by Alexey Stepin
02/15/2004 | 05:00 PM
Just a year ago the mainstream class of graphics processors from ATI Technologies was represented by two R300-based products: RADEON 9500 and 9500 PRO. The latter VPU was a real success. With its eight pixel and four vertex pipelines, the chip was only limited by the memory bus width, but anyway, it performed excellently in most games and applications. However, this situation couldn’t last for long: it was not profitable to use one VPU for all product families. The company drew a thick border line between them by releasing the top-end R350 and the mainstream RV350.
<%BANNER[article]%>The RV350 chip (officially named RADEON 9600) was a pretty progressive one: it was manufactured with 0.13micron technological process and worked at 400MHz frequency, but the developer left only half of the pipelines: four pixel and two vertex ones. The performance degenerated accordingly; the RADEON 9600 PRO would lose to the RADEON 9500 PRO across a majority of applications. However, even this “cut-down” chip left no chances to its main rival from NVIDIA, the GeForce FX 5600 Ultra.
The progress in the graphics market didn’t stop at that, of course. ATI Technologies and NVIDIA Corporation develop and launch their new mainstream solutions, RV360 and NV36, respectively. You can find more information on these two graphics chips in our article called ATI RADEON 9600 XT vs. NVIDIA GeForce FX5700 Ultra. Just to remind you: the RV360 chip is designed with a new insulation material, the so-called low-k dielectric, which helped to add 25% to the core frequency, increasing it up to 500MHz. At the same time, the memory frequency remained the same as with the RADEON 9600 PRO, 300MHz (600MHz DDR). Of course, ATI implements very efficient technologies for better utilization of the memory bandwidth, but anyway 600MHz is rather low compared to the 950MHz of GeForce FX 5700 Ultra cards. Fast memory showed its advantages in high resolutions: graphics cards on NVIDIA’s GPU often beat RADEON 9600 XT in such cases, notwithstanding the overall more efficient architecture of the latter (R360).
Our today’s review is dedicated to an uncommon product. GeCube RADEON 9600 XT “Extreme” graphics card uses a new design of the PCB and comes equipped with a faster memory than ordinary products on this VPU. We are going to answer two questions: what advantages fast memory can bring to a RADEON 9600 XT-based card, and how the new PCB design influences the overclockability of the card (it’s no secret that the simplified PCB design has been the brake to overclocking memory chips installed onto such cards).
The retail package of the graphics card is decorated in the GeCube style: a red band to the right carries the company’s name, while the rest of the box front cover depicts a composition like “yet another visitor from hell”. A few slogans and stickers are scattered around:
The “Extreme” sticker is the most important as it tells you that the card is working at increased frequencies. The back side of the box contains some info about the features of the card like supported resolutions. When we open the box we find the following things:
Yes, the accessories are not so numerous, but it is quite enough for many users. It’s not the first time we see a universal RCA/S-Video cable included with a graphics card: Gigabyte RADEON 9600 PRO comes with a cable like that, too. The basic drawback of this cable is that one of the connectors will be hanging idle, adding more confusion to the heap of cables at the back of your system case.
Note that ATI Technologies has started taking additional payment from the manufacturers who want to enclose a coupon for a free copy of Half-Life 2 with their products. This is a rule for graphics cards on all ATI VPUs, save for the RADEON 9800 XT. As a result, RADEON 9600 XT-based cards are very likely to lose this free accessory pretty soon, too.
The cooling system of GeCube RADEON 9600 XT “Extreme” has nothing in common with any other they usually install on RV360-based cards. Frankly speaking, I wouldn’t call this cooling system completely new. It looks more like a variation of the reference cooler installed onto RADEON 9800 XT.
This time we have a smaller fan, while the heatsink is made of aluminum rather than copper. This should be enough, considering the low heat dissipation of the R360. If we dealt with an ordinary RADEON 9600 XT, I would even call this solution superfluous. We deal, however, with an Extreme product, so the choice of the cooling system may be appropriate. We will learn if this is true or not shortly.
The cooling system of GeCube RADEON 9600 XT “Extreme” contacts the VPU surface through a layer of thermal compound, which resembles the Cooler Master Premium stuff from Shin Etsu in its color and consistence. There is also a protective frame that should prevent the fragile edges of the graphics chip against physical damage. That’s good as the cooling system is heavy enough.
The memory chips give their heat to the heatsink through rubber-like thermal pads. I really doubt their heat-conductive properties, but their elasticity helps to create a tighter contact with the chips than when ordinary thermal paste is applied in between. The memory on the backside of the PCB has also been taken care of: a small aluminum heatsink is mounted on each pair of chips. The heatsink is small, but many RADEON 9600 XT-based cards go around without any heatsinks on memory chips at all.
The PCB design of GeCube RADEON 9600 XT “Extreme” breaks away from the original reference design of RADEON 9600 XT:
As you see, nearly every square inch of the PCB has gone through redesign, which is especially visible in the components of the power circuitry (they must have been enforced to provide stability at higher frequencies).
We don’t see a spot for the daughter card, but there’s a landing-place for a Rage Theater chip instead that’s responsible for the VIVO functionality. Curiously, the place for the power connector remained intact, although all RV350/360-based cards don’t actually require external power. The back side of the PCB also differs from the reference design, however slightly: we have different chips and in different locations than by the reference design.
Now about the “Extreme” title. When applied to a graphics card, it usually means that the card works at higher frequencies or has memory with lower access time onboard. In this case, we have both: chips from Samsung with an access time of 2.5ns work at 350MHz (700MHz DDR) (the graphics chip frequency remained standard, 500MHz). Theoretically, this memory can work at 400MHz (800MHz DDR), but the simple PCB design may become a hitch. We will check this out soon.
ATI’s OVERDRIVE technology can cause your graphics card no harm, but it can hardly make you happier. Practice suggests that RADEON 9800 XT and 9600 XT VPUs rarely receive any perceptible performance gain from this technology (it dynamically adjusts – overclocks – the frequency of the graphics chip depending on its temperature). It’s much simpler and more efficient to overclock the VPU manually (i.e. using special utilities) rather than hope for any bonuses from 13MHz higher chip frequency.
People at The Inquirer found out that ATI OVERDRIVE needs a special chip from National Semiconductor, the LM63, to work on RADEON 9600 XT-based cards. This chip is installed on graphics cards “Built by ATI”, but it is missing in RADEON 9600 XT solutions from other manufacturers (we didn’t find it in our GeCube RADEON 9600 XT “Extreme” either). Anyway, do you really need OVERDRIVE if you’ve got faster memory instead?
Just take a look at the diagram above to make up your mind about OVERDRIVE: if it matters to you more than the advantages of fast graphics memory.
GeCube RADEON 9600 XT “Extreme” graphics card sped up to 640MHz graphics chip and 400MHz (800MHz DDR) memory in our overclocking tests. The memory did work at its nominal frequency, which is a rare thing with many RADEONs. The system worked stably at those frequencies, without any visual artifacts, but another 10MHz above that resulted into an inevitable crash. As you see, the new PCB design is absolutely perfect. I think the overclockability of this card will draw the attention of those overclockers who are looking for a mainstream product in order to squeeze the maximum out of it.
The noise from the cooling system was not too loud, but anyway you can hear it against the noise produced by other system components. As for 2D quality, the image was sharp in all resolutions including 1600x1200@85GHz. Overall, we found no problems with this graphics card at all. There is only one thing that should be checked: how fast is this card with its 700MHz memory?
This is the testbed we used in our tests:
One remark is necessary. Our preliminary testing with the Catalyst 4.1 driver from ATI revealed one very unexpected peculiarity: the performance with enabled full-screen anti-aliasing and anisotropic filtering turned out much lower in many games than in case we tested with Catalyst 3.9. As soon as we installed Catalyst 4.1, the speed would go down irrespective of the testbed configuration. We are now waiting for ATI Technologies to comment on this fact.
At the same time, we should acknowledge that Catalyst 4.1 provides an overall notable performance gain except a few given cases. Since we haven’t yet got any comments from ATI on the strange performance of our RADEON 9600 XT with CATALYST 4.1 driver and enabled FSAA, we decided to offer you the performance results for both: driver version 3.9 and version 4.1.
We have also discovered one more interesting peculiarity of the GeCube RADEON 9600 XT “Extreme” working with the ATI CATALYST 3.9 and older version of the driver. The thing is that when you use any version of CATALYST driver different from 4.1, the card works unstably in a number of applications, so that they freeze or force the system to reboot. Moreover, the frequency of this unpleasant freezing effect didn’t depend on the working frequencies of the card. This phenomenon could have been considered the drawback of our particular sample, if it had revealed itself with the driver version 4.1, too. But with this version of CATALYST the card went through all the tests several times without any problems even at higher working frequencies, not to mention the nominal ones. There is some mystery about it, which we cannot explain logically yet. Maybe it is somehow connected with the performance difference we observe with CATALYST 3.9 and 4.1. Even when we installed the drivers included into the supply (CATALYST 3.6), the situation didn’t turn any clearer: the card continued to have some stability issues at work from time to time.
