Gigabyte Radeon X800 Pro Review

Introduction

The start of 2004 was marked by anticipation of new products by the two video giants: nVidia's NV40 and ATI's R420. nVidia was the first to present its chip and horrified all the reviewers. The performance of that chip was almost twice as much of all the previous chips made by both nVidia and ATI. After all calmed down, the attention was focused on ATI. The main intrigue was about whether R420 would be able to become equal at speed with GeForce 6800 or even surpass it?

So, on 5 May ATI releases the X800 series of video cards on R420 chip. As the first tests with ATI Radeon X800 XT top models showed, it is not inferior at speed to its competitor, and even surpasses it at benchmarks making intense use of shaders. R420 also demonstrated superb results with the FSAA and AA enabled.

However, a lot of reviews of new video cards caused a negative effect on the sales of previous-generation high-end products (primarily, Radeon 9800XT). Practically all users who planned an upgrade of their video systems decided to wait for the release of X800 XT. But there passed a few weeks before such cards hit the retail.

On the whole, the situation with new chips looks very interesting. It is already customary for users to expect another 5-10%, maximum 15%, performance boost with the release of every new chip. A typical example is the evolution "ATI Radeon: 9700=>9700Pro=>9800=>9800 Pro=>9800 XT." But in fact all the listed video cards are based on the R300 architecture. At every stage, engineers of that Canadian company modified the structure a bit, introduced local amendments, fixed bugs. In parallel, the process technology was being debugged, which allowed gradually raising the core speed. As a result, the performance rise was going in small increments.

In the case with the R420 chip, the situation is absolutely different - its architecture is completely new, and ATI intends to realize its potentials for quite a long time. As before, once every six months a new chip with small amendments made to the structure and a bit increased clock speeds will be released. This is highly probable, of course unless nVidia raises the tempo of releasing new products.

As I already said, Radeon X800 XT is the fastest video card made on the R420 chip. However, because of its too high price most users are interested in it only as a virtual joy. That is the fate of every top-end product, and manufacturers are not trying to change the situation somehow, simply because their task is not to sell these boards in mass quantities. They serve primarily a means for the marketing struggle against competitors. Certainly, if such a card sells, manufacturers gain their profits, but the major portion of profits come from sales of the middle-end sector (within 200$ to 350$) that sell in millions of pieces.

It's just for this sector ATI releases a cut-down modification of the R420 chip late in May. Video cards made on this chip are dubbed X800 Pro. To illustrate the differences between X800 Pro and X800 XT, we summed up their specifications into the below table:

	Radeon X800 Pro	Radeon X800 XT
Q-ty of transistors	160 mln	160 mln
Core speed	475 MHz	520 MHz
Memory frequency (DDR3)	900 MHz	1120 MHz
Q-ty of pixel pipelines	12	16
Memory capacity	256 MB	256 MB
Bus width	256 bit	256 bit

As is easy to notice, the differences between the Pro and XT versions are about the cut down number of pixel pipelines and different core speed. The chips are fully identical among themselves and are cut up of the same wafers. At the stage of sorting, chips with broken (i.e. inoperative) pipelines are marked for the production of X800 Pro. With the well-established process technology, the number of these chips may be quite small, and in order to maintain the production plan fully operative R420 chips will be used for the manufacture of chips for X800 Pro. As a result, overclockers are facing a wide area of activities for turning X800 Pro into X800 XT. This sort of re-make should bring quite sizeable profits: from a video card costing ~450$ (with the recommended 400$) we get a video card of much higher performance and price about 600-650$ (these models have not yet gained widespread occurrence in the retail).

In practice, turning X800 Pro into X800 XT is about linking one bridge that was cut up with a laser. For that purpose, conductive adhesive or a lacquer would suit.

Of course, there arises the question "if the new chip is so fast, why overclock it?" The user would say "The speed is more than enough..". Indeed, of all the existing games, only FarCry which makes active use of version 2.0 shaders demonstrates a decent level of graphics. All the other games offer quite primitive graphics in order to expand the circle of potential buyers. Many buyers may have quite old video cards (like nVidia Ti4200 :) which are an "obstacle to progress" (a joke, of course :). By the way, FarCry also suffers from simplifications - take for instance the immovable sky, coarse and scanty machinery models etc.

