by Alexey Stepin
09/01/2003 | 10:38 PM
It’s no secret that most high-end graphics cards follow the reference design from the graphics chip developer. This is true for both: ATI Technologies and NVIDIA. I can’t say this is bad, because the graphics chip maker polishes off its reference design, eliminating any bugs, and thus provides higher product reliability.
Alas, every medal has its reverse side: high-end graphics cards have turned out as like as two peas. It is not good for those manufacturers that should attract the customers somehow. Those who were smart enough found ways to do it. A graphics card can appeal with rich accessories, a beautiful box and an original cooling system. Albatron was among the smartest, and its GiGi GeForce FX5900 PV graphics card will be the main hero of our today’s review.
This card arrived into our test lab in a quite unusual way. It was packed into a big black bag instead of a paper or plastic box.
None of us has ever seen a package like that. Well, the card’s performance is of course in no way connected with its wrapping, but it’s always nice to see some extra accessories. By the way, this bag is large enough to fit for a notebook, for example. The biggest section of the bag contained the card itself, while cables, CDs and the user’s manual were all neatly assorted into respective pockets.
The plastic box, in which the bag itself was wrapped, also deserves a closer look.
The fantasy style is very popular nowadays, but Albatron managed to stand out remarkably here, too. Many manufacturers often depict various monsters on their product package: dragons, gargoyles, hellhounds, and other aggressive specimens from the medieval mythological zoo. The monster should evidently reflect the infinite power of the graphics accelerator. Albatron went another way, however. The package of GiGi GeForce FX5900 PV displays no dragon, but a cute and attractive pointed-ear elfin girl. It bears a striking resemblance to another elf – the Dawn from NVIDIA’s demo program.
Well, although the fragile girl can hardly be associated with the raw power of the NV35 graphics core, I felt more pleased to look at her rather than at all those hackneyed monsters.
Now please have a look at the list of accessories you get with the graphics card:
Not bad all in all, but I think that they might have included more and better games into the software bundle.
The PCB layout of this card doesn’t in fact differ from NVIDIA’s new reference design, save for its blue color. The only distinguishing feature of this solution is the three-fan cooling system. The heatsink mounted onto the card is covered with a plate shaped like a wingspread bird. Although the cooling system seems to occupy one slot, you will anyway have to keep the next PCI slot free to provide air for the fans. Otherwise, considering the hot temper of NV35, the card may behave unpredictably, and even go down altogether. Besides, the heatsink that seems to be of copper is not a real cooper piece in fact. When we removed it, we found that the thermal paste layer is not everywhere thick enough to provide tight contact to the memory chips. So, we had to scrub it away and apply new silver paste from Titan.
When you see that there are three fans, you may expect this cooling system to be really noisy, maybe even as noisy as the legendary original FlowFX cooling solution. However, there is no need to worry that much. In the nominal mode, only two first fans are rotating, while the rear one is off. It is powered up by the Wise Fan system in case one of the main fans fails. The controller of this system is placed in the centre of the heatsink and looks like a small board with a couple of chips. A bright-blue LED is located right next to it. Actually, we don’t quite understand the logic of the cooling system designers. In fact, the fan is very unlikely to go down. It could be better to make the third fan turn on after some critical temperature has been reached. GiGi GeForce FX5900 PV did produce very little noise at work, though. Under the heatsink, we found memory chips from Hynix with 2.2ns access time. This gave us some hope for successful overclocking.
Unlike ASUS V9950, Albatron GiGi GeForce FX5900PV graphics card is equipped with the VIVO function. The PCB carries a Philips SAA7108AE chip that is responsible for Video-out, while Video-in is provided by the controller integrated into NV35.
Is it possible for a graphics card to have poor 2D image quality when the PCB carefully follows the reference design? Sometimes, yes. Unfortunately, the picture in 1600x1200 resolution at 85Hz refresh rate could hardly be called good. Of course, it was not corrupt altogether, but the obvious “fuzziness” made it impossible to work comfortably in a word processor. Having reduced the refresh rate to 75Hz, we got a better image, but it became clear-cut only in 1280x1024. For your reference, the ASUS V9950 graphics card produced crystal-clear picture in all resolution, including 1800x1440@75Hz (the upper limit of the display). Again, the quality of 2D also depends a lot on a given display and graphics card.
Overclocking, on the contrary, was quite enjoyable. Using additional cooling, we sped up the card to 480/940MHz. As all GeForce FX owners should know, these graphics cards work at different clock-rates in 2D and 3D modes. Our GPU did 2D at 300MHz. It was rumored on the Web that overclocking the GPU in this mode may bring artifacts into the picture. Well, they were right. We increased the frequency to 450MHz to find that if you move a window along the Desktop, it will get covered with stripes and speckles. Although in 3D the card worked fine at this frequency.
It’s rather hard to determine what’s wrong with the card: if these are some internal bugs of the GPU or RAMDAC, or bad design of the card’s PCB. Anyway, this problem does exist for many graphics cards.
Now a few words about the VIVO functions.
The VIVO functions of this graphics card work as they should do. After you install appropriate drivers and re-boot Windows, the Device Manager lists four new devices: nVidia WDM A/V Crossbar, nVidia WDM TV Audio Crossbar, nVidia WDM TVTuner and nVidia WDM Video Capture. From this point on, you can work with the TV-input in any software that knows to use it. The quality of video capture is quite average, as in any devices based on the Philips SAA7108 controller.
ASUS V9950 graphics card had BIOS version 220.127.116.11, while Albatron GiGi GeForce FX5900 PV used version 18.104.22.168. Moreover, an NVIDIA representative once told us that all GeForce FX 5900 cards raised memory timings if you overclocked the graphics memory above 900MHz. It is of course done to ensure higher stability. So, we decided to check out how it tells on the performance and carried out a few extra tests at different frequencies. The testbed configuration remained the same:
And here is the list of our benchmarks:
We also tested the graphics card in Codecult CodeCreatures Benchmark Pro using unofficial 44.90 drivers. It is rumored on the Web that NVIDIA is “sharpening” its drivers for this test (as well as for a few others). This news proved true. When we used Detonator FX 44.90, the card’s performance dropped significantly in the CodeCreatures test. Well, now you will be able to see, how big is the influence of the drivers on the performance: even a minor change in the driver may push the performance to change dramatically in either direction.
We will start with the promised investigation on the “dependence of GeForce FX performance on the GPU and memory working frequencies”. It was carried out in the following benchmarks:
Here are the results we got:
So, what do we have? The graphs show that we profit most from overclocking the GPU from 400MHz to 450MHz. Increasing the graphics memory frequency to 900MHz is advantageous, too. After that, there is no perceptible performance gain from overclocking. A few extra fps can be written off to the measurement error. Our verdict is that overclocking the graphics memory in GeForce FX 5900 cards above 900MHz doesn’t practically bring any performance gains: the accompanying growth of the memory latency negates the effect of higher clock frequencies. So, people who are into overclocking should pay more attention to the GPU rather than memory. You should also keep in mind that the latency depends on the card’s BIOS version. Thus, graphics cards with different BIOS’s may show somewhat different results.
Now, let’s turn to traditional tests.
Albatron’s baby feels at home in the old, but robust Quake 3 Arena. ATI’s GPU is left behind. Overclocking makes the gap even wider.
It’s all quite contrary in Return to Castle Wolfenstein. The game seems to dislike GeForce FX with 44.03 drivers, putting RADEON 9800 PRO on top. We also see a curious picture in 1280x1024 resolution: Albatron card did better at regular frequencies than at overclocked frequencies. We double-checked this phenomenon by reinstalling the drivers and rebooting the system. The reason for this may lie in the depths of NVIDIA drivers or in the already mentioned high memory latency.
GiGi GeForce FX5900 PV wins in this test, but with a very slight advantage. This game is rather indifferent to the type of the graphics card, so we are going to replace it in our test package in the near future.
GeForce is slower than RADEON in Serious Sam, although its overclocked version is the fastest of all in all modes and resolutions. High GPU and memory frequencies are the ones that matter here.
The competitors go neck and neck in nominal modes, although overclocking again makes GiGi GeForce FX5900 PV the winner. However, RADEON 9800 PRO is the best in the hardest mode:1600x1200, 4x full-screen anti-aliasing and 8x anisotropic filtering.
As you may remember, the graphics cards from ATI and NVIDIA should only be compared in Splinter Cell in modes without FSAA and AF (thanks to NVIDIA software guys). The weak ALUs of the Albatron card that cannot perform shaders quick enough drag it down in this game. As for the modes with FSAA and AF, they only serve to evaluate the advantages provided by overclocking. For details about this game and NVIDIA GPUs, please refer to our ABIT Siluro FX 5600 Ultra DT Review.
RADEON 9800 PRO wins the first gaming test from the 3DMark 2001SE suite, save for a few modes where the overclocked GiGi GeForce FX5900 PV gains the upper hand. They are light modes without extra workload (such as FSAA + AF).
This time the graphics chip from ATI Technologies remains unrivalled. Even the faster, overclocked Albatron cannot do anything about it.
It’s slightly different in the third gaming test: RADEON 9800 PRO is fast, but the overclocked GiGi GeForce FX5900 PV is faster in the heavy modes.
Here, the picture is quite contrary: GiGi GeForce FX5900 PV beats the competitor in the heavy mode. However, in easy modes our hero manages to beat the rival only after we overclock it to 480/940MHz.
The new 3DMark definitely favors ATI Technologies’ graphics chips, but in the light modes only. When full-screen anti-aliasing and anisotropic filtering are enabled, GiGi GeForce FX5900 PV proves better.
The overclocked graphics card from Albatron is the winner of the second gaming test. The gap is even wider when we turn on FSAA and AF. At its nominal frequencies, this graphics card is slightly slower than RADEON 9800 PRO in the light mode.
The third game test ranks the cards up just like the previous one. However, RADEON 9800 PRO looks better here.
The fourth game test, Mother Nature, will remain a shame for NVIDIA forever. As you see, the NV35 graphics core is perfectly helpless in the test that uses version 1.4 and 2.0 shaders. However, new drivers from NVIDIA may “improve” this situation.
Comparing the results of the GiGi GeForce FX5900 PV card with different drivers, we see that NVIDIA software developers have had a hand in here, too. The unofficial driver version 44.90 shows just what we should see and the RADEON 9800 PRO wins the test.
The new official version of drivers from NVIDIA may either be free from any “optimizations” or (which is more probable) bring even more of them.
The graphics card we got for our tests left an ambiguous impression. It is pretty, comes with many accessories, and shows good performance and excellent overclockability. On the other hand, it produced poor picture in 2D and had this minor defect of the cooling system: loose contact between the heatsink and the memory chips surface. The new BIOS version is also less speedy, but not dramatically so.
The packaging (bag) looks original, but is of little use. You can only carry a notebook in it, but notebook owners usually have better bags, designed exactly for their computers. However, the package of Albatron GiGi GeForce FX5900 PV will undoubtedly draw customers’ attention, which is exactly what the manufacturer wanted. The cute elfin gal on the package serves this purpose, too. By the way, Albatron GiGi GeForce FX5900 PV may make a good gift due to its nice exterior.
The VIVO functions work as they are supposed to. Owners of analog camcorders who do not demand superfluous quality may use them to their advantage. Games that come with the card are relatively old, but maybe you didn’t play them when they were younger?
As for performance, we should say the same thing as in our ASUS V9950 Review. Albatron GiGi GeForce FX5900 PV is very fast in modern games and applications, but slow shaders execution makes its future rather vague. This is true for all GeForce FX-based graphics cards. So, if you don’t care about the games of tomorrow, and resolutions of 1280x1024 and higher, or want to make a gift to your friend, consider this graphics card. It may suit you well.