by Alexey Stepin , Anton Shilov
05/05/2005 | 02:10 PM
Not so long ago we would consider the 256MB of onboard graphics memory to be excessive luxury: additional chips would only make the product more expensive and didn’t ensure any performance growth, because the games of those days simply didn’t have that much graphics data. Some graphics card manufacturers took advantage of the users’ affection with big numbers and equipped not very high-performance graphics accelerators with 256MB of slow memory. As a result, the customer got attracted to the beautiful number, overpaid for the memory he would actually not use and never got the anticipated performance gain.
A lot of things have changed since then. Nowadays, 256MB of onboard graphics memory has already become a common thing: contemporary games boast much more detailed graphics. Besides, most gamers prefer to play with full-screen anti-aliasing (FSAA) and anisotropic filtering enabled, which also increases the requirements imposed upon the graphics memory subsystem. Almost all graphics cards from the performance-mainstream market segment are currently equipped with 256MB of GDDR3. As for the larger capacity of the graphics memory, solutions with over 256MB onboard have so far been only seen in the professional solutions designed specifically for 3D modeling and CAD/CAM tasks.
The first sign that the new consumer type of graphics accelerators equipped with 512MB of graphics memory is about to hit the streets was the Doom III 3D shooter. The game developers claimed that a graphics adapter with so much memory was required to achieve acceptable gaming performance in Ultra high Quality mode. Our tests however, have shows that this statement was by far exaggerated and the cards with 256MB of memory worked perfectly well in this mode running about as fast as in High Quality mode. But the green light has already been turned on and it was just a matter of time how soon the graphics cards equipped with more onboard memory would appear in the market. Especially, since the software developers didn’t sit still with their hands in pockets and kept working on new projects with better quality and more complex graphics than in such popular shooter games as FarCry, Half-Life 2 and Doom III.
512MB of graphics memory may actually become a must for most upcoming games. But do we really need it now? Will it ensure any significant performance improvement in the contemporary games? Will it affect the price point of the cards greatly? Today we will do our best to answer all these questions for you: our today’s article will be devoted to a new ATI graphics accelerator with 512MB of graphics memory.
On
But let’s get to benchmark results later in this article and in the meanwhile we will try to answer a question that is very likely to be of interest to all of you: why RADEON X800 XL? Why did ATI double the amount of onboard graphics memory by this particular solution and not by the faster RADEON X850 XT/XT Platinum Edition? At first glance this seems to be a pretty strange decision to make: extra graphics memory will hardly ensure any significant performance improvement in contemporary games, while in the upcoming games the performance potential of RADEON X800 XL may turn out insufficient, so the extra graphics memory will not help.
The answer to this question is quite simple: the recommended price for the ATI RADEON X800 XL 512MB is $449 in the
So, ATI’s decision looks quite smart and justified from this point of view: it is much easier to find customers willing to buy a $450 card. By the way, NVIDIA sticks to a completely different strategy: they are going to supply the partners with a 512MB GeForce 6800 Ultra working at higher clock frequencies. It is still hard to say how much it is going to cost, but it will undoubtedly get over the $599 bar.
In fact, the potential customers are very excited about the 512MB graphics accelerator models and retail stores also express their vital interest in them already. We should keep in mind that there will always be a certain percentage of customers willing to buy “big numbers”, i.e. willing to get a graphics card with 512MB of onboard memory at any rate. So, it means the demand for solutions like that is there, and they will definitely find their niche.
Now let’s meet one of the cards like that: ATI RADEON X800 XL 512MB.
The 512MB version of ATI RADEON X800 XL doesn’t differ that much from the 256MB version, as you may have expected it to. Take a look yourselves:
ATI RADEON X800 XL 512MB | ATI RADEON X800 XL 256MB |
 |  |
 |  |
All the major design differences have to do with the location of additional memory chips. Since the 512MB version is equipped with 16 GDDR3 chips instead of 8, this resulted into a considerably bigger amount of various tiny electronic components on the PCB around the memory chips. Also the number of stripes grew up significantly, too.
Here I have to point out a few special things. RADEON X800 XL is designed on 8- and 10-layer PCBs. Some manufacturers are preparing simpler 6-layer PCBs for RADEON X800 XL and they are currently going through some testing. Since there is only one manufacturer using 10-layer PCB design, according to ATI, then we dare suppose that the RADEON X800 XL 512MB is based on an 8-layer PCB. Some stripes leading to the memory chips are laid out inside the PCB, that is why it is impossible to use 6-layer PCB for the 512MB RADEON X800 XL solution because of the signal termination problems.