It seems like the only reason for purchasing a video card of X800 Pro level is the expectation of future demands - in order to be ready for Doom3, HalfLife2 and Stalker, or maybe two-three more games. But game developers keep on deferring the release dates, so we have to wait more and more. All developers except ID Software defer the dates - their standard reply is: "the game will be released once it is ready :)". Doom3 is already finished and will hit the retail in a week or two. The first tests showed that previous-generation Radeon 9800XT video cards do not provide acceptable gaming level at the High-Quality level. That is, purchasing a X800 ProXT is recommended solely to Doom3 gamers.

But hardcore overclocking is the fate of a narrow part of users. Today, we'll be looking at the nominal performance of X800 Pro video card. To this end, we'll be testing the GV-R80P256D model made by Gigabyte.

The vide card is packaged in a very big cardboard box with a handle to carry it around.

Apart from the card, there are the following components in the card:

• Power supply adapter (splitter);
• DVI-to-D-Sub adapter;
• S-Video cable, CA cable
• S-Video=>RCA adapter
• S-Video=>HDTV adapter
• User's Manual in English and Chinese;
• A Catalyst drivers CD + technological demos.
• Rainbow Six3 Raven Shield (full version - 2 CDs);
• SpellForce (full version - 2 CDs);
• Licensed version of PowerDVD 5.0;
• Licensed version of 3D-Album 2.03;

The card offers AGP x8/x4 interface and is a precise replica of the ATI Radeon X800 Pro reference design.

There is 256 MB of GDDR3 memory onboard. Eight chips altogether are installed on the front and reverse sides of the card.

The memory chip is made in the BGA package and offers 2 ns access time, which is equivalent to 500 (1000DDR) MHz operating speed.

The real memory frequency is a little bit different from the requirements of the reference design. It amounts to 445(890) MHz, and not 450 as it should be according to the X800 Pro specifications. The chip frequency is 3 MHz less than the required: 472MHz versus 475 MHz. In any case, both the chip and the memory offer some margin for frequency increase (i.e. for overclocking :).

Overclocking and heat emission

Despite the complication of R420 (increase in the number of transistors), developers succeeded in maintaining power consumption within reasonable limits. At that, it is the achievement of 0.13 mk process technology with the use of copper connections and low-k dielectrics. As a result, the power consumption of the X800 Pro card is 50-60W, and 70-75W for X800 XT.

Clearly, the need for external power source in Radeon X800 still remains. That is why there is a 4-pin molex connector to which a power cable from the PSU is plugged in.

Once I received the video card, I ran a series of overclocking experiments. As a result, we get the following maximum frequencies: the chip clock speed = 545 MHz, memory frequency = 600(1200) MHz. At first glance, the results are excellent, but we haven't yet tested the cards of other manufacturers.

It is worth noting that the overclocking results is a purely subjective thing, and for each specific card the maximum frequencies may be varied. Anyway, the scatter of readings is rarely too large. This holds true not only for video cards of a single manufacturer, but for the vast majority of cards built on the same chip (in this case, it is ATI Radeon X800 Pro).

At these frequencies, the nominal cooling system hardly copes with the heat load. The copper radiator is too hot to touch without the risk of getting burnt. This implies the need for additional cooling during overclocking especially if such expensive card is fitted inside the poorly ventilated PC housing.

Without additional ventilation during overclocking, the following may happen - the paint on the cooler casing may crack. Clearly, this spoils the good look of the product, which may entail problems in upgrade or getting your money back.

On the whole, the cooling system hasn't changed at all compared to Radeon 9800XT. It is a large copper plate with fins covered with a plastic casing. On the edge, there is a large 70 mm fan that runs air through the fins. This is done to eliminate the "dead spot" and maximize the efficiency of using the air flow.

The cooler is not heavy and is fastened with two pins and one screw.

On the reverse side of the cooler, there are special pad for memory chips. The pads are not meant for heat transfer and more likely for smooth installation of the cooler without warps.

The thing is that GDDR3 chips emit relatively little heat, which allowed the developers not to install radiators on the chips at the reverse side of the board.

The fan offers a nice blue highlight that varies in intensity depending on the rotational speed.

To make it possible for the cooler to change the speed, it is plugged in via a 3-pin connector.

And the rotational speed varies depending on the chip temperature. By the way, the R420 chip has a thermal sensor through which data can be read with various utilities.

In practice, the cooler proved to be very quiet in operation. It is significant that it has compact dimensions which let the nearest PCI slot leave free. But that entails a rise of the core temperature as a tradeoff.

Performance

To test the performance, we assembled the most powerful system using available components. In particular, it is the Abit IC7-MAX3 (Intel 875P) motherboard. Then, we installed a Prescott Pentium4 processor and set the clock speed = 3.6GHz (225õ16). At that, the memory (PC3200 Corsair TwinX) was running in the asynchronous mode at 180MHz with the latency timings 2-3-6-3. Such a system allowed us to load the video card to the full and estimate its true potentials.

We compared performance of the card relative to the previous-generation video card - GeXCube Radeon 9800XT. We ran tests at nominal frequencies, in the overclocked mode, and in the maximum quality mode (4x FSAA and 16x AF).

We used the following set of benchmarks:

• MadOnionFutureMark 3DMark2001 SE;
• MadOnionFutureMark 3DMark2003 v340;
• CodeCult Codecreatures v1.0.0 (DirectX 8.1 application, shaders, Hardware T&L);
• Digital Extreme/Epic Games Unreal Tournament 2004 Demo (Direct3D, Hardware T&L, vertex shaders, Dot3, cube texturing. We use the "DM-RANKIN" demo);
• Yeti studious Gun Metal Benchmark 2 v1.20s (DirectX 9.0 benchmark , Vertex Shaders 2.0, Pixel Shaders 1.1, Hardware T&L);
• Egosoft X2: The Threat (Direct3D, multitexturing, Dot3);
• Microsoft HALO: Combat Evolved (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L);
• Massive Development AquaMark 3 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L,);
• Legend Ent./Epic Games Unreal II: The Awakening v1403 (Direct3D, vertex shaders, Hardware T&L, Dot3, cube texturing, the quality set to that suggested by BenchemAll).
• Core Design / Eidos Interactive Tomb Raider: Angel of Darkness v49 (DirectX 9.0, Vertex Shaders 2.0,Pixel Shaders 2.0, quality set to default, demo reels - "Paris5_4", "Paris3");
• Demo Crytek / UbiSoft FarCry (DirectX 9.0, Pixel Shaders 2.0).
• ID Software Return to Castle Wolfenstein (OpenGL)

Let's first take a look at the results of synthetic benchmarks.

3D Mark 2001 is already outdated, and we used it simply as a tribute to tradition :). The performance in the first three tests ("terminator", "dragotic" and "matrix" :) is already much more than enough (the fps rate exceeds 300-400), so we restricted to only the last test - "Mother Nature".

And the last synthetic (or semi-synthetic :) benchmark. We will be using real-world gaming applications.

Performance at Comanche game is in strong dependence on the processor's clock speed. Besides, the type of the processor is very important - for example, the clock speed of Pentium4 (Prescott) was approximately by 15% slower than Northwood, similar at the core speed. Therefore, we can see the difference in speed (1600x1200: ~55%) between 9800XT and X800 Pro only with the FSAA and AA enabled.

In the CodeCreatures benchmark, the speed of the platform does not play any part. So the difference in speed is seen in all the modes and resolutions (1600x1200: ~72%).

The "spice of the show" is FarCry. Again, the card based on the X800 Pro chip shows higher performance (1600x1200: ~53.3%). What is more important, it's just due to X800 Pro the minimum fps level has gone up, which allows playing comfortably even in complex scenes.

At the GunMetal benchmark, the performance boost is even higher - up to 63% at 1600x1200. However, the totals index of speed is still below the desired 60 fps despite the use of a 3.6 GHz processor.

At Halo2, the performance boost is about 51% at 1600x1200.

Now look at the performance in the OpenGL application. While the classical Quake3 is an ideal testing environment for processors, memory and motherboards, it does not work for testing video cards because the graphics is too simplified. That is why we used Return to Castle Wolfenstein which, albeit based on the Q3 engine, offers a higher quality graphics. At Return to Castle Wolfenstein the performance boost is minimum - 18% (at 1600x1200). This once again shows that high-end cards should be tested with the most recent and advanced hi-tech games.

The SeriousSam SE game is a bit more complex than RtCW, but anyway the fps counter exceeds 100 at all the applications. The performance boost at 1600x1200 is ~25%.

Another, very demanding and hard for video cards benchmark that makes intense use of DirectX 9.0. That's where X800 Pro shows its full strength: the performance boost is 45-55% at 1600x1200. Remarkably, nVidia products lose seriously at this benchmark to ATI chips. The lag is so serious that programmers at nVidia were unable to eliminate it even through the use of proprietary "optimizations/cheats". As a result, pressed by nVidia, the game developers had to disable the benchmark in the latest patch (game "Tomb Rider: Angel of Darkness" takes part in nVidia's initiative "The way it's meant to be played").

There is one more interesting benchmark, indicative in terms of nVidia vs. ATI opposition. Since the game offers no integrated benchmark, programmers at nVidia did not do any cheats for that (like, e.g., for Unreal Tournament). However, Alexandr Kondratyuk, a developer of the BenchemAll, found a way to enable the benchmarking feature (for which he deserves special respects :), and the program showed that nVidia's new drivers absolutely don't raise performance in this game.

In Unreal Tournaments 2004, the boost is even higher - around 53%.

Half-Life2 is its alpha version is another benchmark with active use of shaders. The result is that X800 Pro demonstrates a 20-70% superiority depending on the demo-recording used.

The last, fairly hard test, albeit not using DirectX 9.0 to the full, but is able putting a serious load upon your video card. At that, X800 Pro surpasses 9800XT by 23.5% at 1600x1200.

In terms of performance on the whole, we can say that Radeon X800 Pro is 1.5 times (or 50%) more powerful than high-end cards of the previous generation (Radeon 9800XT).

As regards the board's performance in the overclocking mode, the boost is about ~20% on the average. This is without regard for the fact that the card hasn't undergone hardware modifications (i.e., only 12 of 16 pipelines were activated). The cooling system has not undergone essential changes either. We only added an additional fan and replaced the nominal thermal paste with Zalman paste. That provided further overclocking margin.

The only what I can say regarding the maximum quality mode (antialiasing and anisotropy) is that it makes sense to enable it only with games whose speed exceeds 100 fps. In this case, the speed will definitely drop (by 30-50%), but this change will not affect the gameplay quality.

Final Words

Summing it up, we say that at performance, the main factor, Radeon X800 Pro is 1.5 times more powerful than the previous-generation high-end model (Radeon 9800 ProXT). However, at the "price/performance" ratio it loses to 9800Pro. I presume the reasoning of "price/performance" ratio are somehow inappropriate when we talk about new products or a game like Doom3 at the maximum image quality.

Clearly, for Doom3 the X800 XT version is even better (at its 600$ price). But X800 Pro offers one attractive feature - the possibility to enable 4 pipelines and turn it into a full-featured X800 XT. This will definitely of appeal to overclockers. As a result, the X800 Pro card is the most optimum choice between the slowest and outdated 9800ProXT series on the one hand, and the very expensive X800 XT, on the other.

Specifically, regarding the Gigabyte GV-R80P256D model I can say it is a superbly made product based on the reference design. The board offers a good package bundle and a quite low-noise cooler. The board also demonstrated perfect overclocking results, albeit through immense heat emission.

Read more on this topic: