ATI RADEON 9600 PRO Graphics Chip in Detail: FIC R96P Graphics Card Review

We will test new solution based on the ATI RV350 VPU (RADEON 9600 PRO) from FIC. We are going to figure out how the performance of this graphics solution depends on the texturing quality settings, and to study the theoretical performance potential behind the RV350 chip compared with its closest rival: NVIDIA GeForce FX 5600 Ultra.

by Alexey Stepin
09/16/2003 | 04:25 AM

We managed to get a new graphics adapter based on the ATI RV350 VPU (RADEON 9600 PRO). This time it was FIC R96P graphics card. We decided to undertake a detailed test session in order to figure out how the performance of this graphics solution depends on the texturing quality settings, and to study the theoretical performance potential behind the RV350 chip.

 

Well, we will tackle these questions later in or today’s review. And in the meanwhile get acquainted with our today’s hero: FIC R96P graphics card.

Package and Accessories

The card came to us in a retail package, i.e. packed in a colorful attractive box designed in black-and-red palette. It looked a little bit similar to FIC A92 (RADEON 9200) package. However, instead of the jumping silver panther we saw some kind of  furious hell creature looking like a bull but with metal horns and predatory teeth. Besides, the box is decorated with ATI Technologies’ logo and indicates the amount of onboard graphics memory: 128MB.


Here is what we found inside the box:

As you see, this product can’t boast a rich accessories set, however, everything one may need to install and use it effectively is present.  The user’s manual is pretty detailed, it is packed into a tight plastic film together with the software CD.

Closer Look

When I took a closer look at the FIC R96P graphics card the bright-red PCB immediately caught my eye. Another thing that I also noticed right away was a pretty modest cooling system used on this solution, which was very similar to the original RADEON 8500 cooler. It seems that 0.13micron production technology allows to do with a compact and noiseless cooling system like that.

As you see, the PCB of FIC R96P is very close to the reference design from ATI. Only the PCB color is different, which is not much of a deal, actually. When I removed the cooler I found a layer of standard “chewy” thermal interface, which is very efficient in nominal working conditions and also protects the graphics chip die from damage during cooler installation and removal. Nevertheless, I immediately wiped it off and replaced with a generous layer of thermal paste, because I was going to do some overclocking later on.

The PCB is equipped with 8 BGA memory chips with 2.8ns access time working at 300MHz (600MHz DDR) frequency. The core was certainly working at 400MHz, just as it was supposed to.

All in all, FIC R96P doesn’t boast any outstanding peculiarities: it is a well-done RADEON 9600 PRO, with a quality PCB design following the reference from ATI.

Overclocking, Noise and 2D Image Quality

Having installed the graphics card into the AGP slot and started the system, I didn’t hear any noise from the fan: it was unbelievably quiet! Against the background of contemporary large and noisy cooling systems this solutions seemed to me nearly ideal. I have to point out that when the graphics card worked at nominal frequencies the heatsink didn’t heat up too much, however, during overclocking the default cooler seemed to be nor quite safe for proper cooling that is why we provided additional side airflow coming from a standard 80mm fan.

The memory chips with 2.8ns access time gave us some hope for pretty good overclocking results, and I have to admit that all our expectations came true. The graphics card from FIC worked stably at 500MHz chip and 760MHz (380MHz DDR) memory frequencies. Nevertheless, I started noticing some weird artifacts in gaming applications after a while that is why  we had to reduce the frequencies to 490MHz for the chip and 740MHz (370MHz DDR) for the memory. I think I have every right to say that the graphics memory of FIC R96P showed simply excellent overclockability. However, as for the graphics processor, the achieved result is not very impressive, especially against the background of TYAN Tachyon G9600 PRO, which overclocked up to 525MHz chip frequency (see our TYAN Tachyon G9600 PRO Graphics Card Review for more details). As you remember, our hero managed to prove really stable only at 490MHz core clock, however, a guaranteed 90MHz frequency gain above the nominal frequency deserves all our praise.

As for the 2D image quality, it appeared simply perfect, just like by FIC A92 (see our FIC A92 (RADEON 9200) Graphics Card Review for details). The image was very clear-cut in all resolutions including 1600x1200@85Hz.

