Graphic Card Review
Kiriman oleh Najib pada 28 November 2006
In the battle for graphics card supremacy, there are no permanent leaders, especially if the war happens to be between ATI and NVIDIA.
Ever since the GeForce 6 series debuted, NVIDIA has held the performance crown undoubtedly. Then it launched the GeForce 7 series and the king became the overlord and made NVIDIA the God of Graphics with hard launches and excellent performance. Despite the R520's best efforts, the 7800GTX 512 bettered it by far and put it in its place. Things, however, have changed.
Within a span of three months, ATI has gone from the X850 series to the R520 based X1xxx series and now the R580. But can it beat the green men from NVIDIA? Read on and find out...

Before we delve into the architecture and performance side of things with the R580, the first and foremost thing to point out is ATI's promise to hard launch the X1900 series of graphics cards. We are glad to say that the company has finally come true on its promise and the X1900 series of cards are available from numerous retailers online. At this year's CES, ATI was frustrated itself, since it promised to hard launch products numerous times, but never delivered. And as we now know, ATI was serious this time around.
Physically, there isn't anything exciting to talk about. The sample we received from ATI was fairly standard. However, what's interesting to note are the X1900 series details. The X1900XT and XTX versions will be equipped with eight vertex pixel pipelines, 48 pixel pipelines, 512MB of onboard memory, 16 texture units, render backends and Z compare units with 512 maximum threads. The only difference between the XT and XTX versions is their price points and core/memory speeds. The XT comes with 625MHz core/1.45GHz memory while the XTX runs at 650MHz core/1.55GHz memory. ATI has officially priced the XT and XTX editions at $550 and $650, respectively. However, since the launch, the prices have fallen to an extent. As you will later find out, we managed to find an XT version for as low as $500.00 and the XTX edition at $600, though the XTX was near ATI's suggested retail price.

R580 Architecture:
While the overall architecture is not that big a departure from the R520, the main discussion point about the R580 is going to be the sheer pixel processing power. It has true 16 pipelines each with 4 processors giving you a thumping 48 pixel processors. Compare this to the paltry 16 we had in the R520 and you begin to appreciate the difference. It seems its not Intel but ATI that should use Leap Ahead as its tag line.
Even though the R580 is similar to the R520, it's much more than an overclocked version of the R520, an important element to keep in mind.
The 48 pixel processors have not come about in a drunk state of engineering drawing. Rather, ATI has done serious research and common logic to implement a new solution. To understand this, lets flashback to the days when shaders were first introduced in graphics cards. At that point in time, such cards were the prerogative of the rich or someone brave enough to take a loan to get a graphics card with 32MB of memory. Then, there were one or two games utilizing shaders the way they should be (much like HDR today, which is available in a few games, but not exactly widespread).

Another important thing to note is the usage pattern of these shaders. To render a graphic, two kinds of operations are performed over each pixel: texture operations and arithmetic operations. Earlier, the ratio between the two was 1:1, that is, they were both being performed in equal numbers. However, as games get more sophisticated and graphics get more and more realistic, we are seeing an exponential jump in the arithmetic operations being performed today. The growth is so much that the ratio, which was once 1:1, now stands at 5:1 and can, at times, be as high as 7:1 as well.
In fact, a technique called procedural textures allows you to perform most shader operations mathematically, thereby converting it to an arithmetic operation. If you are familiar with the way Macromedia Flash works, then you can draw a crude analogy between the two.
The interesting thing is that texture operations arent largely dependent on the processors in the pixel/vertex pipelines, but are more dependent on factors such as memory bandwidth, size, and speed, therefore, to increase these, you need to upgrade the whole system to a new plane.
Arithmetic operations, on the other hand, are mere calculations. Thus, the more processors you have, the faster the calculations can be done. Since 5 out of 6 operations today are arithmetic, it would obviously lead to a huge performance boost. This is exactly what ATI has done. By increasing the number of pixel processors from 16 to 48, it has essentially tripled the capabilities and, we hope, the overall performance.
Each pixel shader processor is made up of:

- One vector ALU (ADD+Input Modifier)
- 1 Scalar ALU (ADD+Input Modifier)
- 1 Vector ALU (ADD/MUL/MADD)
- 1 Scalar ADD/MUL/MADD
- 1 Flow Control Instruction

Each pixel shader can process between 1 and 5 shader operations per clock. The Ultra Threaded Dispatch Processor assigns these processors with work to do. The UTDP is always trying to re-order instructions for optimal performance, so we suspect that overall, the architecture is out-of-order-execution friendly.
Since the texture memory can't ever be enough, there is an enormous amount of simultaneous threads that keep feeding data/instructions to the processor. So what ATI is saying is that two heads are better than one even though they might be thinking slower. It works, and to NVIDIAs discomfort, it works rather too well.

Test Setup:

Now that you have briefly understood they key changes in the R580 architecture, let's get on with the tests. Our test bed consisted of an AMD Athlon X2 4800+, 2GB Corsair XMS RAM (2-2-2-5), 120GB 7,200RPM Seagate SATA HDD, ASUS A8N-SLI motherboard and a reference Radeon Xpress200 board.
Our tests include:

1. Splinter Cell 3: Chaos Theory
2. Half-Life 2
3. Far Cry
4. F.E.A.R
5. Quake 4
6. 3DMark05

As for the resolutions, we changed things a bit. Since these are phenomenally high performance cards, we chucked the low level settings and upped the quality levels to full in every game. Then, we tested the games at the following resolutions:
1. 1280x1024 (occasionally 960)
2. 1600x1200
3. 1920x1440
4. 2048x1536

Performance Summary:

The results are quite interesting actually. Let's get the obvious out of the way, ATI is the new performance king and has effectively taken the crown from NVIDIA. The X1900XTX comprehensively and mostly convincingly beats the 7800GTX 512, the only card that couldve given it a performance match.
We did have some interesting results though. In Far Cry, at the basic settings, the 7800GTX 512 matched the X1900XTX frame for frame, and in Quake 4, it just marginally bettered it. For some strange reason, the 7800GTX 512 consistently did better than the X1900XTX. This could possibly be due to a driver bug with Catalyst drivers at those specific settings. Plus, the NVIDIA engine favors OpenGL (Quake 4) games anyway, which is another reason for ATI's unfavorable performance. We are sure ATI will be launching a patch quite soon to improve its performance in OpenGL games.
Almost everywhere else, ATI has completely decimated NVIDIAs top offering. With news of the G71 being launched around March, we suspect that the throne will be up for grabs yet again. All we hope for are reasonable prices and sustained supply.
As far as the X1900XTX is concerned, we must mention that ATI has launched two cards, the X1900XTX and X1900XT, and from what we've seen, the X1900XT has the ability to overclock to X1900XTX speeds quite comfortably in most instances. The X1900XT currently retails for $500.00 on average while its high-end counterpart is available for $600, and as you can clearly see, the price difference of $100, or 20 percent, is more than enough for us to recommend the X1900XT as the value purchase. The X1900XTX is targeted at those who can truly afford the card, but the X1900XT is what the enthusiasts should seek if value is a top priority.

Review :