This is the list of the games and applications we used during our test session for benchmarking purposes:
First person 3D shooters:
Third person 3D shooters:
Simulators:
Real-time strategies:
Semi-synthetic benchmarks:
Synthetic benchmarks:
The graphics quality settings were the highest in each application.
The very first benchmark reveals the advantages of the approach used by GeXcube Company in their GeCube product: the positive effect of the faster memory is undeniable. Overclocking is rewarding, too. You may notice that it’s more beneficial to increase the memory frequency, rather than the graphics chip frequency, since RADEON 9600 PRO is clocked just a little slower than the standard 9600 XT.
In this case the use of the new driver version led to quite a significant performance drop in heavy work modes. If we are talking about “fast memory” and its influence, then I should admit that it is not so tangible here, though it definitely exists.
We see a similar picture in this 3D shooter. GeCube card shows its best in 1280x1024 resolution, but lacks power for 1600x1200 (if you don’t resort to overclocking). This highest resolution is for graphics cards with a wide memory bus (RADEON 9800 PRO) or with an ultra-fast memory (GeForce FX 5700 Ultra).
The VPU frequency affects the performance, too. As you see, our overclocked RADEON 9600 XT “Extreme” won two first resolutions against RADEON 9800 PRO! In fact there is nothing to be surprised with: the game is pretty simple, there are a lot of textures and nice geometry, but it doesn’t practically involve any exclusive features of modern VPUs. Well, the Quake 3 engine, which runs Elite Force 2, is a few years old already.
In Elite Force 2 the use of CATALYST 4.1 driver on the contrary resulted into a slight performance growth. Here you can also see the effect made by faster memory, though it is not so evident.
The UT 2003 engine is much heavier on the graphics card than Quake 3, so no participant can compete with RADEON 9800 PRO in this test. The game is rich in textures, so GeCube leaves the standard RADEON 9600 XT behind, especially in low resolutions. GeForce FX 5700 Ultra with its improved vertex unit and fast memory upholds the banner of NVIDIA.
In Unreal Tournament CATALYST 3.9 driver behaved quite strangely when we enabled FSAA and anisotropic filtering: the performance our testing participants showed with this driver appeared unexpectedly low. Moreover, even the complete uninstallation, registry clearing and repeated installation of the driver had no effect on the results. CATALYST 4.1, on the contrary performed extremely correctly and demonstrated very nice performance: our hero left GeForce FX 5700 Ultra simply no chances.
Halo: Combat Evolved shows a completely different picture than UT2003: the engine of this shooter uses pixel shaders very actively and the number and frequency of the pixel pipelines determine the speed. As you see, there is a very small difference between the ordinary RADEON 9600 XT and GeCube card. Meanwhile, we can give GeCube card a boost by overclocking the graphics chip.
Deus Ex 2: Invisible War is a first-person 3D shooter with a pretty intricate and puzzling plot. The game features the Unreal Warfare engine and uses most of the capabilities offered by contemporary VPUs. Alas, the game is too hungry for resources: you can play it comfortably either in resolutions below 1024x768 or on a top-end graphics card. This shooter formally supports FSAA, but I couldn’t make it work normally even a single time: the typical “jaggies” indicated that anti-aliasing hasn’t been enabled. Even when we tried forcing it from ATI’s control panel, we failed.
The performance in Tron 2.0 depends on the graphics processor frequency as well as on the speed of the memory subsystem. The above diagrams show that the 128-bit memory bus becomes too narrow for high resolutions. Overclocking GeCube RADEON 9600 XT “Extreme” brings a negligible benefit here.
The new driver ensured a pretty good performance gain in Tron 2.0. No visual problems have been noticed here.
The next generation of computer games will have ravenous appetites, this is hardly doubtful. They will require an efficient graphics processor architecture and high graphics chip and memory frequencies to run acceptably fast. In this case, the Highly Anticipated DirectX 9 Game is more sensitive to the VPU frequency, and GeCube RADEON 9600 XT “Extreme” is just a little ahead of the ordinary RADEON 9600 XT. (For more details on the performance of contemporary graphics accelerators in this highly anticipated next generation game see our article called A Next-Generation DirectX 9.0 Game Graphics Performance Preview)
The scene displayed in the second demo shows the advantages of the fast memory used in GeCube graphics card. The four-pipelined architecture of RADEON 9600 hits upon its ceiling in 1024x768 – even the full overclocking of GeCube RADEON 9600 “Extreme” gives no reward here.
There is very small difference between the two versions of RADEON 9600 XT VPU in the third demo of the upcoming DirectX 9 game, but overclocking brings considerable performance gains. The scene consists of underground passages filled with water, and the amount of textures is relatively small. That’s why the speed of processing geometrical data and executing pixel shaders matters more here, while the bandwidth of the memory subsystem isn’t crucial in this demo.
Another DirectX 9 game we are all impatiently waiting for (with some reservations, it can be referred to the next generation of games, too) appreciates the faster memory of the GeCube, but VPU overclocking is even more rewarding. (For more details about the performance of contemporary graphics accelerators in this game please see our article called Yet Another DirectX 9 Game: Lost Oblivion in Chernobyl).
Unfortunately, we couldn’t pass the tests in this game with CATALYST 3.9 driver that is why you do not see the results for graphics cards tested with this driver version here. As for the version 4.1, the performance of our testing participants is quite good.
This demo is rich in high-resolution textures, but there are no characters, save for the player. The advantage of the GeCube RADEON 9600 XT “Extreme” over its standard companion is obvious and immense.
Here the new driver also guarantees excellent performance.
The third demo resembles the second one, but it is geometrically heavy. The Extreme version of RADEON 9600 XT leaves the standard version behind.
In the third gaming scene GeCube solution is an indisputable leader.
Splinter Cell is a game that extensively uses various shader-based effects. Performance is determined by both: graphics chip and memory frequencies. The positive effect of the 700MHz memory is evident.
The new Prince of Persia is a beautiful game with complex graphics. Its numerous visual effects are created by pixel and vertex shaders. GeCube RADEON 9600 XT “Extreme” is good here, too, outperforming the standard version of the RADEON 9600 XT. Note that this game also doesn’t work correctly with FSAA.
The game is great and GeCube RADEON 9600 XT “Extreme” is great, too, leaving behind all the competitors of its own class. Overclocking is surprisingly satisfying: GeCube gets close to RADEON 9800 PRO.
The CATALYST 3.9 driver failed to work fine in Max Payne 2 only in case of extreme overclocking of the card. As you see, CATALYST 4.1 didn’t affect the performance of our testing participants even with enabled FSAA and anisotropic filtering, while the effect made by faster memory is more than evident.
The genre of flight simulators is rather specific from the technical point of view. A typical scene usually includes a few textures of a very high resolution, which represent the sky and the earth. No wonder fast memory becomes a huge bonus here, as our diagrams confirm.
In Il-2 Sturmovik our solutions also benefit in heavy work modes from the CATALYST 4.1 driver, while faster memory hardly has any effect on the performance at all, unlike the “pure” mode.
X2: The Threat runs faster if you have a graphics card with fast memory, while graphics chip overclocking is of little effect. If you are a fan of this game and want to get a higher FPS rate, you should buy something like RADEON 9800 PRO/XT or GeForce FX 5950 Ultra.
X2: The Threat also froze a few times when we used CATALYST 3.9 driver, even when our solution worked at the nominal frequencies. The performance with the installed CATALYST 4.1 driver is very optimistic, but still insufficient for comfortable playing.
GeCube RADEON 9600 XT “Extreme” graphics card works faster than the ordinary RADEON 9600 XT in this Formula 1 simulator, too, especially in high resolutions.
The new drivers do not tell in any way on the performance, although the advantage over the regular RADEON 9600 XT is indisputable. When we tried testing the overclocked card with CATALYST 3.9 driver, the system refused to work and rebooted.
The same goes for this real-time strategy game.
Unfortunately, with the new drivers installed and forced FSAA and anisotropic filtering we got almost the same results as in the pure mode. Moreover, we saw the aliasing effect, which indicated that the FSAA mode hadn’t been enabled. I have to admit that CATALYST 4.1 driver can hardly be considered bug-free even despite their nice performance.
The difference between the ordinary RADEON 9600 XT and the “Extreme” version is minimal in Aquamark3. Graphics processor overclocking brings more performance gains (compare the results of RADEON 9600 PRO and RADEON 9600 XT here).
Unfortunately, we couldn’t complete the test session in Aquamark3 with CATALYST 3.9 driver installed. In case of newer CATALYST 4.1 driver version everything worked fine and the card performed at a very nice level with the competitors.
Final Fantasy is a highly anticipated online multiplayer role-play game and it also runs faster with faster memory. GeCube RADEON 9600 XT “Extreme” and GeForce FX 5700 Ultra are favorites here, and overclocking allows GeCube to score more than 5000 points!
With enabled FSAA the faster memory working at 700MHz has a noticeable effect. Some performance improvement is also noticeable with the CATALYST 4.1 driver installed.
GeCube graphics card surpasses the standard RADEON 9600 XT in 3DMark03. Let’s take a closer look at the results obtained in this test package.
The first test fits into the flight simulator category with the appropriate results: fast memory means improved performance.
GeCube RADEON 9600 XT demonstrates stable advantage over the regular card version, however we also see a significant performance difference between the two CATALYST versions: the card works slower with the new driver than with the older one. We saw something similar in RTCW: Enemy Territory, although this time we deal with Direct3D API and not with OpenGL. Let’s see what other gaming tests rich in pixel shaders will show.
The second game test is actually a 3D shooter, and the advantages of faster memory are not obvious until we get to high resolutions.
In the second gaming test when we enable FSAA and anisotropic filtering we can also notice the difference between the two CATALYST driver versions. Of course, the demonstrated numbers are not very high in both cases, but the card with CATALYST 4.1 is evidently lagging behind. As for the faster memory, it does improve the performance of the card, despite the overall low performance. However, the three last gaming 3DMark03 tests have never been remarkable for high FPS rates.
In the third gaming test the use of faster memory is more advantageous than in the second one.
The situation we observed in the first two benchmarks with enabled FSAA repeats itself in the third one too: CATALYST 4.1 again shows its weaknesses demonstrating lower results than CATALYST 3.9. Speaking about the performance difference between the cards equipped with 600MHz and 700MHz memory, our supposition about RADEON 9600 XT lacking faster memory once again proves true.
It is not the VPU frequency any more, but rather its architecture that determines the winner in the fourth test. Overall, the advantage of faster memory is small, but anyway evident.
The advantage of GeCube card over the standard RADEON 9600 XT is again evident, just like the advantage of CATALYST 3.9 driver over the 4.1 version.
All in all the new ATI CATALYST driver version provides a significant performance gain in most cases although it doesn’t arouse any superior joy: the problems which might occur during the driver installation are more than evident. So far we encountered some performance issues in certain OpenGL games, such as RTCW: Enemy Territory and 3DMark03 test package. We also discovered some problems with the implementation of forced FSAA, namely we failed to enable FSAA in Command & Conquer Generals.
We are still expecting an official comment from ATI Technologies officials, so that we could finally find out exactly what lies behind the above described problems. As soon as we receive this info, we will immediately share it with you.
I would also like to point out that a lot of contemporary games using new API features, such as DirectX9 cannot work correctly with enabled FSAA. Sometimes even anisotropic filtering works incorrectly in them or doesn’t work at all. All this forces the users to give up the means for image quality improvement. We hope that the software developers will do their best to improve the situation. So far, CATALYST 4.1 can only be recommended if you made sure it works fine in those applications you are planning to run. Otherwise you may get a performance drop instead of performance improvement.
It’s easy to make a verdict here: fast memory working at 350MHz (700MHz DDR) instead of the standard 300MHz (600MHz DDR) is exactly the thing RADEON 9600 XT has been lacking since the launch of its competitor, GeForce FX 5700 Ultra, with its ultra-fast GDDR-2 memory.
Although RADEONs feature efficient technologies for more optimal use of the memory bandwidth, 300MHz proved to be not enough to compete with the rival in applications that don’t use pixel shaders. GeCube RADEON 9600 XT “Extreme” graphics card is a fitting answer to NV36-based products. Its bonus - the faster memory - makes it more competitive against GeForce FX 5700 Ultra, especially in high resolutions. On the other hand, the memory frequency remained relatively low, thus allowing simple and inexpensive PCB design. The new graphics card proves exceptionally good in games using shaders.
Unlike cards on GeForce FX 5700 Ultra, RADEON 9600 XT-based products equipped with fast memory should boast lower power consumption and heat dissipation. Such cards don’t also require additional power supply, which makes them more appealing to the end-user.
Talking about this particular card, GeCube RADEON 9600 XT “Extreme”, I should acknowledge that it is an exciting product: high performance in modern games, simple design, excellent cooling system and good overclockability are the options you usually look for in a high-quality and fast graphics card of the mainstream class. So the card itself is good, but you should be aware of the humble set of accessories. Of course, accessories do not tell on the quality and technical characteristics of the card, but anyway.
Highs:
Lows: