CONTENTS

• 1. Video card's features
• 2. Testbed configuration, test tools, 2D quality
• 3. Test results: Quake3 ARENA
• 4. Test results: Serious Sam: The Second Encounter
• 5. Test results: Return to Castle Wolfenstein
• 6. Test results: Code Creatures DEMO
• 7. Test results: Unreal Tournament 2003
• 8. Test results: Unreal II: The Awakening
• 9. Test results: RightMark 3D
• 10. Conclusion

Since production cards based on NVIDIA's fastest processor GeForce FX 5900 Ultra have only started penetrating into the market, we will focus on a bit less powerful cards based on the FX 5900.

We are well aware that in the game tests the FX 5900 has equal scores with the ATI RADEON 9800 PRO. Also, we know that both ATI and NVIDIA succeed in making cheats in both FutureMark packets: 3DMark2001 and 3DMark03. In the list below all these issues are discussed:

Theoretical materials and reviews of video cards which concern functional properties of the GPU NVIDIA GeForce FX

• Analysis of the architecture of NVIDIA NV30 (GeForce FX)
• NVIDIA GeForce FX 5800 Ultra (NV30) - single-page review
• NVIDIA GeForce FX 5800 Ultra (NV30) - multi-page review
• NVIDIA GeForce FX 5600 Ultra (NV31) and GeForce FX 5200 Ultra (NV34) - single-page review
• NVIDIA GeForce FX 5600 Ultra (NV31) and GeForce FX 5200 Ultra (NV34) - multi-page review
• ASUS V9900 Ultra on NVIDIA GeForce FX 5800 Ultra - AA and anisotropy quality
• Gainward FX Powerpack Ultra/1000 Golden Sample and Gainward FX Powerpack Pro/660 TV/DVI on NVIDIA GeForce FX 5800 Ultra and 5200 - scaling (performance vs. CPU clock speed) of GeForce FX 5800 Ultra, performance of GeForce FX 5200
• Leadtek WinFast A300 Ultra MyVIVO on NVIDIA GeForce FX 5800 Ultra - performance of GeForce FX 5800 Ultra vs. CPU clock speed in heavy modes with AA and anisotropy enabled
• MSI FX5800 Ultra-TD8X on NVIDIA GeForce FX 5800 Ultra
• Albatron, Chaintech, Gainward, InnoVision, Leadtek, Palit and Prolink video cards on NVIDIA GeForce FX 5200
• ASUSTeK NVIDIA GeForce FX 5200/5600 video cards
• MSI FX5600-VTDR128 (MS-8912) card on NVIDIA GeForce FX 5600
• Albatron, Leadtek and MSI video cards on the NVIDIA GeForce FX 5200 Ultra
• Prolink PixelView GeForce FX 5600 256MB Golden Limited on NVIDIA GeForce FX 5600
• Gainward FX PowerPack Ultra/760 XP Golden Sample on NVIDIA GeForce FX 5600 Ultra (new revision)
• Leadtek WinFast A310 Ultra MyVIVO on NVIDIA GeForce FX 5600 Ultra (old revision)
• ATI vs NVIDIA: where are fair duels? or Dishonest Treatment of 3DMark
• MSI FX5900-VTD128 ha áaçe NVIDIA GeForce FX 5900 - more on dishonest treatment of 3DMark (fruits of collaboration of NVIDIA and FutureMark in the light of the "peaceful" agreement)
• Albatron Gigi GeForce FX 5600EQ 256MB, Albatron Gigi GeForce FX 5600P Turbo 128MB, Albatron Gigi GeForce FX 5600 Ultra 128MB (350MHz version), Albatron Gigi GeForce FX 5800 128MB ha áaçe NVIDIA GeForce FX 5600/5800

The recent article by Aleksei Nikolaichuk and theMSI FX 5900 video card review showed that the 3DMark doesn't reflect anymore the real cards' performance, that is why we will hardly use one of these packets in future.

Today, beside examination of the Gainward's card and its features we will look once again at how NVIDIA drags its baby up the ladder.