In fact, our evaluation shouldn’t be regarded as absolutely objective, because this parameter depends a lot on the monitor you are using and on the quality of the cabling. Nevertheless, I would strongly recommend FIC R96P for work with two-dimensional graphics at high resolutions.

Testbed and Methods

The focus of our today’s article will be the influence of texture quality settings on the graphics card performance. Well, let’s pass over to our exciting investigation. As you know, the control panel for Catalyst drivers is rich in all sorts of settings. We are mostly interested in the sliders responsible for texture rendering quality (Texture Preference menu item). We undertook this investigation in order to find out how the above described settings affect the performance of our graphics accelerator in 3D. for a better performance analysis we also took both versions of NVIDIA GeForce FX 5600 Ultra graphics solution working at 350MHz/700MHz and 400MHz/800MHz respectively.

Our testbed was configured as follows:

For our test session we used the following benchmarks and applications:

Influence of Texture Quality Settings on the Graphics Card Performance

Since this is mostly a theoretical investigation, we decided to pay due attention to the was FIC R96P will behave in synthetic benchmarks and to its performance in those applications. At first, we carried out a few fillrate tests, in order to find out how the graphics chip frequency affects it. Here are the results obtained:

You can easily notice that the memory frequency influences the fillrate in case of single texturing more than the graphics chip frequency does. Just compare the testing participants working at 500/600MHz and 500/500MHz. The fillrate drops as evenly as the level of anti-aliasing grows.

With enabled multi-texturing the results turn out more interesting. The graphics core frequency becomes a very important factor: just look at the performance gain provided by the frequency growth from 400MHz to 500MHz! You can also see that the growth of the FSAA level, increases the dependence of the fillrate values on the memory frequency. This dependence is especially evident on the last two graph lines (for 400/500MHz and 400/600MHz).

For a better comparison have a look at the results obtained during a similar testing carried out for NVIDIA GeForce FX 5600 Ultra:

Well, we don’t notice anything interesting here. The fillrate is all in all higher than by FIC R96P except the FSAA 6xS mode. As I have already mentioned to you, flexible architecture of NV31 allows it to change the pipeline configuration. When we have only one texture, this graphics processor works according to the following formula: 4 x 1, i.e. 4 pipelines with 1 TMU on each. However, as soon as we enable multi-texturing the chip switches to 2 x 2, i.e. 2 pipelines with 2 TMUs on each.

As you see, GeForce FX 5600 Ultra gives in when multi-texturing is enabled. It is probably because the 2x2 scheme it uses is not so efficient in those modes that work with FSAA.

Fillrate Investigation

Besides, 3DMark 2001SE we also decided to undertake a little investigation of the fillrate speed by FIC R96P and GeForce FX 5600 Ultra with the help of a special benchmark. This test draws a polygon on the screen and then there are from 0 to 4 textures to be laid onto it with 512x512 pixels big. In the first case, that is when there are 0 textures to be laid, the color of the pixel is calculated as interpolation of polygon vertex colors. This program tests the graphics card in three modes. So, here comes the first mode: color writes enabled, Z-writes also enabled:

As you see, the result is pretty common: since FIC R96P based on RADEON 9600 PRO  features 4 pipelines with 1 TMU on each, it spends an extra clock to process any new texture. It means that the fillrate drops down almost evenly. And GeForce FX 5600 Ultra suffers a sharp performance hit, when there are two textures laid, so that it goes even beyond FIC R96P. this happens because of the 2x2 mode (2 pipelines with 2 TMUs on each).

And now let’s disable Z-writes:

The result is almost the same. Now let’s enable the Z-writes and disable the color-writes:

In this case the textures cannot be laid at all that is why all the testing participants almost reach their  theoretical performance maximum. The winner this time is FIC R96P.

Performance and Texture Quality in Synthetic Benchmarks

Now let’s pass over to synthetic 3DMark tests. Here is what we obtained in 3DMark 2001 SE:

High working frequencies help GeForce FX to beat its rivals in this geometry test, however, the overclocked FIC R96P almost restores the parity. High working frequency of the new GeForce FX 5600 Ultra helps it to outperform the rival.

When we enable 8 light sources in the 3DMark 2001SE High Polygon count benchmark, FIC R96P based on the RADEON 9600 PRO processor leaves the rival behind, even though pretty modestly. A similar thing happened with the original RADEON 256 when it managed to work faster than the competing GeForce2 GTS in situations with dramatically increased workload because of numerous light sources.

In DirectX 8.0 Vertex Shader test ATI’s RADEON 9600 PRO (FIC R96P) manages to outrun the GeForce FX 5600 Ultra. There are absolutely no surprises here, as vertex shaders processing has always been the bright side of the whole R300-family of VPUs. Note that performance of the GeForce FX 5600 Ultra (400/800MHz) does not drop significantly as the resolution increases.

The same picture may be noticed in DirectX 8.0 Pixel Shaders benchmark. The advantage of the RADEON 9600 PRO (FIC R96P) gets bigger as the resolution grows.

This is a dramatic failure of GeForce FX 5600 Ultra, really! In fact, this is not at all surprising, as this test uses pixel shaders version 1.4, which is definitely not NVIDIA’s trump. If you take a closer look at the results, you will see that GeForce FX is exactly twice as slow as FIC R96P. Probably the reason is connected with the architectural peculiarities of each graphics chip.

During Dot3 bump mapping the rivals go almost neck and neck.

On the other hand, the RADEON 9600 PRO-based FIC R96P boasts clear leadership in 3DMark03 EMBM benchmarks, however, the gap decreases in high resolutions.

Now let’s pass over to synthetic 3DMark03 tests:

In DirectX 9.0 Vertex Shader benchmark FIC R96P substantially outstrips the competitor once again. Well, architectural peculiarities are not 3DMark dependent, are they?. :)

As you see, FIC R96P on RADEON 9600 PRO works perfectly well with pixel shaders version 2.0, which unfortunately cannot be said about NVIDIA GeForce FX 5600 Ultra. Once we start thinking about the future games, we understand that the today’s generation of NVIDIA’s GPUs will be hardly suitable for them because of the extremely low pixel shaders processing speed. This is just a theoretical test, anyway, but 24fps in 64x480 can never be called an acceptable result. Do you agree?

Ragtroll test is pretty interesting, because it is not only a good performance test for vertex shader unit, but also a well-balanced benchmark for the entire system, as it uses a real physical model for falling trolls, which loads the CPU quite a bit. Both FIC R96P and the GeForce FX 5600 Ultra demonstrate pretty adequate results with RADEON 9600 leading in low resolutions and GeForce FX 5600 Ultra winning in high resolutions.

All in all, 3DMark03 is exactly the place where RADEON 9600 PRO (FIC R96P) shows its best. The results obtained in synthetic benchmarks show very well, which of the two has a more promising future ahead.

In addition to the 3DMark benchmarks, we tested the RADEON 9600 PRO and GeForce FX 5600 Ultra using the software you are already familiar with from our NV35, NV30 and NV31 reviews (see our Video section fot details). The results were fully predictable.

Performance and Texture Quality in Gaming Benchmarks

Now let’s pass over to the investigation of performance dependence on the texturing quality settings. For this purpose we took Quake3 game, which serves as a basis for the whole bunch of games based on the same engine.

As you see playing with Texture Preference engine hardly tells on the performance of our FIC R96P: in the worst case the difference makes about 2 frames per second. We see a totally different picture in case of GeForce FX 5600 Ultra, which performs well only when working at 400/800MHz. Of course, enabling Performance and High Performance modes leads to a significant performance growth, however, the price we pay for this growth is very high: tri-linear filtering gets practically disabled, the textures start “rippling” and the borders between the MIP-levels become very distinct alongside with more other “nice things”. The quality of the final image becomes very low, that is why we wouldn’t recommend turning on these modes if you care at all about your image quality.

In Unreal Tournament 2003 there is bit different picture compared to the Quake 3 Arena. Texture quality influences performance a little bit more significantly in this benchmark, even though the largest gap between the lowest and highest Texture Quality is just 10 fps – there is hardly a point to sacrifice image quality for just 10 fps…

The winner here is the FIC R96P, while the GeForce FX 5600 Ultra can only compete with it in 1600x1200 because of its 800MHz memory.

And now let me make a small announcement. In the ongoing review of contemporary graphics accelerators based on RV350 and NV31, we are going to discuss how greatly their performance depends on the FSAA level and anisotropic filtering quality

Performance: Gaming Benchmarks

Quake3 Arena

FIC R96P falls behind in light modes and dashes forward when FSAA and anisotropic filtering are enabled. The new version of GeForce FX 5600 Ultra looks much better than the old one.

Return to Castle Wolfenstein

This game still doesn’t like GeForce FX graphics solutions, even though the new drivers have recently come out. The laurels get won by FIC R96P.

StarTrek: Elite Force 2

We decided to replace the Jedi Knight II game with StarTrek: Elite Force 2, which is also a sci-fi first person 3D shooter based on the StarTrek universe instead of StarWars. However, this is not the major advantage of this game. The most important feature is its high scalability and pretty high workload it lays upon contemporary graphics adapters, despite the fact that it is based on the same Quake3 gaming engine. Although you will have hard time trying to find similarities between Elite Force 2 and Quake3: the game is very beautiful, rich in high-quality textures and complex models of futuristic weapons. I am sure that all fans of sci-fi shooters will simply love this one.

We tested our graphics cards in a demo program, which we recorded ourselves in SoloMatch mode. Here are the results obtained:

As you see, the obtained numbers are far from what we usually see in Quake3, and we are not talking about hundreds of fps anymore. GeForce FX 5600 Ultra looks much better than FIC R96P here, although overclocking helps to almost improve the situation completely.

Serious Sam: The Second Encounter

Again GeForce FX outperforms FIC R96P. However, the overclocked RADEON 9600 PRO receives a second wind and manages to prove really great with enabled FSAA and anisotropic filtering.

Unreal Tournament 2003

This game belongs to a new generation, however, unfortunately FIC R96P falls behind its rival here as well. Overclocking helps to improve the situation a little bit, although our hero still fails to take the lead.

Splinter Cell

The situation in Splinter Cell, which is a real scourge for all contemporary graphics adapters, is completely different. Due to a powerful shader unit, FIC R96P completely defeats GeForce FX 5600 Ultra, which is taken down by its slower ALU’s. Overclocking allows RADEON 9600 PRO to achieve more or less acceptable gaming performance at least in 1024x768.

Performance: Synthetic Gaming Benchmarks

3DMark2001SE

As you see, the advantage here belongs mostly to GeForce FX 5600 Ultra. However, overclocking allows RADEON 9600 PRO to get ahead in many cases. The graphics adapter probably lacks fast graphics memory working at 750MHz and up.

3DMark03

3DMark2003 rearranges the situation. In heavy modes with enabled FSAA and anisotropic filtering GeForce FX 5600 Ultra manages to win due to higher working frequencies. However, in light modes FIC R96P on RADEON 9600 PRO takes the lead due to more advanced architecture. No doubt that the “pure performance” of the latter is much higher than that of its rival.

Codecult Codecreatures

In this pretty test with the whole bunch of complex graphics scenes the victory belongs to FIC R96P on ATI RADEON 9600 PRO, because the new official Detonator version has no more optimisations to ease GeForce FX’s life.

Conclusion

Well, we have just tested FIC R96P graphics card based on ATI RADEON 9600 PRO (RV350) graphics processor. What did our investigation show? And it showed that the architecture of this chip is much more fit for new generation games and applications, which actively use Pixel Shaders. As far as the performance dependence on the textures quality is concerned, it is not that high and can be neglected for the sake of higher image quality. Unlike GeForce FX 5600 Ultra, which slows down significantly when the Quality mode is enabled.

As for the FIC R96P graphics card, we really liked it a lot due to excellent 2D image quality and high overclocking potential. Of course, we should also give credit to silent cooling system doing its job at nominal frequencies well enough for the graphics chip to stay cool. As for overclocking, I would strongly recommend to get an additional cooler.

This graphics card will suit perfectly for those of you who are looking for good gaming performance in contemporary games and do not want to spend too much money on a high-end graphics solution. Unlike GeForce FX 5600 Ultra, this product will also do well in the upcoming future games, which will actively use shader operations. Moreover, FIC R96P is a good choice for work on large displays due to very good 2D quality. Unfortunately, unlike TYAN, FIC doesn’t have a modification of the same graphics card equipped with two DVI-I Outs that is why if you are a happy owner of two LCD panels, this graphics solution will not suit you.

Highs:

Lows: