Also vertex shader fetch would reportedly increase the latency what the developers wanted to avoid.

Elderly native ATi used a procedure called render to vertex buffer, short R2VB. This is to have the same effect as VTF, taking the pixel shader must take on this task. The pixel shader writes the data to a buffer in memory, which can then in turn be read by the vertex shader. Since render to vertex buffer until now but still not an integral part of the DirectX specification, must the developer much like in the feature “Geometry instancing” on a Radeon X 8 x 0, that previously query hardware, if R2VB exists. Eventually, these tasks in later versions of the catalyst of the driver is taken. Lack of vertex texture fetch will be expected to headaches the average user at any time, because almost no game uses the feature at this time. Only exception is “Pacific Fighters”, which runs on ATi cards in the highest level of detail.

Because, as already mentioned, it is hardly possible to measure the vertex-shader performance without depending on other resources, this section with regard to the benchmarks will be very shortly. As a small note we have let pass through the vertex-shader test in the 3DMark05, in which the Radeon X 1800 XT as expected the GeForce 7800 GTX far behind can. This is the significantly higher geometry throughput making noticeable except in the 3DMark05 – but hardly a game.

High dynamic range

“High Dynamic Range Rendering” is most certainly one of the features which will allow a grandiose splendour of graphics in games in the future. The player with HDRR will enjoy an extremely high colour gamut, high color accuracy, a dynamic adjustment of light on the reaction of a human eye when looking in a light source and much more – first tasting for a (nowadays) full dynamic range rendering are to examine in far cry and splinter cell 3.

The NV4x-chip made by nVidia supports as first GPU HDRR, which is realized by FP-blending enabled ROPs and FP16-enabled TMUs in hardware. Another important point the name “FP Filtering” and allows you to filter floating point textures on request trilinear or bilinear. The NV4x and its successor G70 support FP filtering, the R520, as well as the smaller offshoot of ATi are but all not in a position to filter FP textures – they can only read them. Therefore, HDR applications manually, say by the software developer, must be adapted. The R5x0 has no problems, however, when you filter by integer textures (FX8 or FX16), which appear but with a lower precision. Because all the currently available games which support HDRR, put on FP16 filtering, this rendering must be – amended first happen is this, for example, in the current beta patch to far cry.

The alternative for the game developer is the filtering of the textures using pixel shader, which is significantly more complex and may also be slower. It is also possible that the calculations block the pixel shader for some time. Positively noted the possibility for FP16 HDR on a Radeon X is X, however, 1800, 1600 or X 1300 to activate multi-sampling-anti-aliasing. This is not possible on a G70 in return. ATi has justified the lack of FP filtering with a design. It was according to own statements as possible to save transistors and to allow the MSAA at HDR.

Performance

To measure performance when enabled dynamic range rendering, we selected the games Splinter Cell 3, serious Sam 2 and far cry, that explicitly support this feature. ATi was able to kindly provide us with an early beta version of a patch for far cry which HDR on a R5x0 allows. With this patch the game of CryTek developers on a Radeon is also the first 3D application,-X 1800 card HDR including multi-sampling aliasing allows. Being again the Radeon X used 1800 XT and the GeForce 7800 GTX for this comparison. As a resolution, we opted for 1600 x 1200.

In Splinter Cell 3, the victory is clearly on the map from Canada. Radeon X 1800 XT is much faster than the equivalent of nVidia with enabled dynamic range rendering. Fortunately, a liquid playing in HDR mode is possible on both graphics cards. A similar picture emerges in serious Sam 2, only the manufacturer have swapped roles. ATi has little problems, get on with it 7800 GTX to the GeForce both with and without HDR. This plays only a minor role in however in the cards, because none of the two in this resolution can be play with HDRR liquid.

In far cry both producers with their top models in 1600 x 1200 without anti-aliasing, anisotropic filtering, and high-dynamic-range rendering at a performance level, whereas the result enabled transitions effects changes. The GeForce 7800 GTX is clearly in the lead, and you can here the Radeon X 1800 XT no chance – we suspect that some reserves available in newer drivers are ATi. Declared with in addition the MSAA of is war while effectively the edges of the polygon, the performance decreases logically but further into the cellar. On the ATi card, you can play far cry thus not smooth in the selected quality settings. Nevertheless, ATi receives a bonus point to offer still anti-aliasing enabled HDRR. The playability with a GeForce 7800 GTX is on the razor’s edge and is dependent on the level, as well as the situation.

Image quality

As you can see from the screenshots, the benefits of high dynamic range rendering in serious Sam 2 is very low, since the game uses our information no dynamic “tone mapping” effect. Thus, no “getting used” to the high light intensity is simulated, which is why the strong brightness decreases after a few seconds, as is the case, for example, in far cry. The Radeon X 1800 has on a minimally different image quality to the GeForce 7800. However, you cannot talk about a better or worse image. In serious Sam 2, we recommend to leave the HDR option generally turned off. The only minimally better image quality is in no relation to the low performance.

A completely different result shows, however, Splinter Cell 3, in which there is already clearly visible differences between the two graphics cards without HDRR. So, among other things the spotlight on the R520 is in a unnatural picture. The differences are even more apparent with enabled HDDR. The entire image appears brighter on the GeForce 7800 and better lit than on the ATi counterpart.

In the “Jungle shooter” far cry, there are only slight differences in image quality without HDR rendering. So the end of the path to the R520 is not displayed properly, prepares the G70 nVidia problems when the shadow appearance on the weapon that unnaturally hard fails on the other side. Otherwise, there is however no significant differences. The same effect as in splinter cell is evident also in far cry with enabled HDRR. The entire image appears brighter lit on the GeForce 7800, just as the Sun would seem strong. When the Radeon X 1800 the entire image is too dark, although the settings regarding the HDR mode ‘ are completely identical. The image on the R520 with HDR and the anti-aliasing, which is technically not possible on a GeForce 7800 is worth a look.

Our theory for different lighting at HDRR: It is located on the procedure in the “tone mapping”. Dynamic tone mapping converts the generated image into a so-called LDR (low dynamic range) image on the basis of the overall brightness, as there is currently no monitors that have a high contrast to HDR without represent tricks. While that simulates tone mapping the response of the human eye, which is only needs to set to the high difference in light – is thus reached in far cry and splinter cell 3, that the image, after it has looked in a light source, again abdunkelt is. Serious Sam 2 used, however, no dynamic, but only a static tone mapping. As the game shows no difference in brightness with enabled HDR on two graphics chips, the conjecture makes sense that other calculation of tone mapping the different illumination is responsible.

256 MB vs. 512 MB

In the latest high end Accelerator, the question whether you should grab a cheaper 256 MB model, or better the more expensive to 512-MB variation arises now. So the top models are delivered 1800 7800 GTX 512 by default with a 512 MB VRAM XT and the GeForce from ATi and nVidia, Radeon X, from the ATi card, there will be also a 256 MB card. The choice has the buyer and the situation is not unknown, after all there was same discussions also when moving from a 64 MB card on a 128 MB adaptation and as well when deciding whether 128 MB or 256 MB be right now. The same crystallized this has in both cases: when the direct purchase is better off because of the cost savings with the “smaller” RAM size, half a year later the whole might be but again different. We hold 512 MB mid-range or even low end offers, however, for a long time as useless. The question of whether the increased texture memory in today’s 3D games already pay, we will cover in this section.

VRAM usage

First, we want to measure the RAM usage of the video card in current games using the “Video Memory Watcher” by nonatainment. This, however, we must point out that there can be depending on the application to incorrect results, as some games have a “strange” way of dealing with the memory of the graphics card and these vollschreiben until there is no space. As a quality setting we opted 1600 x 1200, in combination with two anti-aliasing for the resolution, so that the textures are as large and maximum saturate the memory. In addition, we will add 6xAA ATi and nVidia 8xSAA to the measurements, since both modes are officially available in the driver.

Graphics card XT and the GeForce are as the Radeon X 1800 a 512-megabyte memory have since both adaptations 7800 GTX 512 used.

In the first six Testapplikationen, the magic 256-MB limit is reached mostly not even close. The exception is the 3DMark05 da, is also using 6xAA at ATi or 8xSAA at nVidia – at the traditional 4xAA still easily fit the entire textures in the 256 MB framebuffer. The difference with the RAM allocation between the ATi-and the nVidia graphics card, which should not be actually is striking. This is a procedure called “Filter@Scanout”. Means: nVidia operates only in the Scanout, the Downfiltering which you while memory bandwidth can save, consumption of VRAM but is increasing. In age of Empires 3 the ATi card in conjunction with the memory video watcher crashed regularly and in Splinter Cell 3 we could 7800 GTX no 8xSAA set on the GeForce, which is why these measurements could not be performed.

The measurements in the second part of the test matches, which consume almost more memory than 256 MB depending on the setting are much more interesting. So, far cry, half-life 2, serious Sam 2 and Doom3 at four x anti-aliasing on the nVidia card almost consume all memory on a 256 MB card. It is still critical at 8xSAA and 6xAA, as must for these settings need to almost all the games over a 256 megabyte of VRAM and outsource the textures accordingly. Also here is the difference between the two producers to once again clearly see. Especially with Doom 3 and Riddick, the difference is glaring.

In theory a 512 megabytes large VRAM is so quite reasonable – depending on the application but only, as long as extremely high-quality and memory-eating-quality modes are enabled. Whether is the measurements in games benchmarks reflect, is clarified in the next section.

VRAM speed

Even if current games theoretically already benefit from more than a 256 MB memory on a graphics card, the practical profit will need to show only in games. Testing 7800 GTX and a Radeon X are XL a GeForce 1800 with a 256 MB frame buffer used. 512-MB-Boliden XT and the GeForce be used as a Radeon X 1800 7800 GTX 512 on the basis of 512-MB memory, which are each heruntergetaktet to the frequency of the 256 MB cards. We use 1600 x 1200, in connection with four x anti-aliasing and 16-time anisotropic filtering quality settings. Say we have that the simulated cards may be minimally differ from the performance of a “true” 512 MB version with lower frequency, this well can be equipped with other memory timings.

In the first part of the 3D games can benefit from their larger VRAM the 512 MB graphics card, and is almost go through at the level of the 256 MB version. Only the 3DMark05 shows a minimal performance gain by the larger frame buffer, but only on the ATi card and there’s lead is just once just 110 points. Interestingly, the 512 MB card from ATi cuts better than the equivalent from nVidia to the cards with the smaller store off something. This suggests that ATi to nVidia not has reduced the memory timings on the other hand.

Largely an identical image is reflected in the second part. However, there are also some exceptions. Half-life 2 surprisingly shows a slight performance gain by the 512 MB frame buffer, regardless of whether the ATi-or the nVidia derivative is used. This lead turns out approximately 1.5 fps but very low. Also benefit from the larger store fear – what one could – also suspect as a result of the measurements of memory consumption, but also here the difference is one or two FPS at low. Here we had expected quite a large difference. Also Doom 3 shows a small increase in the speed, but only on the X 1800. Same result is evident once again at the 3DMark05.

Overall, the larger VRAM in today’s games brings what will change in the future however still no (or no more than a minimum) plus performance. If the newly acquired high end graphics card you want to keep some time, a graphics card with 512 MB of memory is already useful, otherwise you can easily access to a cheaper Variant.

“Dual core” driver

The new dual-core processors from AMD and Intel are currently hotly sought-after topic. The CPUs but provide a disadvantage: in current software that mostly has no adjustments for two-core processors, AMD Athlon 64 X 2, nor the Intel Pentium D bring a performance benefit. This seems nVidia but not held to have to adjust the ForceWare driver of version 81.xx for GeForce graphics cards on the dual-core CPUs, which the after game more performance from the GPU out to tickle is quite a few percentage points. We have already given a first impression of performance of the driver on a Pentium D processor, now follows a post-test on an Athlon 64 X 2.

As a special treat, we can present first benchmarks of an early beta version of a customized driver for ATi graphics cards, the Canadians have kindly made us available. This driver is based on the official catalyst 5.11 according to the Catalyst Control Center, however, still some optimizations for dual-core processors were added. NVidia we use 7800 GTX the ForceWare 81.94 on the GeForce, while for ATi that being 1800 – used Radeon X XL with the new beta driver a Radeon X 1800 XT us at the time of the test no longer available. As a two-core processor we use an Athlon 64 X 2 3800 +, while the single-core variants an Athlon was chosen by 64 3800 +, which was heruntergetaktet to the beat of the X 2 3800 +, i.e. 2,0 GHz. Unfortunately, we have to say that the Athlon based 64 3800 + still on the old NewCastle core, which has a slightly slower Memorycontroller.

Performance part 1

In the first part of the 3D games, the dual-core driver in the resolution of 1280 x 1024 without anti-aliasing and anisotropic filter, especially in the 3DMark2001 SE, Aquamark3, and in age of Empires 3 bring a strong increase in speed. In the latter, this is the case but only on the nVidia card since the Radeon X 1800 works on XL same performance level as with the single-core CPU. On the other hand the ATi card can beat, even if there are only two frames per second, 2160 profit out of the Athlon 64 X 2 in Earth, while there the ForceWare driver not a plus in speed brings.

In the second part the nVidia card boasts on and serious Sam 2 and Doom 3 WINS in far cry, half-life 2, some FPS. The beta-Catalyst driver be able to see only in far cry and half-life 2, in addition he works 64 3800 + in fear slightly faster than on the traditional Athlon.

Overall the GeForce can benefit more as long as no anti-aliasing 7800 GTX by nVidia from a dual-core CPU as ATi with the early version of catalyst – and no anisotropic filtering is used. Here, however, we must say that there is still hope for the ATi graphics cards because the driver is probably still in an early stage of development.

Performance part 2

Additional quality increase by 4-faches anti-aliasing and 16-time anisotropic filtering, same results as without the quality-enhancing features show up in the first part of the 3D applications, only these drop out percentage slightly lower due to the higher GPU load. So can both the Radeon X 1800 XL, as well as the GeForce 7800 GTX in the 3DMark2001 SE and Aquamark 3 of the Athlon 64 X 2 benefit, while in age of Empires 3 only the nVidia card and some frames per second can win in Earth 2160 only the ATi equivalent.

In the second part most games it is irrelevant whether a dual-core processor is used, or only a single core CPU. In far cry, the two chip giants can benefit from a corresponding CPU still quite well, while shrinking profit in half-life 2 to only one or two frames per seconds. Age of Empires 3 remains a domain of the Californian with 4xAA and 16xAF, while fear 1800 shows now virtually no performance difference on the Athlon 64 X 2 and the Radeon X XL by ATi more.

The better quality settings show that the CPU plays only a minor role in these settings. Both producers can benefit no longer of the two core processor from ATi is again nVidia easily – but here is the note that the Canadians on newer versions of the driver can get expected to somewhat more speed.

Driver comparison

A new generation of smart cards – how the new R5x0 from ATi is largely a is – offers initially not only (usually) some new features, but also a driver, which is rather less mature; When the Radeon X1k series. Now, some weeks after the presentation have already passed and the Canadian manufacturer has significantly developed the catalyst drivers so that there are three executable versions: the launch driver, a beta driver called catalyst 5.10a, as well as the official catalyst 5.11. While the leek driver from Ibiza not only with performance problems, but also with problems at the power and the image quality has to fight, the catalyst 5.10a at least at the power brings relief. ATi from this driver has the “Memory Mappings” in the new basic storage controllers (not to be confused with programmable!) optimized, which is why OpenGL games relating to four x anti-aliasing run significantly faster. Without anti-aliasing or Direct3D there are no basic changes of speed against it, however, they can follow through further improvements.

The catalyst 5.11 is identical to its predecessor while close to one hundred percent, but there was still some performance changes. As the three drivers when compared to the competition the ForceWare drivers (ForceWare 78.03, 81.89 and 81.94) beat be, is clarified in this small comparison, which deals not only with the performance, but also with the image quality.

Image quality

As you can see from the screenshots, the ATi Catalyst driver in the synthetic benchmark 3DMark05 from the House of Futuremark provides an inferior quality of the representation of the smoke. It has a coarse, wrong gradient and is represented on a GeForce 7800 graphics card from nVidia korrekt(er). Also ATi chips of the R4x0 generation have no problems with the representation. Otherwise there are no problems with the image quality in the benchmark both on the Radeon-X 1800-, as well as on the copy of GeForce 7800. Since this is a synthetic benchmark, we can ignore these difficulties, in addition, ATi is already informed of the error.

Nanu, what’s that? Not bad, as was the famous Aquamark 3 same image errors such as the latest 3DMark-Version we extol. Also here the Radeon X displays 1800, a wrong gradient in contrast to the GeForce 7800. Regardless of whether the launch driver or the catalyst 5.11 is used. Thus the Fri comes naturally