First of all, a bit of the history of events. A year ago ATI unexpectedly released its new product RADEON 9700 PRO. It was so sudden that even the people close to the Canadian company was taken aback. The fans and PR agencies of NVIDIA poured out sceptical comments about the overpriced R300 and a too low percentage of valid so huge and complicated dies. They also pointed out that that was an unprofitable deal, and boasted the fact that NVIDIA was working on the 0.13 micron process and was going to release a killer of the R300 which would have lower power consumption and lower cost.

The history showed that the PR agencies of the NVIDIA's camp had to dodge for a long time and look for new reasons to explain the delay of the NV30. Finally, in autumn when the RADEON 9700 PRO reached the peak of sales, NVIDIA brought out its GeForce FX 5800 Ultra.. but in very small quantities.

Why just few cards? The processor turned out to be too expensive, and NVIDIA's president has to look for reasons why the NV30 wasn't served by Xmas. It was caused not only by the problems with the processor and 0.13 micron technology (which are not done away with yet), but also by the desire to make this product look better than the ATI R300 due to the DDR-II memory. That was a fatal mistake. The PCB coupled with such memory was so expensive and complicated that a 256bit memory bus couldn't be realized, and NVIDIA had to look for third firms to place orders for such cards with. NVIDIA's partners were left out of work: the processors were supplied by NVIDIA, PCBs came from NVIDIA, the memory were bought by NVIDIA and then supplied by it as well. The GeForce FX 5800/5800 Ultra released were made in actually the same place and were copies. Poor vendors had no choice but stick their logos and pack them in boxes. The cost of such construction (with the FlowFX cooler for deaf) was very high, and profitability of the NV30 production was very low. That is why NVIDIA had to limit production of these processors by 100,000 pieces.

These cards hit the stores only in April, but the prices didn't justify their capabilities at all. Besides, by that time ATI had a new 3D king RADEON 9800 PRO released. Today it's no problem to find FX 5800 based cards, simply because nobody buys them. Why would one want such a card if in November 2002 NVIDIA's President played a whole performance attracting the ghosts of Voodoo and 3dfx to represent the NV30 in a favorable light, and then in April 2003 he called the NV30 a failure and even a mistake. Certainly, all the PR trumpets changed the tune, forgot about the FX 5800 and started heralding a soon release of the NV35.

The new processor was released in May. You wonder why so soon? But just count the time gap: November 15 to May 15 (approximately) makes half a year. Well, NVIDIA was cunning enough not to deceive its shareholders: the semi-annual cycle wasn't broken! However, the PR agencies held back the fact that the cards reached the market not after the chip release but only in April. How could they sell the NV30 if it became outdated by the beginning of sales and if a new product was promised in a month? It wasn't the first time NVIDIA let down its partners.

Well, some loopholes were closed in the GeForce FX 5900 (the memory throughput was increased, the heat generation was lower and the cooling system was less complicated, though the core clock was also brought down to 50 MHz). The ALU was optimized as well. But the worst problem - a low shaders performance - remained untouched. Although 32-bit floating-point operations could be executed at the same speed as 16-bit ones, did it satisfy anyone? All tests that loaded the GPU with shader technologies proved that the NV35 touted as the fastest 3D accelerator fell behind ATI's RADEON 9800 PRO. We have to admit that the GeForce FX 5900 (even the Ultra version) has a lower shader performance than the Canadian.

That fact gave NVIDIA no rest. What benchmark could reveal all the loopholes of the FX 5900? It's clear that it was the 3DMark03. First of all, the packet itself was oriented towards shaders 1.4 in most subtests (which was NVIDIA's Achilles' heel since the times of the GeForce3 Ti), - it immediately puts all ATI's latest cards into a higher position; secondly, even NVIDIA's DX9 accelerators execute shaders 1.4 by emulating them with shaders 2.0. That is why 3DMark03 is the Enemy Number One for NVIDIA. That is why the company used all the means to raise the FX speed. That resulted in the well known scandal that ended up with the defeat of... FutureMark, when all cheats revealed and then disabled in the patch 3.30 were announced to be optimizations. NVIDIA became one of the privileged beta testers again and released new beta versions of its drivers where the speed in the 3DMark03 rose again, and it was even higher as compared to that before the patch 3.30. All the facts are well described in this article.

So, NVIDIA used cheats in the popular benchmark to bring performance of the FX 5900 to the level needed for NVIDIA's the PR managers. The reputation of FutureMark, as the benchmark developer, suffered. The 3DMark2001 reveals cheats from both companies. That was proved in our review.

Now come the game tests. You will see that NVIDIA decided to raise the speed in the UT2003 as well by making cheats for the demo benchmark. We pointed out before that optimizations made in games and causing no quality degradation are welcome. But it's unfair to make optimizations for only a demo benchmark (for example, DM-Antalus FlyBy) when the speed goes up only in this benchmark, not in the game!

That is why we decided to play safe and created our own demo benchmarks in many game tests (thanks to Danil Gridasov AKA DegustatoR). Only the Code Creatures BenchMark Pro is left 'as is'. Later you will see that all. And now we come back to the hero of our review - Gainward FX Powerpack Ultra/1200 Golden Sample.

Gainward is well known throughout the world; its products made a good showing, and its prices are even the lowest on our local market.

Card

Comparison with the reference design, front view
Gainward FX Powerpack Ultra/1200 Golden Sample	Reference card NVIDIA GeForce FX 5900

Comparison with the reference design, back view
Gainward FX Powerpack Ultra/1200 Golden Sample	Reference card NVIDIA GeForce FX 5900

Here are some aspects to be pointed out: 1. The new FX5900 design differs a little from the reference 5900 Ultra card, - the right section that controls the power supply is redesigned. 2. The memory size was cut down from 256MB to 128MB - on the backside there are empty space for missing BGA chips. But the 256bit bus wasn't shortened because the card has 8 32bit memory chips (256 bits in all). The card just doesn't support the dual-bank memory mode. Now, please, have a look at the cooler. Gainward FX Powerpack Ultra/1200 Golden Sample This is a massive aluminum heatsink equipped with two fans and LEDs for illumination. Such cooler is pretty effective in spite of a single heatsink for the memory and GPU. But the memory doesn't generate much heat (for 2.2ns memory the clock speed of 425 MHz is not that high), the heat generation of the NV35 is significantly reduced, and too complicated coolers are not needed. If video cards were installed with their coolers on top and PC cases were transparent, the LEDs could have a real effect. But the LEDs illuminate only what's under them. The card is equipped with the Philips 7114 codec to control the Video-In. The TV-out is supported by the NV35 itself. Here it is: This product was released on the 18th week of this year (beginning of May), i.e. before the GeForce FX 5900 was announced. Hence the codename. The box contents:

Gainward FX Powerpack Ultra/1200 Golden Sample
User guide, CD with drivers and utilities, VIVO software, S-Video-to-RCA and DVI-to-d-Sub adapters, a VIVO splitter, and an external power supply adapter.

Gainward FX Powerpack Ultra/1200 Golden Sample
This is a stylish box in the style of the latest cards.

Note that Overclocking requires additional card cooling (for its memory, in particular): Overclocking depends on a certain sample, and you shouldn't extend single-card results to the entire series or trade mark. Overclocking results are not obligatory characteristics of video cards. Test results Before we start examining 2D quality, I should say there are no complete techniques for objective 2D quality estimation because: 2D quality much depends on certain samples for almost all modern 3D accelerators; Besides videocards, 2D quality depends on monitors and cables; Moreover, certain monitors might not work properly with certain video cards. As for the samples tested, together with the ViewSonic P817 monitor and BNC Bargo cable they showed excellent quality at the following resolutions and clock speeds:

Test results: comparison of the cards' performance Conventional signs: ANISO 8xP - Anisotropic 8x Performance (earlier it was called Balanced), ANISO 8xQ - Anisotropic 8x Quality, ANISO 16xQ - Anisotropic 16x Quality. Test applications: Return to Castle Wolfenstein (MultiPlayer) (id Software/Activision) - OpenGL, multitexturing,ixbt0703-demo, test settings - maximum, S3TC OFF, the configurations can be downloaded from here Serious Sam: The Second Encounter v.1.05 (Croteam/GodGames) - OpenGL, multitexturing,ixbt0703-demo, test settings: quality, S3TC OFF Quake3 Arena v.1.17 (id Software/Activision) - OpenGL, multitexturing,ixbt0703-demo, test settings - maximum: detailing level - High, texture detailing level - #4, S3TC OFF, smoothness of curves is much increased through variables r_subdivisions "1" and r_lodCurveError "30000" (at default r_lodCurveError is 250 !), the configurations can be downloaded from here Unreal Tournament 2003 v.2225 (Digital Extreme/Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality Code Creatures Benchmark Pro (CodeCult) - the game that demonstrates card's operation in DirectX 8.1, Shaders, HW T&L. Unreal II: The Awakening (Legend Ent./Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality RightMark 3D v.0.4 (one of the game scenes) - DirectX 8.1, Dot3, cube texturing, shadow buffers, vertex and pixel shaders (1.1, 1.4). Note that the Anti-Detect is applied only to the Direct3D applications, the OpenGL games have simply changed their demo benchmarks. Quake3 Arena The tests with the new demo benchmark show that the FX 5900 takes the lead only in the modes with the AA and anisotropy disabled and in the Performance mode with the anisotropic filtering used. But taking into account the way of realization of anisotropy in the RADEON 9800 PRO, the maximum level of the ATI's products can be compared to the Performance level (former Balanced) of the NVIDIA's solutions. Serious Sam: The Second Encounter The FX 5900 loses the battle, though the performance gap is not that great. Return to Castle Wolfenstein (Multiplayer) Like in the Quake3, the NVIDIA wins only in the ANISO Performance mode (but it seems that the CheckPoint demo benchmark wasn't ignored by the guys at NVIDIA, and the FX 5900 had much higher scores than now). Code Creatures First off, the cheat detection revealed that NVIDIA did its "best" and raised the speed by 30%. Besides, without such cheats the FX 5900 would suffer a crushing defeat (that was expected because the Code Creatures aggressively use shaders). Unfortunately, it's impossible to take two identical screenshots here (this is a rolling demo, without interactive support), that is why we can't compare quality with the cheats and without. But even if there are some useful optimizations that boost performance without damaging quality, they could make sense only if this application were a game. But this is a benchmark (even though it's built on the game engine), and it's not fair to make changes and make this test work differently from what was planned by its developers. I must say that we found no cheats in ATI's RADEON for this test! Look at the anisotropic filtering in this test! As there is no speed drop, this function doesn't work! So attractive figures demonstrated by the card in comparison with the RADEON 9800 PRO are wrong. Although it's not a direct evidence of the fact that anisotropy doesn't work here, this function is never costless. Unreal Tournament 2003 The FX 5900 fails again. The Anti-Detect also shows that the cheats increase the speed by 23-37% on the standard benchmark DM-Antalus FlyBy! The RADEON 9800 PRO reveals no cheats again!!! It refers both the the standard test DM-Antalus and our demo. However, some of NVIDIA's optimizations are useful: just compare the performance with the Anti-Detect and without it on our demo benchmark (not the standard one). Since it's impossible to make cheats for our demo version, the difference is caused by useful optimizations which were disabled by the Anti-Detect. Below are several pictures that prove that the quality is the same with the AD and without (but completely identical screenshots can't be attained for per-pixel comparison because the game uses random actions). ATTENTION! The full-size screenshots are recorded in the BMP format! EACH SCREENSHOT MEASURES 2.3 MB and packed into a RAR archive of about 1 MB!

Driver 44.65, Anti-Detect	Driver 44.65, no Anti-Detect

Unreal II: The Awakening The picture for the FX5900 is pretty sad. Fortunately, no cheats were found this time (probably because this game has no standard benchmarks :-) ). RightMark 3D The most shader-intensive test raveled how weak the FX 5900 is in this respect. Fortunately, none of the companies brought in cheats for our test :-), though after this review they can take revenge.

Article navigation: