Part 3: Performance in Game Tests

TABLE OF CONTENTS

1. Part 1 - Theory and Architecture
2. Part 2 — Practical examination
3. Video cards' features
4. Testbed configurations, benchmarks
5. Synthetic test results
6. Game test results (performance)

Let's proceed to the game performance analysis. It's no secret that many games have already reached the limits of CPU and system capacities on the whole, so we don't see the potential of top video cards even in AA 4x and AF 16x modes. That's why we introduced a new HQ (High Quality) mode, which means:

• ATI RADEON X1xxx: AA 6x, plus Adaptive AA, plus AF 16x High Quality
• NVIDIA GeForce 6xxx/7xxx: AA 8x, plus TAA, plus AF 16x.

A note to those indignant readers who say that it's impossible to compare differently implemented AA modes. We have a single criterion: maximum quality. ATI offers better AF, NVIDIA — AA. Trust me, you won't see any difference in quality, unless you scrutinize enlarged screenshots. That's why we actually can compare both modes (with some approximations, of course).

Test results: performance comparison

We used the following test applications:

• Splinter Cell Chaos Theory v.1.04 (Ubisoft) — DirectX 9.0, multitexturing, test settings — maximum, shaders 3.0 (for NVIDIA cards)/shaders 2.0 (for ATI cards); HDR OFF!
  
  
• Half-Life2 (Valve/Sierra) — DirectX 9.0, demo (ixbt01 . The tests were carried out with maximum quality, option -dxlevel 90, presets for video card types are removed from dxsupport.cfg.
  
  
• FarCry 1.33 (Crytek/UbiSoft), DirectX 9.0, multitexturing, demo from the Research level (-DEVMODE startup option), Very High test settings.
  
  
• DOOM III (id Software/Activision) — OpenGL, multitexturing, test settings — High Quality (ANIS8x), demo ixbt1 (33MB!). We have a sample batch file to start the game automatically with increased speed and reduced jerking (precaching) d3auto.rar. (DO NOT BE AFRAID of the black screen after the first menu, that's how it should be! It will last 5-10 seconds and then the demo should start)
  
  
• 3DMark05 1.20 (FutureMark) — DirectX 9.0, multitexturing, test settings — trilinear,
  
  
• 3DMark06 1.02 (FutureMark) — DirectX 9.0, multitexturing, test settings — trilinear,
  
  
• The Chronicles Of Riddick: Escape From Butcher Bay 1.10 (Starbreeze/Vivendi) — OpenGL, multitexturing, maximum texture quality, Shader 2.0, demo 44.

  I wish to thank Rinat Dosayev (AKA 4uckall) and Alexei Ostrovski (AKA Ducche), who have created a demo for this game. I also want to thank Alexei Berillo AKA Somebody Else for his help.
  
  

• F.E.A.R. v.1.02 (Multiplayer) (Monolith/Sierra) — DirectX 9.0, multitexturing, maximum test settings, Soft shadows disabled.
  
  
• Call Of Duty 2 DEMO (Ubisoft) — DirectX 9.0, multitexturing, test settings — maximum, shaders 2.0, tested in Benchemall, demo and a startup script, readme contains necessary instructions
  
  

Considering the capacity of the latest generation of accelerators, performance analysis should be performed only with the AA+AF load. Otherwise, it would be a competition of processors or system units rather than video cards. Sometimes in HQ mode only. Of course, in heavy applications, where powerful video cards demonstrate mediocre performance even without AA+AF, we shall analyze these modes as well.

We have also compared the X1900 XTX and the X1950 XTX, operating at the frequencies of the former card, in order to find out whether there were some changed made to the core structure.

Game tests that heavily load vertex shaders, mixed pixel shaders 1.1 and 2.0, active multitexturing.

FarCry, Research

Test results: FarCry Research

Tests in this game should be run in HQ mode only, because modern Hi-End accelerators offer decent gameplay in this mode. The X1950 XTX was utterly defeated not only by the dual-GPU 7950GX2, but also by its direct competitor (in price) — 7900GTX.

Increased memory bandwidth of the X1950 (in comparison with the X1900) added up to 10% of performance in this mode. In case of regular AA+AF — up to 12%. We should do justice to the new card from ATI - it takes up the lead under such a load (AA4x+AF16x in high resolutions).

Game tests that heavily load vertex shaders, pixel shaders 2.0, active multitexturing.

F.E.A.R.

Test results: F.E.A.R.

HQ mode in this game is slower, but it's still very fast in 1024x768. On the whole, the X1950XTX is also outperformed by its competitors. But in case of AA4x+AF16x, it goes on a par with the 7900 GTX.