Other than that the PCB layout of two RADEON X800 XL versions is very similar. Even the location of many individual components and the PCB size are identical. RADEON X800 XL 512MB is just a little bit longer because of the technological lug at the back of the PCB, which doesn’t carry any electronic components or stripes.
The voltage regulators of the cards are also almost the same except for a couple of chips for additional power supply for the 512MB version. In fact, we were surprised to see it there, as ATI claimed that the power consumption of the new 512MB RADEON X800 XL would be only 5-7% higher than that of the regular RADEON X800 XL graphics card. We carried out a small experiment and disconnected the additional power supply from the card, so that it had to be happy with the power it could get through the PCI Express connector only. The graphics card didn’t react to it in any way and continued working just fine at the same clock frequency without the external power supply. So, I would assume that ATI Technologies provided this additional power supply line to ensure higher stability of the device, and not because they were concerned with its power consumption level.
Since this graphics card belongs to the top price category, the presence of RageTheater chip implementing VIVO support and two DVI connectors are also quite justified. One of these connectors is implemented by the transmitter built into the R430 core, while the other one – by the stand alone Silicon Image SiI1162 chip. You can actually see this chip on all contemporary ATI RADEON graphics cards featuring two DVI-I outs.
Now that the cooling system has been successfully removed we can see the VPU:
The marking suggests that this piece was manufactured during week 48 of last year. The chip is equipped with a metal frame typical of all top ATI chips, which protects the fragile edges of the die from physical damage. We also removed the layer of thermal paste from one of the memory chips:
This is a classical Samsung K4J55323QF chip with 2.0ns access time. It is very often used on high-end and mainstream graphics cards today. The card is equipped with the total of 16 chips 256Mbit each, which makes 512MB. Samsung and Infineon already have 512Mbit GDDR3 chips at their disposal, so with these ones they could do with only 8 chips per card thus simplifying the PCB layout significantly. However, these chips are still pretty expensive and are not that widely spread. Therefore ATI decided to go for better value 256Mbit chips that were widely available when they designed their RADEON X800 XL 512MB.
The cooling system of this graphics card deserves a separate mention in our review. It is very much different from the reference cooler you can see on the regular RADEON X800 XL based graphics cards with 256MB of onboard memory. We have already seen something like that by PowerColor X800 XL graphics card (for details see our article called PowerColor X850 XT and PowerColor X800XL Graphics Cards Review). However, the similarity you might see is mostly first glance observation, although both coolers use a pretty noisy ADDA AD4512HB-E03 fan powered by 0.35A current.
The new cooler features a solid copper footing providing tight contact with the VPU surface, while the PowerColor X800 XL features an aluminum heatsink footing.
If you take a closer look at the card, you will also see that there ate two flat heatpipes transferring the heat from the footing above the VPU and distributing it evenly all over the heatsink composed of thin aluminum ribs. The memory located on the front side of the PCB is cooled down with the same heatsink as the VPU, as the footing features special bumps covering the memory chips, too. On the back of the PCB there is an aluminum plate dissipating the heat from memory chips located there. It is not a very efficient cooling solution, however, GDDR3 memory working at about 500MHz (1000MHz) frequency generates not that much heat, so the existing cooling appears more than enough.
The cooler is fastened to the PCB with four screws, three of which also hold in place the above mentioned aluminum plate. Two more screws press a special bracket to the bottom side of the PCB. This bracket is intended to prevent the PCB from curving and also serves as an additional cooler retention. The thermal interface is provided by two types of thermal paste: thick gray paste for the VPU and white silicon paste for the memory chips.
The entire construction leaves very good impression of a well-done thoughtful solution, although the fan still aroused a few concerns: according to our PowerColor X800 XL tests, it becomes very noisy at its maximum rotation speed. Everything depends on the rotation speed regulation circuit, and if the fan will be working at lower rotation speed most of the time, the card can be really quiet. In the next chapter of our review we will specifically dwell on this matter.
Despite our concerns, RADEON X800 XL 512MB demonstrated pretty good noise parameters. Although the card works very noisy during the system boot-up as the fan rotates at its maximum speed, it slows down significantly in just a few seconds and the noise level drops down to acceptable level. Moreover, you can barely hear the card at all against the background of other system noises. During our test session the fan never sped up, which is actually not surprising at all, as ATI R430 is manufactured with 0.11micron production technology and generates very little heat at work. Moreover, the copper footing and heatpipes of the cooling system dissipate the heat very efficiently.
