by Alexey Stepin
11/04/2004 | 11:05 AM
Right now the market of top-end graphics cards feels a kind of shortfall as products on RADEON X800 XT Platinum Edition and GeForce 6800 Ultra graphics processing units are seldom spotted in shops. NVIDIA’s problems with the GeForce 6800 Ultra are well-known, while the lack of the RADEON X800 XT is rather strange as ATI seems to have mastered its tech process well enough to ensure the necessary chip yield.
<%BANNER[article]%>Whatever the reason, the RADEON X800 XT is really shipped in very small quantities and its price is quite naturally high, lying in a range of $500-700. All we can do is to hope that this situation will be changing for the better, and there will be plenty of RADEON X800 XT cards that offer the highest performance to their users. So far only enthusiasts who don’t mind the price have been hunting for the rare available samples.
Getting closer to the subject of our today’s review, we should confess that we can’t appreciate the reference cooling system of the RADEON X800 or its slightly enlarged variant from ASUS. These cooling designs are good at handling the RADEON X800 PRO, yet they can hardly cope with the hotter RADEON X800 XT chip. If the card is working in a cramped PC case without additional system fans or if the weather is hot, the card may often overheat and hang up.
Saying so, we are basing on our own experience with different graphics cards on RADEON X800 series chips. Some manufacturers like Sapphire or HIS realize the urgency of this problem and equip their cards of that class with powerful and efficient cooling systems, but we are left wondering why ASUS hasn’t done the same (you may remember that ASUS once created an excellent cooling system for their version of the RADEON 9800 XT, for example). Of course, such a cooling system would increase the cost of X800-based cards somewhat, but such devices are anyway positioned by the Taiwanese company as top-end and expensive solutions for hardcore gamers and PC enthusiasts who don’t care much about the price factor.
A little while ago we published a review of the ASUS AX800 PRO/TD graphics card (see our ASUS AX800 PRO/TD Graphics Card Review). Being an overall excellent product with rich accessories and excellent 2D image quality, the card was not completely free from drawbacks like the aforementioned imperfect cooling system and the lack of heatsinks on the memory chips.
Of course a major company like ASUS couldn’t limit itself with making graphics cards on the junior X800 series chips only, so they soon added a major 16-pipelined model into the ranks. So, welcome the ASUS AX800 XT/TVD graphics card as it is going to be the subject of this review. We will pay special attention to the cooling system of this card to see if the manufacturer corrected the defects of the reference cooling design.
Like the ASUS AX800 PRO, the AX800 XT came to us in its retail package – this huge-size box has already become a symbol of top-end products from this manufacturer.
This time the company came up with a unified package for the whole AX800 family. Like the AX800 PRO, the AX800 XT card from ASUS is accompanied with luxurious accessories:
Instead of S-Video and RCA cables the package contains a special splitter which is attached to the all-purpose VIVO connector. It’s not as handy as the switch unit enclosed with ASUS RADEON 9800 XT/TVD cards, but makes its job done, anyway.
The contents of the orange plastic package with software haven’t changed since the AX800 PRO. We’ve got the following discs here:
This time it’s quite natural to see video-editing software bundled in as the ASUS AX800 XT/TVD is equipped with a Rage Theater chip capable of capturing video signal.
Like the AX800 PRO, the reviewed card supports ASUS’ exclusive technologies GameFace, Video Security and Smart Doctor that we described in our previous reviews of graphics cards from this manufacturer (you can refer to our ASUS V9950 Graphics Card Review, ASUS RADEON 9800 XT/TVD Review in our article called ASUS RADEON 9800 XT and LeadTek WinFast A380 TDH Ultra: The Battle for AquaMark3, or ASUS AX800 PRO/TD Review for details). Also there you can get a description of the web-camera included as an accessories bundle.
The ASUS AX800 XT/TVD differs but slightly from the junior model and that’s normal as all X800 family cards use the same reference PCB design.
In this particular case, all outward difference from the AX800 PRO boils down to the AX800 XT’s having a yellow connector near the additional power plug and a Rage Theater chip with its smaller companion logics. The yellow connector looks like an audio input but is actually intended for attaching an additional video input which can be placed, for example, on the front panel of the system case.
Besides these external differences, there are internal ones, which are even more important. Particularly, the ASUS AX800 is equipped with faster GDDR3 memory from Samsung with an access time of 1.6 nanoseconds (against 2ns memory of the AX800 PRO). Like becomes a decent RADEON X800 XT, this one clocks its memory at 560 (1120DDR) MHz and its GPU at 520MHz. All the 16 pixel pipelines are in operation now, too, while the AX800 PRO (and any other RADEON X800 PRO for that matter) has only 12 operational pipelines.