Performance gains from the increased memory bandwidth in the x1950 in a simple AA+AF mode are not very high here - the game is more critical to the GPU power. But the HQ mode really welcomes the increased memory bandwidth, especially in high resolutions.

Splinter Cell Chaos Theory

Test results: SCCT

The situation changes a little here. The HQ mode helps the X1950 XTX card to catch up and even outperform the 7900 GTX. But the 7950 GX2 is still the first. The new card becomes a leader under normal AA+AF load in high resolutions. Having analyzed the above tests, we've come to the following conclusion - the higher a resolution, the more advantage is gained by the X1950 (or the less it's outperformed).

What concerns performance gains from memory bandwidth, everything is like in the previous test.

Call Of Duty2

Test results: CoD2

We can see in this test that it's useless to analyze the results in HQ mode, because the game is not fast enough even without it. In case of the usual AA+AF load, the X1950 XTX defeats the 7900 GTX, being outperformed by the 7950 GX2 again.

Increased memory bandwidth in the x1950 (compared to the x1900) provides 8% performance gain.

Half-Life2: ixbt01 demo

Test results: Half-Life2, ixbt01

Even though the x1950 demonstrates some performance gain from the increased memory bandwidth in HQ mode, the new product is still generally defeated. We speak of the HQ mode, because it's a sin to use any other mode in this game on powerful accelerators.

Game tests that heavily load pixel pipelines with texturing, active operations of the stencil buffer and shader units

DOOM III High mode

Test results: DOOM III

HQ mode has led the X1950 to defeat as well. In the usual AA+AF mode, it's on a par with the 7900GTX, but the 7950GX2 is still victorious.

Chronicles of Riddick, demo 44

Test results: Chronicles of Riddick, demo 44

An utter defeat. NVIDIA products are faster in every respect.

Synthetic tests that heavily load shader units

3DMark05: MARKS

Test results: 3DMark05 MARKS

That's synthetics. Everything is right according to the specifications and expectations. Yep, the X1950 has slightly outperformed the 7900 GTX and hasn't caught up with the dual-GPU 7950. It's only natural.

3DMark06: MARKS Shader Model 2.0

Test results: 3DMark06: MARKS Shader Model 2.0

Execution speed of simple shaders plays an important role in this test. But the new product is not very good at it (as we have already noted). The same concerns texturing performance. Hence a fiasco.

3DMark06: MARKS Shader Model 3.0

Test results: 3DMark06: MARKS Shader Model 3.0

And vice versa. The X1950 is good at SM3 like no other model. That's why the 7900GTX is defeated. Only the dual-chip 7950 did not let the card outperform itself.

Conclusions

ATI RADEON X1950 XTX is obviously an improved modification, revision, of the previous flagship. According to our tests, the product fairs on the level of the 7900 GTX, sometimes it gets close to the 7950. We can say that it's almost a leader among SINGLE-SLOT and SINGLE-GPU accelerators. The 7950 GX2 still remains a leader among SINGLE-SLOT solutions (that is those cards that take up only one PCI-E slot).

But! The announced price of $449 will encourage fans of the Canadian company! Its price being comparable to that for the 7900GTX (it may be even cheaper from the very beginning), the new product from ATI outperforms the former in a number of tests (the 7950GX2 is faster, but it's more expensive!), it offers excellent characteristics — price/performance/features, and provides the best 3D image quality. Besides, don't forget that it's the first video card with GDDR4 memory.

Another point - the X1950 XTX is now equipped with a very efficient and quiet cooler. That's good news for potential buyers, as the noise generated by powerful system units has become one of the major problems. It's even close to a disaster. We can only welcome the new product in this respect. But it should be noted that power consumption of the card remains on the same level, the card gets very hot. You should keep it in mind. Have a look at the environment temperature from the monitoring data above. You'll get the picture.

Thus, if the retail prices come up to the recommendations, the X1950 XTX may become a very good bargain. But considering that the cards will appear in stores only by mid September (most likely later), we cannot forecast a price drop for the competing products from NVIDIA.

I'll stress the important point once again: although these cards have become much quieter, they still dissipate much heat, so PC cases must be ventilated well. You will also need at least a 450-500 Watt power supply unit that can output up to 18-20A at the PCI-E end. A less powerful PSU may suffice sometimes, but it must offer honestly good characteristics.

You can find more detailed comparisons of various video cards in our 3Digest.

for the excellent design of the card, its cooling system in particular.

ATI RADEON X1300-1600-1800-1900 Reference

NVIDIA GeForce 7300-7600-7800-7900 Reference

PSU for the testbed was kindly provided by HIPER

Memory for the testbeds is kindly provided by