RADEON X800 XL based graphics cards do not overclock that well, as a rule, as ATI didn’t use low-k technology during R430 graphics processor manufacturing. It didn’t make any sense to hope that ATI RADEON X800 XL 512MB would work wonders. And no wonder happened, although I still wouldn’t call the overclocking results we managed to achieve a failure. The maximum frequency the graphics card worked at stably and reliably reached 460MHz for the VPU. The memory overclocked to 550 (1100)MHz, which is also not bad at all, especially keeping in mind that the PCB design has become considerably more complex and that the memory chips onboard feature 2.0ns access time.
RADEON X800 XL 512MB demonstrated worthy 2D image quality, just like all other ATI graphics adapters ever tested in our lab. The image was crisp and clear in all resolutions up to 1600x1200x85Hz and 1800x1440x75Hz. Unfortunately, we didn’t manage to test the 2048x1536 resolution, because our lab monitor, Dell P1130, provides only 60Hz refresh rate in this case, which is unacceptable for comfortable perception of the image.
Since large amounts of graphics memory can be actually used only in high resolutions with enabled full-screen anti-aliasing and anisotropic filtering, we decided to introduce a new test mode, which would load the memory subsystem most heavily. The thing is that in low resolutions with disabled FSAA the size of the graphics memory of the card doesn’t have any influence on the performance of the accelerator. If you would like to play games in 1024x768 or 1280x1024 without FSAA, you should be OK even with 128MB of graphics memory. This statement is proven by the benchmark results we obtained for GeForce 6600GT, which is usually faster than ATI RADEON 9800 XT in this case, although the latter is equipped with 256MB of onboard graphics memory and features a 256bit bus. However, once FSAA is enabled, the situation changes completely: RADEON 9800 XT is very often ahead of GeForce 6600GT in this case due to faster bus and higher graphics memory capacity.
Contemporary high-end graphics accelerators are so fast that even FSAA 4x will hardly reveal the benefits of the 512MB of graphics memory. That is why besides the pure performance tests and FSAA 4x + AF 16x, we also introduced the test mode with the maximum FSAA level: 6x for ATI graphics cards and 8xS for NVIDIA graphics cards. We believe that the tests in this mode will allow us to find those games where doubled onboard graphics memory of ATI RADEON X800 XL 512MB will be truly beneficial. We assume that with disabled FSAA the performance gain will be minimal compared with the results shown by RADEON X800 XL 256MB, or will simply be null. Our benchmarks will help prove or deny this statement very soon.
For our tests we used our standard PCI Express platform with the following configuration:
We used the following software:
For a more illustrative comparison and better performance analysis we used the following graphics cards alongside with our today’s main hero:
According to our standard lab methodology the CATALYST A.I. optimizations were enabled as Standard, Mipmap Detail Level option was set to Quality. The optimizations in the NVIDIA ForceWare drivers were also activated, except Anisotropic mip filter optimization. Image settings engine was set to Quality, too. This way, ATi and NVIDIA graphics cards were running at equal conditions.
As for the game settings, we set the image quality to the maximum possible level, which was the same for all graphics adapters. If the game allowed adjusting FSAA and anisotropic filtering settings, we used this opportunity, otherwise the required settings were forced through drivers. Since most games supporting FSAA level adjustment do not allow enabling 8xS for NVIDIA graphics cards, we would force it from the ForceWare driver.
During our tests we used the following games and applications:
First Person 3D Shooters:
Simulators:
Strategies:
Semi-synthetic Benchmarks:
Synthetic Benchmarks:
Please note that we excluded games that could not enabled full-scene antialiasing (FSAA) properly due to the fact that high-end graphics cards with 512MB of memory onboard are currently meant to run games with FSAA enabled.
If you are well versed in modern 3D graphics, you should already know that larger amounts of graphics memory make higher resolutions and higher levels of full-screen antialiasing possible in games. This should theoretically improve the image quality. We decided to check if the image quality was really improved with the maximum possible FSAA levels – 6x and 8xS – on ATI’s RADEON and NVIDIA’s GeForce6 cards, respectively. So we took a number of screenshots and compared them assiduously. The games were the popular shooters Doom 3, Half-Life 2 and Far Cry as well as the strategy game Warhammer 40.000: Dawn of War and the rally simulator Colin McRae Rally 05. The screenshots were all made in 1600x1200 with 16x anisotropic filtering enabled.
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Doom III | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA |
|
|
FSAA 4x |
|
|
FSAA 6x/8xs |
|
|
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Doom III | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
Doom 3 is a rather dark game and does not generally show the advantages of full-screen antialiasing well. Yet we found one level where one could feel the difference. The antialiasing algorithms from NVIDIA work obviously better in Doom 3 than ATI’s ones. The slanting edges of polygons look less jagged on the GeForce 6800 Ultra than on the RADEON X800 XL 512MB. As for the difference between 4x and 6x/8xS antialiasing modes, it’s hard to see any, but the performance degenerates dramatically.
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Half-Life 2 | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Half-Life 2 | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
Half-Life 2 looks obviously different without FSAA and with 4x FSAA since the game’s scenes are generally brighter than Doom 3’s dark corridors of the UAC station. We think the FSAA 6x mode available with modern RADEONs gives a better antialiasing quality than NVIDIA’s FSAA 4x, but it’s not so easy to see that, especially during the play. The FSAA 8xS mode brings you the best quality, just like in the previous case, but it is hardly playable due to the very low frame rates. You can only play comfortably in 1024x768 with this mode enabled. But here, it would be better to use a higher resolution and a less resource-consuming FSAA mode – you’ll have a higher image quality in this case.
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Far Cry | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Far Cry | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
It’s hard to tell between 4x and 6x/8xS FSAA modes in Far Cry, but it’s clear that enabled antialiasing improves the image considerably.
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, CMR Rally 05 | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
Colin McRae Rally 05 also benefits from using full-screen antialiasing. It’s again difficult to catch the difference between ATI’s and NVIDIA’s antialiasing, but we think NVIDIA’s algorithms work better on slanting edges, while there’s some jaggedness visible in the 4x FSAA mode on the RADEON X800. Like in the rest of the cases, it is even harder to see the difference during the actual play. The FSAA 8xS mode brings you the best quality, but also a terrible performance hit: high resolutions cease to be comfortably playable.
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Warhammer 40K | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
FSAA Quality Comparison: ATI RADEON vs. NVIDIA GeForce, Warhammer 40K | ||
| ATI RADEON X800 | NVIDIA GeForce 6800 |
No FSAA | ||
FSAA 4x | ||
FSAA 6x/8xs | ||
It’s possible to improve the image quality in modern strategy games using full-screen antialiasing. In games of this genre you usually see numerous buildings and units with complex geometry on the screen, and antialiasing comes in most helpful. It’s harder to see the difference between 4x FSAA and 6x/8xS FSAA since the objects are relatively small, but the difference between disabled antialiasing and 4x antialiasing is very conspicuous. Note that in case of Warhammer 40.000 game the quality of NVIDIA GeForce 6800-series antialiasing is much worse compared to that of ATI RADEON X800 hardware.
So, it is really possible to improve the image quality by enabling full-screen antialiasing, but extreme modes like 6x and 8xS can hardly be recommended for use since they bring just a minor quality improvement over 4x FSAA but at a tremendous performance hit.
Such extreme antialiasing modes can hardly be playable in resolutions above 1024x768, but if this is the optimal resolution of your monitor (e.g. if you own a 15” TFT panel), you can try to improve the quality by turning on 6x/8xS FSAA. The latter mode ensures the best image quality, but is resource-consuming and only available to owners of NVIDIA GeForce 6x00 graphics cards. The 6x FSAA mode, available with all modern RADEONs, doesn’t improve the quality much over 4x FSAA, the latter less affecting the speed of the card.
The next section of our review is all about the performance of graphics cards with extreme full-screen antialiasing modes enabled.
As might have been expected, the 512 megabytes of graphics memory don’t show up in the “pure speed” mode, even though we tested the new RADEON X800 XL at the “Ultra High Quality” settings of the game.
We don’t see any performance benefits from more graphics memory in the “eye candy” mode (i.e. with enabled full-screen antialiasing 4x and anisotropic filtering 16x), either. The speeds of the two RADEONs X800 XL coincide to a tenth of fps.
Surprisingly there’s no performance gain even in the highest FSAA modes, but the speed has dropped below playable in 1024x768 already.
So, id Software’s claim that a graphics card with 512MB of graphics memory is necessary to play Doom 3 with the maximum image quality settings is not confirmed. The GeForce 6800 Ultra/GT are fast enough for the FSAA 8xS mode to be playable in 1024x768, but we can’t say the same about the ATI RADEON X800/X850 – even the topmost models cannot overcome the 40fps barrier using 6x antialiasing.
We didn’t observe any speed bonuses on the d3dm4 map in the “pure speed” mode.
Our turning on 4x FSAA and aniso-filtering didn’t change anything. The 512MB version of the RADEON X800 XL is just a single frame per second faster than the ordinary one in 1600x1200, but this lies in the measurement error range.
There are no bots on the d3dm4 map, and the speed of our demo is generally faster even with extreme antialiasing modes in use. The game is quite playable in 1024x768, but none of the cards can produce more than 40fps in higher resolutions.
The graphics cards all hit at the speed ceiling imposed by the central processor of the system.
The same thing happens in the “4x FSAA plus 16x AF” mode.
The participating cards from NVIDIA slow down suddenly in the extreme quality mode due to the resource-consuming 8xS FSAA algorithm, but maintain playability even in 1280x1024. The RADEON X800 XL 512MB has a certain advantage in speed, but it’s no more than 2.6fps, absolute frame rates being about 65-70fps.
The results on another level, Metallurgy, also suggest that Unreal Tournament 2004 is an easy task for top-end graphics cards of today. The GeForce 6800 Ultra/GT cards are ahead of the RADEON X850/800 models in high resolutions, though.
The RADEON X850 XT Platinum Edition wins the high resolutions of the “eye candy” mode, while the two versions of the RADEON X800 XL have the same results in all the resolutions. That is, 512 megabytes of graphics memory don’t give any bonuses to the new card in this game, either.
We switch to 6x/8xS FSAA mode to see the RADEON X800 XL 512MB being just 2fps ahead of its 256MB version. ATI’s cards managed to make even 1600x1200 resolution comfortably playable while the speed of the GeForce 6800 Ultra/GT dropped below the critical mark of 40fps.
The Chronicles of Riddick is a relatively fresh game. It is quite resource-consuming, while its engine generally works better on NVIDIA’s GeForce6 series cards. The RADEON X800/X850 cards can’t boast high performance here, being barely capable of providing for comfortable play in 1600x1200. There’s no difference between the two versions of the RADEON X800 XL: their results coincide to a fraction of fps.
There’s no difference in the 4x FSAA + 16x aniso-filtering mode, either. Then, the game is only playable on the RADEON X800 XL in 1024x768 at such settings.
We didn’t find any performance gains in the maximum quality mode. The game doesn’t seem to operate with as many complex textures as to demand for more than 256 megabytes of graphics memory in typical resolutions.
The scene recorded on the Pier map contains a flight on a hang-glider high above the water surface. The area that fits in the screen is huge and the performance in the “pure speed” mode depends on the speed of the system at large. That’s why all the tested graphics cards have almost identical results, although there’s some advantage on the part of the RADEON X850 XT.
Both RADEON X800 XL models are on the same level with the GeForce 6800 Ultra in the “eye candy” mode. As you see, 512 megabytes of graphics memory can’t give you any advantage in Far Cry.
The RADEON X800 XL 512MB doesn’t seem to profit much from its large memory even in extreme antialiasing modes. By the way, ATI’s RADEON X850/X800 cards are fast enough even here, while the GeForce 6800 Ultra/GT don’t provide for comfortable play in resolutions above 1024x768.
The demo on the Research map is the opposite of the one made on the Pier map. This one depicts a walk through an underground cave lit with numerous light sources. The RADEON X800 XL cards are both a little slower than the GeForce 6800 GT here.
In the “eye candy” mode (4x FSAA + 16x AF) the RADEON X800 XL are successfully competing with the GeForce 6800 GT, there still being no difference between the 512MB and 256MB versions.
There’s no difference in the 6x FSAA + 16x AF mode, either, but the RADEONs again prove their ability to ensure playable frame rates even in this, hardest mode.
The results of all the cards are identical here as they are all limited by the speed of the system’s central processor.
The ATI RADEON X850/X800 cards shoot ahead in higher resolutions thanks to their advanced memory controller, technologies for an efficient use of the available memory bandwidth and the traditionally “fast” FSAA method. Yet we still see no profit from the new RADEON X800 XL’s 512MB of onboard memory.
When the highest quality modes are activated, the NVIDIA GeForce 6800 cards fall far behind their competitors that are traditionally better at extreme loads. Well, there are no downright low results here, while the extra 256MB of memory on board the RADEON X800 XL 512MB remain unused.
The d1_canals_12 map abounds in pixel shaders, so graphics cards based on the ATI R480/430 processor are in the lead here.
It’s in the “eye candy” mode that we for the first time see a really big difference between the two versions of the RADEON X800 XL: the 512MB version is 5% faster than the older one in 1280x1024 and almost 20% faster in 1600x1200! It is no surprise the RADEON X800 XL 512MB enjoys a speed bonus from having the extra memory in Half-Life 2 rather than elsewhere since this game abounds in high-resolution textures other games, for example Far Cry, don’t have. Added the enabled full-screen antialiasing and anisotropic filtering, each megabyte of graphics memory becomes valuable, and the 512 megabytes of the new RADEON come in most handy here.
The RADEON X800 XL 512MB enjoys supremacy over the ordinary RADEON X800 XL in the “extreme quality” mode, too. Here, the gap varies from 15% in 1280x1024 to 20% in 1600x1200. Note also that the performance is high even in 1600x1200 – it’s comfortable to play at 70fps. The GeForce 6800 Ultra and GT are good enough in the first two resolutions, but slow down below the acceptable limit in 1600x1200.
The scene recorded on the d3_c17_02 map depends much on the speed of the system’s central processor, but ATI’s RADEONs are undoubtedly leaders here, even in the “pure speed” mode. The 512MB of memory don’t give any special bonuses, but that’s expectable in modes without full-screen antialiasing.
This map isn’t as large as d1_canals_12 and is simpler in terms of graphics. The RADEON X800 XL 512MB is only 5% faster than the 256MB version, and only in 1600x1200 resolution.
The advantage goes up to 18-20% with 6x antialiasing, but again in the highest resolution only. Thus, the RADEON X800 XL 512MB even beats the RADEON X850 XT Platinum Edition. The GeForce 6800 GT and Ultra are far behind the others in this test: the enabled 8xS FSAA mode affects the performance negatively, but you can still play 1024x768 resolution.
F.E.A.R. is a completely new game now under development at Sierra Entertainment. We use its beta version in our tests that is limited to the multiplayer mode. Yet even the beta version of the game is highly impressive. It raises the graphics quality bar at a whole new level. This shooter is definitely more beautiful than Far Cry or Half-Life 2. The game has an integrated benchmark which is, however, not very accurate. The benchmarked scene is not always reproduced correctly: sometimes the soldiers refuse to enter the mined room thus strongly affecting the final result. We also warn you against taking the numbers below too seriously: practice suggests that game developers often optimize the final versions of games for the available hardware, and the GPU manufacturers are optimizing the drivers, too.
A number of sophisticated special effects, including realistic water, soft shadows and reflections, make this game – based on a new engine from Monolith – a highly demanding application. We could only squeeze 48fps out of the GeForce 6800 Ultra at the maximum graphics quality settings. The cards of the RADEON X800 and X850 families have worse results probably because the game makes use of the special features of the NV4x architecture, particularly UltraShadow II technology.
With 4x FSAA and aniso-filtering enabled, the RADEON X800 XL 512MB is a little ahead of the RADEON X800 XL 256MB. That’s natural, considering the quality of the image. The gap is about 10%. Given the highest hardware requirements of F.E.A.R., there’s no talking about playability of the “eye candy” mode. You can get 20-25fps at best with a modern graphics card, and not a single frame more. Are we to wait for the next generation of graphics cards to play this game?
The new RADEON X800 XL again shows how important the amount of graphics memory is for upcoming games. Despite the relatively low frequencies, it delivers the performance of the RADEON X850 XT Platinum Edition and is even faster than this top-end model in 1024x768, even though by 1fps only. Like in the 4x FSAA mode, none of the graphics cards was capable of yielding 20fps even, so the full-screen antialiasing feature is practically useless in F.E.A.R., at least with the current version engine and on the currently available hardware.
This flight simulator defaults to using OpenGL, thus giving a certain advantage to NVIDIA GeForce 6800 GT/Ultra cards. We have no profit from the 512MB of graphics memory in the “pure speed” mode here, just like in the rest of the games we used for this review.
The extra memory brings no speed bonuses in the “eye candy” mode, either, despite the enabled full-screen antialiasing and anisotropic filtering. The GeForce 6800 GT and 6800 Ultra are leaders still.
We faced a paradox in the “extreme quality” mode: the RADEON X800 XL 512MB was worse than the ordinary RADEON X800 XL! The gap was no more than 5%, but it was stable and repeatable – we reran the test and had the same results. It’s hard to explain this outcome since both cards have identical clock rates and only differ in the amount of graphics memory onboard. Anyway, it doesn’t matter much since the performance of both cards is too low in this mode.
Another flight simulator – Lock On – doesn’t seem to get anything from the 512 megabytes of graphics memory on board the new RADEON X800 XL.
The same is true for the 4x FSAA mode. Both RADEON X800 XL cards deliver the same performance and are slower than the NVIDIA GeForce6 solutions and the topmost RADEONs.
It is only in the “extreme quality” mode that there’s difference between the two versions of the RADEON X800 XL, but the speed bonus is no more than 5-7%.
We don’t publish the FlatOut results in the “pure speed” mode since the game has an inbuilt speed limiter set to 100fps.
Both versions of the RADEON X800 XL roughly match the performance of the GeForce 6800 GT in this test. The extra memory of the new RADEON X800 XL doesn’t show up at all.
The same situation is in the “extreme quality” mode. The GeForce6 team, like in other cases, fall behind ATI’s RADEON X800/X850 due to their more difficult 8xS way of doing full-screen antialiasing. 1600x1200 resolution is downright unplayable on the GeForce 6800 GT/Ultra – the frame rate sinks below 30fps.
This auto sim makes an intensive use of pixel shaders, and ATI’s RADEON X800 cards run it faster than NVIDIA’s GeForce6 ones do. Both RADEON X800 XL models are far ahead of the GeForce 6800 Ultra, although the gap becomes somewhat narrower in 1600x1200.
The new RADEON X800 XL doesn’t seem to have any speed bonus from its large memory when 4x FSAA and 16x aniso-filtering are enabled.
Here, we have the same thing in resolutions above 1024x768 as in Il-2: Sturmovik. That is, the RADEON X800 XL 512MB is behind the 256MB version of itself, although the gap is only 2fps at most.
Both versions of the RADEON X800 XL perform identically in this strategy game, being slower than the GeForce 6800 Ultra and RADEON X850 XT.
We can’t see any advantage from having 512 megabytes of graphics memory in the 4x FSAA + 16x AF mode.
The 512MB model is slower than the 256MB one by 5-8% in the “extreme quality” mode. Despite the laborious 8xS antialiasing enabled, the NVIDIA GeForce 6800 cards perform well here, although 1280x1024 resolution can hardly be considered playable.
There are no complex textures in Perimeter, but complex shaders do live here, so there’s no profit from the 512MB of graphics memory and both versions of the RADEON X800 XL are far behind the top RADEON models as well as behind the NVIDIA GeForce 6800 GT and Ultra.
It’s only top-end devices that can run this game normally in the “eye candy” mode, and only up to 1280x1024 resolution. The performance is too low for comfortable play in higher resolutions.
The performance goes down ever more with higher FSAA levels, and even the fastest graphics cards cannot ensure a playable frame rate in resolutions above 1024x768. The RADEON X800 XL 512MB again repeats the results of the junior version to a fraction of fps.
Aquamark3 doesn’t have complex textures but loads the GPU with complex geometry instead, so large amounts of graphics memory mean nothing in this benchmark.
We can’t see any positive impact of the larger memory amount of the new RADEON X800 XL even when we enable 4x full-screen antialiasing and 16x anisotropic filtering. It is slower than the RADEON X850 XT and XT Platinum Edition, like in most other tests.
The same is true for the “extreme” mode with full-screen antialiasing of the maximum level.
3DMark03 gave identical overall scores to both versions of the RADEON X800 XL. This is partially due to the default use of 1024x768 resolution in all benchmarks from Futuremark. Let’s see what we have in each test independently.
The first test indicates that the new RADEON X800 XL has no advantages over the older one in high resolutions, and they are both slower than the RADEON X850 models as well as both top-end devices from NVIDIA.
The same is true for the 4x FSAA mode since the first test doesn’t have complex textures.
A certain performance gain can be observed when the maximum antialiasing level is in use, but the gain is never more than 2-3%.
The second test of the 3DMark03 suite doesn’t have high-resolution textures, either. That’s why, like in the first test, the RADEON X800 XL 512MB behaves exactly like the ordinary RADEON X800 XL does.
It’s the same story repeated in the 4x FSAA + 16x AF mode: the difference between the two RADEON X800 XL models lies in the measurement error range, again.
Our tuning on extreme antialiasing modes leads to the NVIDIA GeForce 6 cards falling behind the ATI RADEON X800/X850 models in high resolutions due to the high load on the memory subsystem.
The third test is intrinsically the same as the second, so all the above-said things apply here as well.
There’s no gain from the larger memory amount even with full-screen antialiasing enabled.
Well, there’s no difference between the two RADEON X800 XL models even at the maximum-level antialiasing.
The fourth test contains complex pixel shaders but doesn’t operate with large amounts of graphical data, so there’s again no difference between the 512MB and 256MB versions of the RADEON X800 XL.
The same is true for the “eye candy” mode. The RADEON X850 XT/XT Platinum Edition, with high core frequencies, show their best in the fourth test.
Even with enabled 6x FSAA, the RADEON X800 XL 512MB has no advantage whatsoever – it performs exactly like the model with 256MB of memory. Thus, the separate tests agree with the overall scores: 512 megabytes of graphics memory don’t give any bonus to a graphics card in 3DMark03.
Although the new 3DMark is much more complex than the older version, it doesn’t seem to appreciate the availability of 512MB of graphics memory, either. Is it true for all the tests and resolutions? Let’s check it up.
The first scene of the suite is a typical 3D shooter. It’s difficult in terms of graphics quality, but the extra memory on board the new RADEON X800 XL gives it no speed bonuses.
Our turning on FSAA plus aniso-filtering doesn’t change anything. The RADEON X800 XL 512MB is still slower than the top-end RADEONs with only 256 megabytes of memory.
The performance gain from the additional memory is at last visible at 6x FSAA, but it amounts to 5% at best. Thus, the RADEON X800 XL 512MB still cannot challenge the RADEON X850 XT Platinum Edition.
The distinctive features of the second test are the dynamically generated vegetation and numerous shadows. The texture load is high here, but we are not surprised anymore to see the two RADEON X800 XL models have identical results.
Both versions of the RADEON X800 XL are far behind the RADEON X850 XT/XT Platinum Edition in the “eye candy” mode (4x FSAA + 16x AF).
Well, there’s no visible difference even with the maximum possible level of antialiasing on. On second thought, the gap of 0.2fps may be considered big enough because the absolute frame rates are no more than 10-13fps here.
The gist of the third test is in its pixel shaders that are extremely complex both math1ematically and texturally, and we might have expected some gain from more memory here, but the results refute this supposition.
The situation remains the same in the “eye candy” mode (4x FSAA and 16x AF): the RADEON X800 XL 512MB is still behind the RADEON X850 XT/XT Platinum Edition.
Even the use of the combination of 6x FSAA, anisotropic filtering and high resolutions doesn’t bring anything new to us. It’s probably the GPU speed rather than the memory amount that plays the crucial role in this test. Thus, the lack of difference in performance of the 512MB and 256MB versions of the RADEON X800 XL is confirmed in 3DMark05, too.
So, has our testing of the ATI RADEON X800 XL 512MB answered all the questions? One thing is certain: 512 megabytes of graphics memory on board a graphics card doesn’t bring any significant performance advantages in modern games, with rare exceptions. In our tests such exceptions were Half-Life 2 that operated with many high-res textures and F.E.A.R. Yet even in Half-Life 2 the performance gain would only show up in highest resolutions with enabled full-screen antialiasing and it never exceeded 20%. In all other cases the new device didn’t have any advantages over the ordinary RADEON X800 XL in typical resolutions.
A greater gain may possibly be got in resolutions above 1600x1200, but the bulk of today’s mass-user monitors do not support them. The native resolution of 17” and 19” TFT monitors is 1280x1024, while widespread CRT models rarely support resolutions higher than 1600x1200 at a comfortable refresh rate.
Besides that, the performance of the RADEON X800 XL may be too low to ensure comfortable play in 2048x1536 with 4x-6x full-screen antialiasing. At least we can guess it by the results of the tests in 1600x1200 and are going to check this highest resolution in practice soon.
That said, the market perspectives of the RADEON X800 XL 512MB seem obscure. Priced the recommended $449, it doesn’t give the user any tangible advantages in comparison with the ordinary RADEON X800 XL but it is noticeably costlier than the RADEON X850 PRO which works at higher clock rates but has only 12 pixel pipelines against the RADEON X800 XL’s 16 pipelines. These graphics cards should perform similarly, but the RADEON X800 XL is going to be somewhat better in pixel shader-heavy games and somewhat worse when the GPU clock rate is important, like in scenes with complex geometry.
Does the RADEON X800 XL 512MB suit people who want to play next-generation games? It depends on what games belong here. This year is to bring us such titles as Half-Life 2: Lost Coast, Quake 4, F.E.A.R. and others that use engines oriented towards the current-day hardware. As the developers employ ever more complex textures, normal maps and environment maps, a 512MB local buffer may become needed in some cases. As for games of a more distant future, like those scheduled for 2006 and later, we expect them to require a lot of graphics memory. But when they do arrive, the performance of the RADEON X800 XL processor proper may prove too low, and this performance rather than the memory amount may become the bottlenecking factor. Yet it’s too early to make any predictions. Such games don’t exist, while nearly all modern games are quite satisfied with 256MB of graphics memory if you use typical resolutions.
The new product from ATI may be interesting for PC enthusiasts who just want to have a graphics card with 512MB of memory at any cost, and for owners of big monitors that support resolutions above 1600x1200. But if you just want to enjoy the current day, it would make sense to consider cheaper devices with 256MB of memory.
However, this situation may change because the lifecycle of the RADEON X800 XL 512MB is expected to be at least one year since the date of its announcement. Its price may drop considerably in the meantime as ATI Technologies has resources for driving it down.
