by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
10/07/2005 | 06:19 PM
After taking a look at the RADEON X1600 XT product earlier this week in our architectural preview (for details see our article called ATI RADEON X1000: Brand-New Graphics Architecture from ATI Explored ) of the RADEON X1000 we were impressed by its performance in pixel shader 3.0 branching and some other tests, but were left curious about real-world performance of a graphics processing unit (GPU) that has 12 pixel processors and only for 4 texture units.
The RADEON X1600-series is expected to substitute RADEON X700 family of graphics chips as well as to replace the RADEON X800 GT and possibly the RADEON X800 XL (or GTO2) products from performance mainstream segment. The obvious advantage the RADEON X1600 has over predecessors is support for Shader Model 3.0, HDR and some other technologies. The probably disadvantage is 128-bit memory bus as well as pretty low texel-rate. We will find out the outcome later in the article, but now let’s summarize everything we know about the RADEON X1600 also known as RV530, based on the results in synthetic investigation.
We used the following test system for our today’s test session with the new graphics accelerator family from ATI:
ATI and NVIDIA drivers were configured as follows:
NVIDIA ForceWare 78.01:
The following games and applications were our benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
Being pretty GPU-bound game, Battlefield 2 demonstrates pretty interesting scores.
It appears that the RADEON X1600 XT manages to outperform anything except the GeForce 6800 GT and the RADEON X700 XL. At 1600x1200 resolution, however, neither the RADEON X1600 XT, nor the model X800 XL provide sufficient performance.
Performance drop with active 4x full-scene antialiasing as well as 16x anisotropic filtering does not drop performance significantly, which may mean that 128-bit memory bus does not limit performance of the RADEON X1600 XT much (also due to the fact that ATI has improved its hidden surface removal algorithms). The model X1600 XT manages to leave behind the GeForce 6800, but cannot outperform the GeForce 6800 GT.
Being an OpenGL game which takes advantage of NVIDIA’s UltraShadow technology, The Chronicles of Riddick is not favouring ATI’s RADEON X1600 XT. The only card that the model X1600 XT can outperform is ATI’s RADEON X700 PRO and its results are slightly behind those of the GeForce 6600 GT.
Similar situation remains in the eye-candy mode: the RADEON X1600 XT scores about the same framerate as the 1-year old GeForce 6600 GT that now costs about $149.
Doom III is a yet another game that does not favor ATI’s graphics accelerators, hence, the results are pretty obvious.
RADEON X1600 XT only manages to leave behind the model X700 PRO, but cannot really compete even with the RADEON X800 XL, not talking about GeForce 6600 GT or more powerful products. For users of the RADEON X1600 XT only 1024x768 resolution will offer enough speed.
The situation does not change much with eye-candy activation. The RADEON X1600 XT does not bring the real performance to Doom III, at least in case of maximum quality settings.
Another Doom III demo shows higher framerate, but the same positions for graphics cards: the GeForce 6800 GT is the best card to play Doom III at $299 and below, whereas the GeForce 6800 provides enough speed even for 1600x1200 at the price of below $199.
RADEON X1600 XT, which manufacturer suggested retail price is $249 for 256MB version, cannot outperform even the RADEON X800 XL, which now costs about $249 as well.
The situation in eye-candy mode looks similar, but now the RADEON X1600 XT succeeds in achieving framerate of the GeForce 6600 GT.
Theoretically, RADEON X1600 XT has higher fillrate compared to the RADEON X800 XL, however, in reality it cannot match the latter in high resolutions. This is probably an outcome of only 4 texture mapping units (TMUs) that are present on the model X1600 XT. While framerate is generally not bad even with 1600x1200 resolution, this an alarming sign.
Despite of supposedly sufficient memory bandwidth, the RADEON X1600 XT cannot showcase itself well with FSAA 4x and anisotropic filtering 16x turned on. It offers similar performance with the GeForce 6800 and gives enough speed to play in 1024x768, but it cannot beat the GeForce 6800 GT.
Demo recorded on the Research level utilizes pixel shaders 3.0 in order to optimize performance. ATI’s RADEON X1600 XT scores well, but the only product it can match is the GeForce 6800 that works at significantly lower core clock-speed and provides similar memory bandwidth.
With FSAA and anisotropic filtering the RADEON X1600 XT continues to stay inline with the GeForce 6800 and cannot offer similar performance to the RADEON X800 XL.
Due to the nature of the Canals demo, the insufficient number of TMUs should not limit performance of the RADEON X1600 XT.
The novelty from ATI manages to achieve the same score as the GeForce 6800, but can beat neither the RADEON X800 XL, nor the GeForce 6800 GT.
Even though performance offered by the RADEON X1600 XT should be enough for playing with all the eye-candy in 1280x1024, but from competitive point of view the part is much less preferable than the RADEON X800 XL.
The street fighting scenes in Half-Life 2 require more processing power than the Canals levels do. Our demo record also includes numerous special effects, so the performance is generally lower here.
The RADEON X1600 XT product manages to perform pretty well in this test and get close to the GeForce 6800 GT leaving the rest behind.
With anisotropic filtering and antialiasing activated, the RADEON X1600 XT continues to perform pretty well.
From the technical point of view, Painkiller is a not very complex game, although its graphics quality is quite high, and the gaming experience is very exciting. This game doesn’t require much for the PC graphics subsystem that is why all testing participants perform almost equally fast in low resolutions. In high resolutions RADEON X1600 XT falls slightly behind GeForce 6800/6600 GT. In eye candy mode it is almost as fast as GeForce 6800, and sometimes it is even faster than the latter.
FEAR is the most demanding shooter used by our lab and it heavily uses advanced pixel shaders for special effect.
RADEON X1600 XT manages to outperform the GeForce 6800 and 6600 GT behind, but cannot achieve the same level as the RADEON X800 XL. Given that pixel shader performance of the X1600 XT should be higher than that of the X800 XL, we should again complain about the lack of sufficient amount of TMUs.
Pariah gaming engine is based on the modified Unreal Engine using different lighting and blurring effects as well as bump-mapping. It looks like there are no complex pixel shaders in this game, so RADEON X1600 XT doesn’t have much room for showing off. Nevertheless, in 1024x768 it offers pretty high gaming performance, although in higher resolutions it yields even to GeForce 6600 GT because of the fewer TMUs. We believe that 4 texturing units are evidently not enough for 12 pixel processors and ATI should have left RV530 at least 8 of those in order to avoid frustrating bottlenecks, which wouldn’t let the newcomer reveal its real performance potential.
Project: Snowblind uses a lot of special effects based on pixel shaders 2.0. The new-generation RADEON X1000 family performs these shaders relatively fast although not as efficiently as we wish it would. At least, it is only the high working frequencies that help RADEON come out on top in many cases. In our case RADEON X1600 XT is powerful enough to outpace GeForce 6800 or go neck and neck with it. However, it fails to catch up with the RADEON X800 XL and GeForce 6800 GT. When we enabled FSAA RADEON X1600 XT slows down in higher resolutions and by 1600x1200 it falls behind GeForce 6600 GT.
Well, RADEON X1600 XT copes pretty Ok with the vertex shaders, however, the fillrate still leaves much to be desired. As a result, it only manages to deliver comparable results in 1024x768 – in higher resolutions it gets defeated by all the rivals except RADEON X700 PRO.
In eye candy mode even highly efficient memory controller cannot save the newcomer from defeat.
We see something very similar in the Metallurgy level, although in 1280x1024 with enabled anti-aliasing RADEON X1600 XT manages to catch up with GeForce 6600 GT.
As we have already pointed out in our previous articles this game is very rich in textures and relatively simple pixel shaders. RADEON X1600 XT works pretty well with pixel shaders however the scenes with multiple textures involved are a real pain for it, because it has only 4 TMUs onboard. As a result, the newcomer runs only as fast as RADEON X700 PRO and GeForce 6600 GT in Prince of Persia: Warrior Within . By the way, the GeForce 6600 GT features only 4 rasterization units (ROP).
In Chaos Theory RADEON X1600 XT performs very successfully, although it still cannot really compete with 16-pipeline solutions, which boast much higher computational power. Besides, RADEON X800 XL doesn’t support Shader Model 3.0, which partially explains its relatively high scores. In general, the new RADEON X1600 XT runs at the performance level of GeForce 6800 in pure speed as well as eye candy mode. The texturing speed doesn’t matter that much in this game, and the pixel shader performance of our hero is quite high, as we have already seen in theoretical benchmarks, especially if we compare its results against RADEON X700 PRO. I would also like to stress that ATI’s new solution boasts much higher minimum performance in all resolutions than any of the NVIDIA competitors. It should be the high clock frequency that does the trick.
In Colin Mc Rae Rally05 the texturing workload is not very high either. So the major factor influencing the graphics cards performance here is the number of pixel processors and their working frequency. Of course, the number of texturing units also tells on the scores. RADEON X1600 XT has only 4 texturing units like that against 12 by GeForce 6800, but its high GPU frequency makes up for the lack of TMUs that is why it goes almost neck and neck with the NVIDIA solution in all resolutions except 1600x1200. Here the ATI RADEON X1600 XT yields its positions to competitor. Unlike Splinter Cell: Chaos Theory the highest minimum framerate here belongs to GeForce 6800.
When we enable FSAA, RADEON X1600 secures its positions and even outperforms GeForce 6800 GT in 1280x1024 a little bit. In 1600x1200, however, it steps back to the level of GeForce 6800 due to the influence of the memory subsystem. Even though the memory bus of RADEON X1600 XT is half as wide as the memory bus of GeForce 6800, its practical bandwidth is almost the same: 22.08GB/s against 22.4GB/s.
RADEON X1600 XT performs surprisingly well here even despite the inefficient OpenGL driver and too few TMUs. At least it is up to 50% faster than RADEON X800 XL in high resolutions in pure speed mode. Maybe that’s because Pacific Fighters supports Shader Model 3.0 and ATI’s new solution outperforms the previous generation ones, which support only Shader Model 2.0. In 1600x1200 RADEON X1600 XT comes up real close to GeForce 6600 GT thanks to its new memory controller.
In eye candy mode with full-screen anti-aliasing RADEON X800 XL with 32GB/s memory bandwidth starts to surpass RADEON X1600 XT with its 22.08GB/s. Unfortunately, both ATI solutions can ensure minimum comfortable gaming performance with enabled FSAA only in 1024x768, while GeForce 6800 allows you to play just fine in 1280x1024. Again, they should blame the inefficient ATI OpenGL driver for that.
As the resolution increases the performance of our RADEON X1600 XT drops slower than that of other testing participants, so that in 1600x1200 it overtakes GeForce 6800 GT still offering bigger performance margin. In fact fillrate and memory subsystem performance are the key factors determining the graphics card performance in flight simulators, therefore it is not surprising at all that 16-pipeline RADEON X800 XL and GeForce 6800 GT get to the forefront in game modes with enabled FSAA and anisotropic filtering. RADEON X1600 XT easily demonstrates the performance level of GeForce 6800 due to high GPU frequency.
Due to high pixel shader performance, RADEON X1600 XT proves quite fast in real-time strategies. At least it doesn’t fall that far behind RADEON X800 XL in Perimeter game, in both: pure speed and eye candy modes. I think it could have performed even better than that if it only had more TMUs. I am strongly confident that by limiting the number of texturing units to only 4 ATI created a potential serious bottleneck for its RV530, which is in our case connected with the very low fillrate. High GPU frequency can partially make up for that, but it is not enough in some cases. Note, however, that the Perimeter results were taken manually, so they may be not absolutely exact.
Besides rather complex geometry (2,000 polygons per model, over 100 models per frame), the Dawn of War engine uses fixed TnL functions, stencil shadows and high-resolution textures. As a result, GeForce 6/7 based graphics cards suit ideally for this game because they provide high fillrate, emulate TnL efficiently and process stencil shadows very fast. Unfortunately, RADEON X1600 XT cannot boast any of those features, even though it can process twice as many Z values per clock. In pure speed mode its higher working frequencies and 5 vertex processors help it to compensate the bottlenecks we have just mentioned, however, even in this case its minimum performance is still extremely low: 16fps in 1024x768. For comparison: the minimum performance of GeForce 6800 doesn’t drop below 25fps even in 1600x1200.
As for the eye candy mode with enabled FSAA and anisotropic filtering, the situation turns out even worse for RADEON X1600 XT: it outperforms only the good old buddies RADEON X700 PRO and GeForce 6600 GT. And forget about comfortable gaming experience here.
Most shaders in Aquamark3 are pixel and vertex shaders 1.1: the benchmark has 220 of them. There are very few 2.0 shaders: 3 vertex shaders and 4 pixel shaders, and the number of instructions per shader doesn’t ever go beyond 50. RADEON X1600 XT can show its best in this test: the scenes here feature quite complex geometry, and the new ATI solution features 5 enhanced vertex processors working at 590MHz. Moreover, it can remove hidden surfaces very efficiently, which is very important for a test like Aquamark3 with high overdraw coefficient.
Unfortunately, low fillrate spoils the whole show. RADEON X1600 XT competed somehow with GeForce 6800 in lower resolutions, however in higher resolutions all it can do is try to scare away GeForce 6600 GT.
Here we can observe a pretty funny situation: all testing participants except the leader - RADEON X800 XL, and the loser – GeForce 6600 GT, perform almost equally fast. I dare suppose that it is geometry processing speed that has certain influence on the results in this test: GeForce 600 GT works with geometry slower than everyone else, because it has only 3 vertex processors. If we take this criterion into account, RADEON X1600 XT shouldn’t be falling that far behind RADEON X800 XL. However, as we have already pointed out in our yesterday’s article called ATI RADEON X1800 XT and XL Performance: Crushing NVIDIA's 7800?, the texturing workload in FF XI Official Benchmark 3 is quite high, which could easily prevent the newcomer from catching up with the leader.
Even though 3DMark03 benchmarking suite doesn’t really like RADEON X800 and RADEON X1000 architectures, our today’s hero managed to show some worthy results in the total score chart: over 9,000 points. Its most evident competitor, GeForce 6800, turned out 341 points behind. Now let’s take a closer look at each gaming benchmark separately to see what determined the newcomer’s victory here.
The failure in Game 1 test is not surprising to me: it uses fixed TnL functions, in particular point sprites, and requires high fillrate. Here I would like to make some comments: although the background is created with large polygons covered with a single texture, all aircrafts are rendered with 4 textures. We already know that RADEON family has some problems with TnL emulation, which emerge from CATALYST driver issues. Therefore, it is quite logical that RADEON X1600 XT turned out the last but one in 3DMark03 Game 1 test. The only one slower than our today’s hero is RADEON X700 PRO suffering from cut-down texture caches. However, it still outperforms RADEON X1600 XT in 1600x1200 with enabled full-screen anti-aliasing.
The newcomer feels much better in Game 2 test and outperforms GeForce 6800 in 1024x768 (which is the default resolution in 3DMark03). As the resolution increases, RADEON X1600 XT and GeForce 6800 level out and the situation remains like that until the end of the test. This benchmark is very sensitive to the GPU’s ability to efficiently process the Z-buffer, and RADEON X1600 XT is pretty good at it. Firstly, it features 8 Z-Compare units and hence can process twice as many Z values per clock cycle. And secondly, RADEON X1000 architecture can compress Z-buffer data as 8:1. Of course, high GPU frequency also contributed to the success of RADEON X1600 XT. All these factors together allowed the new ATI solution to come out on top in Game 2 test.
Game 3 test is remarkable for highly complex geometry: there are almost twice as many polygons per frame compared with the previous test. All in all, the performance picture remains the same, with the only difference that in lower resolutions RADEON X1600 XT doesn’t get ahead of GeForce 6800, and even falls behind it with enabled FSAA. In general, we can state that there is evident parity between the ATI solutions and NVIDIA solutions featuring 12 pixel processors.
RADEON X1600 XT looks much better in Game 4 test, which takes real advantage of Shader Model 2.0. This is the test where pixel and vertex processor performance matters a lot, and ATI has it all under control due to new architectural advantages and 590MHz working frequency. As a result, RADEON X1600 XT is almost as fast as GeForce 6800 GT and even faster than the rival in low resolutions. In eye candy mode with enabled FSAA and anisotropic filtering our hero gets limited by the 128-bit memory bus: RADEON X800 XL and GeForce 6800 GT feature 32GB/s memory bus bandwidth, while RADEON X1600 XT can boast only 22.08GB/s.
All in all, the results of individual benchmarks from the 3DMaro03 testing suite revealed the advantages of the RADEON X1600 XT that determined its victory in the total scores chart. However, you shouldn’t forget that in higher resolutions than the 3DMark’s default 1024x768, its performance may be lower.
As we have already said, 3DMark05 is a much more up-to-date test than 3DMark03, because it takes good advantage of pixel and vertex shaders. RADEON X1000 architecture was developed specifically for this type of processing so no wonder that RADEON X1600 XT is the winner here. Even though RADEON X800 XL features 16 fully-fledged pixel pipelines, the new-generation solution outperforms it. What contributes to the triumph of the ATI RADEON X1600 XT in 3DMark05?
In Game 1 benchmark RADEON X1600 XT works somewhat faster than RADEON X800 XL and GeForce 6800 GT. This test generates mostly geometrical workload that is why 5 new-generation vertex processors working at 590MHz are exactly what they need here. Unfortunately, we didn’t compare the performance of GeForce 6800 and RADEON X1600 XT in eye candy mode in resolutions above 1024x768, because the NVIDIA solution is equipped with only 128MB of onboard graphics memory, while you need 256MB of memory to be able to use FSAA in high resolutions in 3DMark05.
Game 2 test also loads the vertex processors quite heavily, although here they are mostly used to create a lot of dynamically generated vegetation. There are also dynamic shadows using 2048x2048 depth maps and procedural light dissipation. RADEON X1600 XT doesn’t yield to 16-pipeline rivals and predecessors. The only exception is again the eye candy mode where our hero lacks memory subsystem performance, because this test seems to be loading memory more in this test than in the previous one.
Game 3 test exhausts the pixel processors almost completely, because one of the major materials imitating rocky surfaces is created with the help of a very long pixel shader pushing Shader Model 2.0 to the limit of its potential. And the water surface also created with complex math1ematical calculations is rendered in 6 passes. On the one hand, pixel processors of RADEON X1600 XT work pretty efficiently, and on the other hand, the memory subsystem is loaded quite heavily that is why the new ATI solution outperforms RADEON X800 XL and GeForce 6800 GT only in 1024x768, and then the previous-generation ATI RADEON takes the lead. When FSAA and anisotropic filtering are both enabled, both RADEON solutions perform equally fast up until 1600x1200 where RADEON X1600 XT again starts to suffer from narrow 128-bit memory bus.
In all three benchmarks RADEON X1600 XT performed either slightly faster or slightly lower than RADEON X800 XL, so that the total difference ins cores made only 117 points in favor of the new ATI solution.
The RADEON X1600 XT 256MB proved to be nothing else but a pretty fine mainstream graphics card that sports some capabilities for the future and packs pretty serious performance for today. But the new mainstream graphics solution from ATI is definitely not the ideal one.
In current games, except some OpenGL titles, the RADEON X1600 XT generally manages to offer similar or higher performance compared to the GeForce 6800 product. This means that the product is supposed to outperform the GeForce 6600 GT, probably the best performance-mainstream offering of the recent years. On the one hand it seems that the RADEON X1600 XT does redefine the market of powerful, but not expensive graphics cards, but on the other hand it has some serious competitor – the GeForce 6800 GT – from the high-end, which is officially priced only $50 above, but can be obtained for about $249, the MSRP of the RADEON X1600 XT 256MB.
As our test results show, the GeForce 6800 GT and the RADEON X800 XL are much faster than the RADEON X1600 XT 256MB, yet, their prices have been dropping for the past several months now and they are likely to cost approximately the same amount of money compared to the RADEON X1600 XT parts. This means that graphics cards makers will have to decrease pricing of the RADEON X1600 XT in order to make it more appealing to customers.
At this time there are numerous advantages the RADEON X1600 XT has over the predecessors and some of the rivals: high Shader Model 2.0 performance, a very efficient Shader Model 3.0 implementation, support for HDR, Avivo video engine and hardware H.264 decoding support, higher-quality anisotropic filtering as well as relatively low power consumption. The obvious disadvantage of the part is a small number of TMUs, which does limit performance in a number of games.
While everything said above seems to be correct just now, the RADEON X1600-series is only expected to be shipped for revenue starting November 30, 2005, which means that some things may be corrected, for instance, ATI may improve its drivers, while NVIDIA may once again decrease official pricing on the GeForce 6800-series. All-in-all, the final word is yet to be said.