The cooling system has been inherited from the AX800 PRO, with the same disadvantage – the memory chips are not cooled down at all, although ASUS could have easily realized that. The memory on the AX800 XT is clocked at a higher frequency than on the AX800 PRO, so it just demands cooling but never receives it. Running a little ahead, we should confess that the graphics card we tested would overheat often during our tests as the weather was rather hot these days here. Overheat led to frequent hang-ups which we only managed to prevent by installing an additional 120mm fan to blow at the card from its butt end. The air stream was being cut in two by the card’s PCB and was reaching to both front and back sides of the PCB, cooling the memory chips. We would have been spared the trouble if ASUS had put thermal pads on the memory chips under the cooler’s sole and passive heatsinks on the backside memory chips. Considering the more intensive heat dissipation compared to the AX800 PRO, we think the AX800 XT needs to be installed into a system case with additional air cooling and a well-thought configuration of airflows; otherwise you may find your card overheating too often, especially in summer.
All the praises we showered on the ASUS AX800 PRO for its noiselessness and excellent 2D quality can be applied to the AX800 XT, too. The cooler is only heard for a couple of seconds after you’ve powered the computer up, but then it slows down to become almost completely silent. On the GPU hitting a certain threshold temperature, the cooler increases its rotational speed, but this hardly ever happens.
2D quality was no worse than with the junior model: the AX800 XT was outputting a crisp image in all resolutions supported by our monitor, up to 1800x1400x@75Hz inclusive.
Overclocking, on the contrary, was rather average. With additional cooling, we managed to overclock the core to 555MHz as the card was unstable at higher core frequencies. The memory pleased us even less, only speeding up to 570 (1140DDR) MHz. Again, at higher memory frequencies, we had all manner of artifacts onscreen and frequent hang-ups. Water or cryogen cooling might have helped us more, but these are not really inexpensive and simple ways to improve the performance.
We rolled out our standard testbed for this session. The system was configured like that:
And here’s the list of games and applications we use for benchmarking purposes:
First Person 3D Shooters:
Third Person 3D Shooters:
Simulators:
Strategy:
Semi-synthetic Benchmarks:
Synthetic Benchmarks:
In each particular case we selected the best possible graphics quality settings, the same for graphics cards on ATI’s and NVIDIA’s chips.
Call of Duty isn’t a complex game by today’s standards, so don’t be surprised at the three-digit fps rates the modern top-end graphics cards have here. The performance of the ASUS AX800 XT matches the GeForce 6800 Ultra, save for 1024x768 resolution. Overclocking is of a very small effect here.
Like in the pure speed mode, the results in 1024x768 resolution look rather strange, but everything returns to the norm since 1280x1024: the efficient algorithms of working with the graphics memory make the ASUS AX800 XT the leader. Overclocking is more rewarding in this mode, as the memory load is higher.
Doom 3 definitely prefers graphics cards with NVIDIA’s architecture, so the ASUS AX800 XT cannot compete with the GeForce 6800 gang here.
The same goes for the eye candy mode with its enabled full-screen anti-aliasing and anisotropic filtering. The new game engine from id Software has too a strong liking towards NVIDIA’s GPUs, while ATI’s OpenGL driver is rather outdated already. New versions of the Catalyst suite may bring a considerable speed boost to Doom 3 but until that ATI Technologies’ camp cannot boast high performance in this game.
Having only 128MB of onboard memory, the GeForce 6800 couldn’t turn FSAA and AF on in 1600x1200 resolution.
Everything we’ve said above about the Hellhole level is true for the d3dm4 map: graphics cards based on the R420 processor from ATI are hopelessly slower than devices on NVIDIA’s NV40 core. Once again, the problem is not only in the mutual suitability of Doom 3 and NVIDIA’s GPUs, but also in the obsolete and inefficient OpenGL driver from the ATI Catalyst suite.
There’s nothing extraordinary in the eye candy mode – the GeForce 6800 family cards remain on top.
Unreal Tournament 2004 is no serious trouble for top-end graphics cards of today. On the Torlan level, they all show the same performance, being limited by the speed of the CPU.
We don’t see any difference among the cards at the eye candy settings, too, save for 1600x1200 resolution where the ASUS turns in the best performance, both at its regular and overclocked frequencies. Well, an advantage of 5-6fps should be considered small when the absolute speeds are about 80fps. The GeForce 6800 is very slow, but due to an obvious reason – its slow memory drags it down.
The Metallurgy level is less complex and less CPU-dependent than Torlan, so we can clearly see how the participating cards differ among themselves. Since the cards are mostly engaged into geometry processing here, the GeForce 6800 Ultra has a better result than the ASUS AX800 XT.
The AX800 XT and the GeForce 6800 Ultra match each other in the eye candy mode; overclocking helps the ASUS to get a short distance ahead of the others.
The GeForce 6800 Ultra wins the low resolution of this game, but is overtaken by the ASUS AX800 XT in 1280x1024. Overclocking helps little to the latter in this race.
Well suited to perform full-screen anti-aliasing and anisotropic filtering, the ASUS card suddenly rushes ahead. The effect from overclocking is small again – almost lost against the absolute speeds that reach as high as 87fps.
The main combatants – ASUS AX800 XT and GeForce 6800 Ultra – have practically the same results on the Research level, although the scene is different here than the one recorded on the Pier level.
There’s nothing new or extraordinary in the eye candy mode. The R420 architecture, more efficient under such conditions, helps the AX800 XT to leave the GeForce 6800 Ultra far behind. Moreover, even the 12-pipelined RADEON X800 PRO shows a performance comparable to that of the top-end cards on NVIDIA’s chips. The 12-pipelined solution from NVIDIA evidently lacks fast memory.
Halo isn’t a new game anymore and, having been ported from Microsoft’s Xbox console, it can’t boast complex textures. But there’s an abundance of pixel shaders in here, so the small advantage of the ASUS AX800 XT is natural as the members of the RADEON X800 family have historically been better at executing pixel shaders than GeForce 6800 series processors. Overclocking is of little effect here, again.
This game is known for its thrifty handling of system resources, and top-end graphics cards often run it at 200-300 fps rates, like in the current test. Efficiently processing pixel shaders and working at a higher frequency, the ASUS AX800 XT outperforms the GeForce 6800 Ultra by about 20fps, but this is no serious defeat considering the absolute speeds of about 250-300fps.
When full-screen anti-aliasing and anisotropic filtering are enabled, the ASUS AX800 XT enjoys a bigger advantage over the GeForce 6800 Ultra. The speed bonus you get by overclocking the ASUS card is also bigger in the eye candy than in the pure speed mode.
It is the first time we use the full version of the recently released Counter-Strike: Source (see our article called Half the Way to the Half-Life 2: Counter Strike: Source Benchmarked). This online first-person shooter shares the same engine with Half-Life 2, and thus features beautiful hi-resolution textures as well as numerous pixel shaders. CS:Source is truly a next-generation game. To check its speed we used the same testbed, but replaced the old processor with an AMD Athlon 64 3200+. We’re going to test all graphics cards with the AGP interface with this CPU in the future.
The GeForce 6800 Ultra wins the first two resolutions of the Chateau level of CS:Source, but the ASUS AX800 XT catches up with it in 1600x1200. This level seems to be less saturated with pixel shaders, although there’re enough water surfaces here. Overclocking brings a small performance gain.
The ASUS card loses the lowest resolution to the GeForce 6800 Ultra in the eye candy mode, then overtakes it in 1280x1024 and increases the gap in 1600x1200. Overclocking the AX800 XT brings more dividends in this mode than in the pure speed one – we enjoyed a 10% speed boost in the highest resolution!
It’s different on the Italy level. The GeForce 6800 Ultra remains the leader in all the three resolutions, although the overclocked ASUS AX800 XT is getting close to it.
It’s similar to what we saw on the Chateau level at the eye candy settings: the ASUS AX800 XT loses the lowest resolution but then overtakes the GeForce 6800 Ultra. Overclocking helps in the chase, but not too much.
All the 16-pipelined graphics cards have the same speed at the pure speed settings of this upcoming game (see our article called A Next-Generation DirectX 9.0 Game Graphics Performance Preview for more details). They all hit the ceiling imposed by the central processor of the system.
The Under Two level is less dependent on the CPU and allows better evaluating the speeds of the graphics cards. Well, there’s anyway parity between the ASUS AX800 XT and the GeForce 6800 Ultra, save for 1024x768 resolution where the CPU rather than the graphics card seems to play the crucial part.
The beta version of this upcoming game (see our article called Yet Another DirectX 9 Game: Lost Oblivion in Chernobyl for details) uses the capabilities of DirectX 7 – pixel shaders do not live here, only fixed T&L functions. NVIDIA’s graphics processors have some optimizations for running applications of this kind, so the results presented in the diagrams are normal.
The ASUS card, however, overtakes its main rival, the GeForce 6800 Ultra, in the eye candy mode thanks to the excellent algorithms of using the available graphics memory bandwidth.
NVIDIA wins again on the Escape Run level, but this victory is less evident as in the previous case: the ASUS AX800 XT is only 8-27fps behind, which is a small gap considering the absolute speeds of about 200fps.
The speeds of the ASUS AX800 XT and the GeForce 6800 Ultra are similar when we turn on full-screen anti-aliasing and anisotropic filtering. The ASUS even wins at overclocked frequencies.
The ASUS AX800 XT deservedly wins this test as it is capable of processing complex pixel shaders this game abounds in, quickly. Beautiful visual effects are created with the help of those shaders, while the textures are less complex – the game was ported from a gaming console, too. This game is tested manually so the results are approximations (it’s almost impossible to repeat the same run with the main character).
This is a popular third-person arcade, not really rich in complex pixel shaders, and again we observe parity, although with a minor advantage on NVIDIA’s part. This advantage weighs only 10fps in the highest resolution when the absolute speeds are about 100fps. Overclocking the AX800 XT improves its performance by 4-6fps only.
All top-end graphics cards with 16 pipelines have the same speed in the pure speed mode of Max Payne 2. The reason is obvious – they all reach the performance ceiling set by the central processor of the system.
We can only see any difference between the top-end cards in 1280x1024 and higher resolution of the eye candy mode. This time the ASUS AX800 XT is slightly faster than the GeForce 6800 Ultra.
The last version of the popular flight simulator IL-2 Sturmovik is a very demanding game, requiring a high pixel shader performance and a high fill rate from the graphics card. The GeForce 6800 Ultra can boast a high fill rate, while the ASUS AX800 XT works faster with the shaders. NVIDIA’s top-end card is ahead of the ASUS in this test, but the gap is negligible. When the ASUS is overclocked, there’s even no gap at all.
Graphics cards with GPUs from ATI usually feel at their ease at eye candy settings (when full-screen anti-aliasing and anisotropic filtering are enabled) thanks to their quick operation with the graphics memory, but we’ve got quite another picture this time: all the members of the GeForce 6800, except the 12-pipelined model, run this game much faster than the ASUS AX800 XT. We are not really sure why this happens; probably ATI’s OpenGL driver is subject to blame.
Lock On doesn’t use complex pixel shaders, but it is anyway a resource-hungry game, producing rather complex special effects like multi-layer clouds. Besides that, the game is rather wayward as concerns its graphics quality settings – sometimes it’s not possible to set the same settings for different platforms and/or graphics cards. The GeForce 6800 Ultra and the overclocked ASUS AX800 XT share the top place here; the AX800 XT at its regular frequencies shares the next step of the podium with the GeForce 6800 GT.
The ASUS delivers the performance of the GeForce 6800 Ultra in the eye candy mode, and adds 10% more at overclocking.
The engine of this auto simulator makes an extensive use of complex pixel shaders to create various visual effects. So, the result can be expected – the ASUS AX800 XT deservedly takes its first place. Overclocking brings a considerable speed boost here as the speed of processing pixel shaders directly depends on the GPU clock rate.
Unlike in IL-2: Sturmovik, the eye candy mode of Colin McRae Rally 04 is without surprises: the higher load on the memory subsystem allows the ASUS AX800 XT to show its best. Overclocking is rewarding here, too.
There’s no difference between top-end graphics cards until 1600x1200. Well, the gap between the AX800 XT and the GeForce 6800 Ultra is a negligible 3 frames per second in the highest resolution, too.
The speeds start to differ since 1280x1024 resolution in the eye candy mode. The ASUS AX800 XT has an advantage over the GeForce 6800 Ultra here and increases it in 1600x1200.
Perimeter is most sensitive to the pixel shader performance, so the ASUS AX800 XT wins easily here. By the way, it seems to be the only graphics card that allows playing this real-time strategy comfortably.
Nothing exceptional happens as we turn on FSAA and AF. The ASUS X800 XT wins this test both at its regular and overclocked frequencies. However, there’s less playing comfort now because the minimal speed occasionally bottoms out to below 25fps. This game is benchmarked manually, so there can be small imprecision in the results.
High overdraw coefficient and complex scene geometry are the distinguishing features of this benchmark. It has few pixel shaders and hardly any complex shaders, so NVIDIA’s GPUs, particularly the GeForce 6800 Ultra, are faster here. The ASUS AX800 XT reaches the same level of performance through overclocking.
The combatants change their places in the eye candy mode: the ASUS gets the lead, thanks to the efficient operation with the memory subsystem. This time the advantage over the GeForce 6800 Ultra is hefty enough, amounting to 7 fps in high resolutions while the absolute speeds are 30-60fps.
The ASUS AX800 XT turns in a higher overall score than the GeForce 6800 Ultra in 3DMark03, but let us examine each of the tests separately.
The first test is simple: DirectX 7 only and no pixel shaders. No wonder the GeForce 6800 Ultra feels better than the AX800 XT here.
The ASUS strikes back in the eye candy mode, though. This card is better than the GeForce 6800 Ultra but the gap is diminishing in higher resolutions to become almost unnoticeable in 1600x1200. Overclocking is of tiny effect.
The second game test is NVIDIA’s territory, but the ASUS AX800 XT doesn’t give up and accelerates to the speed of the GeForce 6800 Ultra, even outperforming it at overclocking. That’s an excellent result, considering the “foreign turf” – the test uses rather simple version 1.4 pixel shaders, but the rendering algorithm is favorable to NVIDIA’s GPUs that can process shadows quickly thanks to their optimized Z buffer.
And here’s the proof that the NV40 architecture is better suited to the conditions of the second test than the R420 architecture: the GeForce 6800 Ultra outperforms the ASUS AX800 XT. The latter is ahead at the overclocked frequencies, though.
The results of the third game test are close to those of the second as both tests use the same rendering algorithms.
Things are different in the eye candy mode, however: the ASUS AX800 XT doesn’t lose at the start and surpasses its rival in high resolutions. Overclocking even makes the ASUS an unrivalled leader in this test.
The fourth game test requires no comments. It is stuffed with pixel shaders that demand complex mathematical calculations, so the result of the ASUS AX800 XT seems right.
The same goes for the eye candy mode with its full-screen anti-aliasing and anisotropic filtering.
Thus, the ASUS AX800 XT’s higher overall score in 3DMark03 is well earned as this card was close to the GeForce 6800 Ultra in the first three tests, and much faster in the fourth test which was the most difficult.
The ASUS AX800 XT/TVD is an example of product that is very close to being perfect: there is only one drawback – the cooling system! Anyway, we guess any enthusiast would enjoy having the tremendous speed and gorgeous accessories of this device. Of course, like all X800 series graphics cards, this one doesn’t support Shader Model 3.0, but this is unimportant as yet, since there’s just one game (Far Cry) that uses shaders of the third version. 3DMark05 also supports them (see our article called 3DMark05: The Future of Computer Games in Numbers for details), but it is not a game, but a benchmarking tool and should be regarded as such.
For you to see the overall performance level of the ASUS AX800 XT we’ve constructed two diagrams with all the test results put together.
As you see, the performance of the ASUS AX800 XT in the pure speed mode roughly corresponds to that of the GeForce 6800 Ultra, save for a few instances like Doom 3 where all NVIDIA’s GPUs have an initial advantage. Sometimes, thanks to its quick execution of pixel shaders, the ASUS leaves the GeForce 6800 Ultra far behind (see the results for Painkiller and Colin McRae Rally 04, for example).
In the eye candy mode, i.e. when 4x full-screen anti-aliasing and 16x anisotropic filtering are enabled, the ASUS AX800 XT has much more impressive results, outperforming the GeForce 6800 Ultra across the majority of applications, save for Doom 3 and IL-2 Sturmovik. It is especially clear in 1600x1200 resolution as the efficient methods of working with the graphics memory allow the ASUS card to feel more confident in the highest resolution than in 1280x1024.
The high price of the ASUS AX800 XT/TVD is another disadvantage we haven’t yet mentioned as, according to Pricewatch.com, you are invited to pay from $585 to over $700 for a sample of that card. Well, highest performance and gorgeous accessories do not come for free – this is a top-end graphics card after all. We guess true enthusiasts won’t be taken aback by the price and we recommend this graphics card to them. However, we’d like to express our concerns to the manufacturer about the cooling system – the current cooling of the card is not always capable of keeping such a rapid device cold.
Highs:
Lows: