There are scheduled articles and unscheduled ones. But sometimes seemingly non-processor materials for internal use spring up funny facts, which we just cannot send to our archive for internal use and forget. So let's not call this "an article". This material is sort of extended news, which appeared due to a funny fact we accidentally came across.

It all started when we were asked a banal question: "Isn't the video card for testing CPU performance too slow?" The logical answer was just as simple as the question: "What do you suggest as a replacement?" The suggested model was an impressive-looking video card from Sapphire based on Radeon X1900XT. Having studied Andrey Vorobiev's section, we decided that the suggestion should be at least given a thorough consideration (even though the number of processors, tested by the new procedure, is getting close to a dozen). But test results of the aspirant to the title of a new standard video card in our CPU performance test procedure have shocked us... Our material will be devoted to this very issue. Perhaps, it may seem incomplete to 3D fans. But we cannot help it, we are not experts in this field, so our opinions may be based only on test results and hackneyed logix

Hardware and Software

Testbed configurations

• Video cards: GeForce 7800GTX 256 MB (Gigabyte), ATI Radeon X1900XT (Sapphire)
• Memory: 2 GB (2 modules)
• HDD: Samsung SP1614C (SATA)
• Coolers: standard models that come shipped with processors
• PSU: CHIEFTEC GPS-550AB A

* — "2x..." means "per each core"
** — in AMD processors it's frequency of the memory controller bus
*** — it's measured differently in Intel and AMD processors, so a direct comparison is not correct

Software

1. Windows XP Professional x64 Edition SP1.
2. 3ds max 7.0
3. Maya 6.5
4. Pro/ENGINEER Wildfire 2.0
5. SolidWorks 2005
6. F.E.A.R. 1.3
7. Half-Life 2
8. Unreal Tournament 2004 build 3339
9. Quake 4 Point Release 1.1

Drivers

1. NVIDIA ForceWare 91.31
2. ATI CATALYST 6.6
3. NVIDIA nForce SMBus Driver 4.50
4. Intel INF Update 7.2.2.1006

Testing

In this case we decided to depart from the concept of "a total score in a subgroup of tests", which is used in our new test procedure by default, and to focus on those results that really differ for different video cards. This approach seems to us the most adequate in the context of these tests: indeed, we changed nothing but a video card in the testbed.

3ds max 7.0

The first ear-splitting "hook", missed by ATI Radeon X1900XT... When a faster (according to its specifications) card is outperformed by a lower one by 32-52%, all you want is to rerun all the tests: "it just cannot be!" OK, we reran them. The fact remains, it's still outperformed. The same percents again.

Maya 6.5

The first shock is over, so SPEC for Maya results just convince us that the defeat of the Radeon X1900XT GeForce 7800GTX is not an absurd accident. But it was still too early to speak of a regularity...

Pro/ENGINEER Wildfire 2.0

Now we can talk regularities. The first one is API. Professional programs practically don't use Direct3D, preferring OpenGL to it — the reason is its multiplatform nature (Direct3D can be used by 3ds max, but we run it in OpenGL mode). The second one — we haven't dealt with any programs, except professional — games will be reviewed later...

SolidWorks 2005

David again defeated Goliath. As we have seen it on the fifth diagram running, there are more than enough grounds for the sentence. It would be naive to assume that all hardware advantages of the X1900XT are false (you will soon find out that it's not true). So the problem must be in something else.

F.E.A.R.

Let's proceed to games — everything alters radically: ATI Radeon X1900XT takes up the lead. As appropriate for a more powerful and modern video card. We have finally seen that the X1900XT is worthy of this title :).

Half-Life 2

ATI's video card wins in this game, its victory is even more pronounced on the Intel platform (percentagewise).

Quake 4

That's an only gaming test so far, where we can see two striking differences from the previous results. Firstly, it's the first game, where a relatively weak video card with an NVIDIA chip managed to defeat ATI Radeon X1900XT (on the Intel platform). Secondly, it's the only test in our test procedure, where platform preferences are divided: ATI is victorious on the AMD platform. The second fact can hardly be explained — it's an exceptional situation, and we are short of statistics. We can try to explain the first case - John Carmack's engines have always focused on the classic OpenGL. Theoretically, it might have led to preferences like in professional programs.

Unreal Tournament 2004

An old game, an old engine — NVIDIA is victorious, even though the break away is not significant. And besides GeForce 7800GTX is considered a much weaker chip than Radeon X1900XT.

Summary diagram

The situation is more or less clear: ATI Radeon X1900XT outperforms NVIDIA GeForce 7800GTX only in two games (both are based on API Direct3D) and makes a draw in one OpenGL-based game (victorious on the AMD platform and defeated on the Intel platform). The X1900XT is defeated in absolutely all professional applications, API — always OpenGL. Conclusions seem evident.

Conclusion

As it regularly (though not very often) happens in our practice, totally harmless and seemingly predictable "internal" testing has sprung a very strange surprise: a weaker video card defeats on points one of 3D monsters. We should ask a classic question: "Whose fault is it?" Having swept aside the fantastic assumption that the Radeon X1900XT chip actually contains Rage 128 Pro and that ATI engineers draw all specifications and block diagrams of the new products with felt-tip pens :), we can only assume that the powerful chip was let down by the drivers. What happened to CATALYST drivers? We came up with three possibilities:

1. ATI drivers are weakly optimized for professional applications that use OpenGL API. That's because the company focused on gamers and does not attend to the sector of professional applications seriously.
2. ATI intentionally leaves out the code optimized for professional applications in the "amateurish" CATALYST drivers, which are written for the RADEON series for home users (this code is included into drivers for the professional series of 3D accelerators FireGL). Here is the argument for it - unlike NVIDIA, ATI does not offer a unified driver for consumer and professional series.
3. A combo of the first and the second possibilities: due to some peculiarities of ATI graphic chips, their drivers can be well-optimized either for professional OpenGL, or for games; it's impossible or very difficult to switch between these modes. That's why the division, which existence we assumed in Possibility 2, is actually forced.

Which possibility of the three is the most likely? Strange as it may seem, we are inclined to stake on the first one. Why? Firstly, it agrees well with Occam's Razor, being the simplest one. Secondly, this assumption is partially backed up by the test results in the latest version of the most famous OpenGL engine (Quake 4 / id Software). And thirdly, problems of ATI drivers even with OpenGL games go back in history, known well to those who are interested in the situation with 3D accelerators on the x86 platform. Even if Possibility 2 is correct, we should still drop the handkerchief to less "greedy" NVIDIA that equips even its consumer video cards with better drivers (from the professional point of view). If Possibility 3 is true (though it's the most fantastic of all), NVIDIA is at advantage again, as its chips turn out to be all-purpose.

What concerns our conclusions drawn from the in-house tests, they are obvious: the fact that the weaker NVIDIA GeForce 7800GTX managed to outperform the ATI Radeon X1900XT by 14-16% in our assortment of programs (when performance depends on a video card) indicates that we chose the right video card for our CPU performance tests.

P.S. (for gamers only): in fact, there are much more of you than those who work with 3ds max, Maya, or Pro/ENGINEER. That's obvious. And Direct3D is becoming a standard for Windows games, slow and sure. So you shouldn't be sad: more pipelines, better Direct3D support, and other features — they hardly stopped being important. We have just drawn attention of a small group of users to some issues that may be of interest to them...

Write a comment below. No registration needed!