Tech-Hounds.com

Deeper with the Radeon X1900 Series

PCI Express x16 graphics cards
Chipset X1900 512 MB, 256 bit

It's only been three months into the year 2006, but the next generation of graphics chipsets are already here. This time around, the first salvo is from ATI, in the form of the Radeon X1900 series formerly known as R580. When the very late Radeon X1800 came out several months ago, most people agree that the chip to watch out for is not the Radeon X1800 (or R520) but the R580. Architecturally, the two are very similar, but the R580 is equipped with more shader units per pixel pipeline than the R520 - 48 shader units in all.

Both the Radeon X1800 and X1900 signify a new move in realtime consumer 3D graphics - a more shader oriented approach. Granted, shaders is not a new thing, it's been present ever since DirectX 8 made its debut in realtime consumer 3D graphics. However, due to limitations of graphics hardware and API, these shaders are very simple in nature and not used very extensively. With the move to DirectX 9 and SM 2.0 / 3.0 and the rise of shader processing power, shaders are now used more extensively than ever before. In fact, some of the newest 'eye candy' effects present today like motion blur, depth of field and high dynamic range rendering which until now remains in the realm of offline rendering is now possible with shader capable graphics hardware.

If shader capable graphics hardware is so abundant this days, what makes the Radeon X1800 and X1900 unique? It's their design, the Radeon X1900 with its 48 shader units in particular. Unlike the GeForce 7 series from NVIDIA, ATI decided to keep the number of pixel pipelines at 16 for both the Radeon X1800 and X1900. Both cards rely on a higher operating speed (MHz) to maintain fillrate parity with NVIDIA's GeForce 7. On the other hand, NVIDIA chose to equip the GeForce 7 series with 20 pixel pipelines for GeForce 7800GT and 24 pixel pipelines for the 7800GTX, respectively. With more pixel pipelines, NVIDIA GeForce 7 can offer higher frame rates than the Radeon X1800 or X1900 on texture dependent games. However, on shader heavy games, the Radeon X1800 and X1900 should provide the same level of performance to the GeForce 7800GTX. Most likely even higher in the case of X1900. ATI is 'betting' that the clock speed difference is sufficient enough for their cards to have the same level of performance compared to the GeForce 7. But when it came to shaders, the Radeon X1900 should be significantly faster than the GeForce 7800GTX. This makes both the Radeon X1800 and X1900 unique, in a manner marking the transition to even more complex and comprehensive use of shaders in realtime 3D graphics.

There's a reason why ATI chose to push shaders. Textures take up space, but more importantly, they take up bandwidth. We're already seeing 512 MB graphics cards today, which should be the norm this year with the coming of new games and Windows Vista. However, unlike capacity that's ever growing, memory bandwidth is relatively 'stuck' and have virtually hit a brick wall - even the fastest GDDR4 available can only offer 10 - 11 GBs of bandwidth. With a 256 memory controller, that's about 40 - 44 GBs of bandwidth total. Now look at the GeForce 7800GTX 512, a card that has about 50 GBs (or so) of bandwidth. This is the upper limits of current memory technology. Our article last month (which you can read here) shows that to get a significant increase in performance, either from a higher clocked core and / or more pixel pipelines out of the current architecture, we need to have more than that. There's no point of putting more pixel pipelines with the bandwidth available today.

Furthermore, games today are focusing on shaders more and more. ATI feels that this is the area next generation of graphics cards should focus on. Not just pixel or texel fillrate. The Radeon X1K series are actually the first SM 3.0 capable parts from the Canadian based manufacturer. The series span across different levels, from the entry level X1300, the mainstream X1600, and the high end X1800 and X1900. It's about time too, since rival NVIDIA have had SM 3.0 parts for more than one year now with the GeForce 6 and 7 series. ATI felt SM 3.0 was not needed a year ago, and they are right with so many games and developers still targeting SM 2.0 hardware or still in the process of transitioning from DirectX 8 to DirectX 9. Games are just beginning to utilize SM 3.0, despite the already widespread use of SM 3.0 capable GeForce 6 and 7 series. Even now, there are only a number of titles available today that utilize SM 3.0. ATI feels the time has come for SM 3.0 to take center stage, that's why they're introducing the X1K series. This move in the desktop arena is in sync with their overall strategy, including developments on the console market. The Xenos chip in the Xbox 360 shares many similarities with the X1K series. The similarities between the two products should ease development on two platforms for game developers, at the very least make porting code between platforms easier and take less time.

Of course, while having more shader power is nice, that power can only be unleashed with the right software. Since games fully utilizing SM 3.0 potential are yet to appear, we really don't know just how good (or bad) ATI's design decision for the X1800 and X1900 is until those games appear on the market. So, the X1800 and X1900 come with more than just raw shader power. It offers higher quality filtering with less optimization, adaptive antialiasing for transparent textures, high dynamic range rendering with anti aliasing and overall faster antialiasing performance. We'll be looking into several of those claims in this article.

Image Quality Comparison

First of, let's examine texture filtering. Below you'll see screenshots of D3D AF tester with the Radeon X1900. The screenshots are arranged in this order: 1x, 2x, 4x, 8x and 16x. In case you're wondering, the number of AF samples is forced from the driver panel.



They're not that much different from the previous generation, the X800. However there is one major difference. ATI took a lot of flak with AF in X800 because it is only applied to the first texture stage (when forced from the driver panel), so the X800 is pretty much only doing bilinear filtering to the other stages. Below you can see that with the Radeon X1900 (and X1800 too), that's not the case anymore. Every texture stage is now correctly anisotropically filtered, even when AF is forced from the control panel. Theoretically this gives us a much higher quality image than on the X800 (under the same settings).



Next up are the application controlled AF screenshots. Just like before, they are arranged in this order: 1x, 2x, 4x, 8x, and 16x. They're pretty much the same thing really, we can see that there are no visible compromises here, except maybe for angle optimization.



Starting with the Radeon X1800, a new high quality anisotropic filtering option is now available. When this option is enabled, angle optimizations are turned off. Below are the MIP filter screenshots with HQ AF enabled (1x. 2x. 4x. 8x and 16x). There's definitely a significant difference between the two, although it may not always be noticeable in gameplay. The most intriguing thing about this HQ AF is that it can be used with both application controlled and driver forced anisotropic filtering levels. But even more so, ATI claims this filtering is essentially free with virtually no performance hit whatsoever.



With the previous generation, we really couldn't say that ATI have the best filtering, although it did provide the better image quality even with optimizations enabled. With the X1800 and X1900, that's simply not true anymore. With the X1800 and X1900 series, ATI does not only offer the best filtering available, but have taken it to a higher level with HQ AF. They are also claiming that the new series have a much more faster 6x anti aliasing performance, nearly close to 4x AA levels. These two improvements should be very welcomed by gamers wanting a higher quality image. We'll take a later look at these performance claims to see if they are true or not. For now, let's take a look at what these image quality improvements offer in real games.

The Real World

Below you'll see several screenshots, both from the Radeon X1900 and GeForce 7. There are three screenshots from the Radeon X1900: 4x AA with 16x AF, 6xAA with 16x AF and 6xAA with with 16x HQ anisotropic filtering. From the GeForce 7, we have 4x AA with 16x AF and 8xS AA with 16x AF, in that order. Let's take a look at them.

Call of Duty

Radeon X1900



GeForce 7



There's very little difference between the Radeon X1900's screenshots. The most noticeable difference is on the fence on the left, between HQ AF and the standard AF shots. However, even 6xAA with 16x HQ AF can not come close to NVIDIA's 8xS. The use of 2 sample supersampling in this mode provide the best image quality for that area. You probably won't notice that in gameplay though. Other than that, we don't really see any improvement over 4x. The reason maybe caused by the lack of faraway objects on the horizon.

Nascar 2003

Radeon X1900



GeForce 7



The most obvious difference between these shots are the slightly sharper, darker texture of the crowd between HQ AF and all other shots. The 8xS do come close, but HQ AF is the definite winner here. If you look down the track, you'll see that the 6x AA 16 HQ AF shot has 'leaner' poles than all the other shots as well. In all fairness, these differences don't amount to much, but they do show that using more samples when anti aliasing will help display faraway objects with slightly less aliasing.

Homeworld 2

Radeon X1900



GeForce 7



Aliasing in motion
Radeon X1900
4x AA
6x AA

GeForce 7
2xQ AA
8xS AA

These static screenshots are pretty much the same. It's not until we see the game in motion do we see the limitations of NVIDIA's anti aliasing compared to ATI's. Even 8xS AA still exhibits aliasing despite the number of samples. ATI is maintaining the tradition of more effective anti aliasing with the X1800 and X1900.

SW: KOTOR

Radeon X1900



GeForce 7



Aliasing in motion
Radeon X1900
4x AA
6x AA

GeForce 7
4x AA
8xS AA

The differences are subtle but they can be seen if you look close enough in the screenshots. They become more apparent when viewed in motion. ATI's 4x and 6x is still more effective than NVIDIA's 8xS in removing aliasing. However, we can see that 4x and 6x antialiasing is not that much different if we were to look only at static screenshots. ATI's 4x mode is that good.

Except for Call of Duty which feature transparent textures, anti aliasing is still better on the Radeon X1900 than on NVIDIA's GeForce 7 series, be it 4x and 8xS. On faraway objects, ATI's 6x and NVIDIA's 8xS does a better job than both's 4x mode, but not that much. So, even if the X1900 6x is still slower than 4x on the Radeon X1900, we actually don't lose that much in image quality. Or put in another way, we really don't gain that much with 6x. The most notable gain is actually the use of HQ AF, though it may not be noticeable in all games.Let's see if those performance claims are true.

Performance

We put off examinations of transparency anti aliasing and HDR performance for the time being - hopefully we can examine them more closely next week. For now, we'll be focusing on performance under default, 4x AA and 16x AF, and also 6x AA and HQ AF settings. To put these scores into perspective, we've included results from both a standard clocked and a higher clocked GeForce 7800GTX (the ASUS EN7800GTX TOP) with comparable settings (default and 4x AA 16x AF). Out of the many cards in the X1900 series, we chose the X1900 XTX for this article, which is the current high end model of the X1900 series. Performance with X1900 XT should be slightly slower, though not by much.

We'd like to thank both Tagan and Kingston for supplying the additional power supply and 1 GB memory modules for this article. We'd also like to thank Gigabyte for providing the Radeon X1900 XTX sample card.

Our test setup
AMD Athlon 64 3500+ socket 939
2 x 1024 MB Kingston KVR 3-3-3 PC3200 DDR-SDRAM
MSI K8N NForce 4 SLI motherboard
Maxtor DiamondMaxPlus9 80 GBs Serial ATA 8 MB buffer
ASUS E-616 DVD-ROM
Tagan TG530-U15 530 watts ATX/BTX power supply

Windows XP Professional with Service Pack 2 installed
ATI Catalyst 6.2 reference driver
NVIDIA Forceware 81.98 reference driver
NVIDIA NForce 4 6.66 reference driver
Creative SoundBlaster Live! 24 bit 5.12.1.512 driver.
DirectX 9.0c

The graphs are pretty self explanatory, but in case you can't see the text, they're arranged in groups of three colors: green for minimum fps, blue for average fps and red for maximum fps. The first group represent test results under default settings, and the second represent results with 4xAA and 16xAF enabled. For the X1900XTX, the third group represent results with 4xAA and 16x HQ AF, while the fourth represent 6xAA and 16x HQ AF. These results are in frame rate per second.

The results:

Obviously we're system limited at 1024 x 768, even with AA and AF. Only at 1280 x 1024 and higher do we see the three cards behave differently. The difference becomes more apparent at 1600 x 1200. It's clear that the GeForce 7800GTX is faster under default settings, but with AA and AF, the Radeon X1900 XTX reigns supreme. We're still seeing very high frame rates with 6xAA and 16x HQ AF in 1600 x 1200 with the Radeon X1900 XTX, much more than we're getting with the GeForce 7800GTX.

Call of Duty is quite old, so we doubt the 512 MB RAM on the Radeon X1900 XTX is making that much of a difference here. ATI retains the lead even when we enabled 6xAA with 16x HQ AF. It looks like ATI's claims has some truth - both HQ and 6x is relatively free. However, we'd have to see if that's true just for old games or all games in general.

	GeForce 7800GTX TOP		GeForce 7800GTX		Radeon X1900 XTX
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	0	0	0
45-60 fps	0	0	0	0	0	0
60-90 fps	3	8	4	32	8	9
90-120 fps	6	58	5	42	9	12
>120 fps	67	10	67	2	59	55
Total	76	76	76	76	76	76

results are in seconds

In case you're wondering, the table above is actually a brokedown of the benchmark results, displaying the time spent throughout the replay organized into several different frame rate ranges. Because this is a replay and not a timedemo, the replay runtime is the same between all cards. Now look at the 60 - 90 fps range, where the minimum fps shown by the earlier graphs are. You'll see both the Radeon X1900 XTX and the higher clocked GeForce 7800GTX spending less than 10 seconds here. We can also see the Radeon X1900 XTX is pretty much system limited, with virtually no difference across the entire fps range with and without AA and AF. In comparison, both the GeForce drop from above 120 fps to the 90 - 120 fps range when AA and AF is enabled.


In Homeworld 2, things looks a bit different although the trend is similar. The Radeon X1900 XTX is faster only at 1600 x 1200 with AA and AF enabled. It has a slightly overall lower minimum fps than both GeForce 7800GTXs. The GeForce 7800GTX is simply faster here because this game rely more on texture than shaders, which suits the GeForce 7800GTX a whole lot better.

The good news is that the Radeon X1900 XTX hardly flinched with even 6x AA and 16x HQ AF at 1600 x 1200. Just like in Call of Duty, HQ AF and 6x AA is free in this game with the Radeon X1900 XTX.

	GeForce 7800GTX TOP		GeForce 7800GTX		Radeon X1900 XTX
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	0	0	0
45-60 fps	0	0	0	1	0	0
60-90 fps	0	3	0	11	3	3
90-120 fps	1	44	0	48	17	17
>120 fps	69	23	70	10	50	50
Total	70	70	70	70	70	70

results are in seconds

Here, we're seeing virtually the same result from the Radeon X1900 XTX both in default and with AA and AF enabled. It spent the same amount of time virtually above 90 fps with and without AA and AF. In comparison, both GeForce 7800GTX is very fast and spent all of its time above 120 fps when running without AA and AF. However, with AA and AF is enabled, both of them took quite a hit and spent most of their time in the 90 - 120 fps and even 60 - 90 fps.


It's not until we hit 1280 x 1024 do we see a difference. With 4x AA and 16x AF enabled, the Radeon X1900 XTX is around 20 to 30 fps faster than the GeForce 7800GTX Performance levels with 6xAA with 16x HQ AF on the Radeon X1900 XTX is pretty much the same with 4xAA and 16x AF on the higher clocked GeForce 7800GTX. We do see a slight difference between 4x and 6x AA and also HQ AF, so they do come with a minimal performance penalty in this game. In 1600 x 1200, this becomes more apparent: we're losing about 10 fps with HQ AF and almost 20 fps with 6x AA and HQ AF.

	GeForce 7800GTX TOP		GeForce 7800GTX		Radeon X1900 XTX
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	0	0	0
45-60 fps	0	0	0	0	0	0
60-90 fps	0	38	0	141	0	1
90-120 fps	19	213	26	117	12	120
>120 fps	240	8	233	1	247	138
Total	259	259	259	259	259	259

results are in seconds

You can see clearly the Radeon X1900 XTX is besting both GTX, both in default and with AA and AF enabled. The Radeon X1900 only dipped slightly to the 60 - 90 fps range, spending 1 second there and continues to maintain a comfortable frame rate above 90 and 120 fps throughout the rest of the replay. In comparison, we can see the higher clocked GeForce 7800GTX dipped to below 90 fps at some point during the replay. The standard clocked GeForce 7800GTX even spent more than nearly two minutes between 60 - 90 fps (141 seconds).


The average fps may mislead you to think the GeForce 7800GTX is faster. That's because the maximum fps from the GeForce 7800GTX is higher than the Radeon X1900. Only when we focus our attention on the minimum fps do we see that the Radeon X1900 XTX offers noticeably higher minimum frame rates. That means less lag than the GeForce 7800GTX. Even at 1280 x 1024, the Radeon X1900 XTX still managed to reach the 30 fps minimum mark where even the higher clocked GeForce 7800GTX dipped below it.

6x AA can still be too much of a strain even for the Radeon X1900 XTX, although HQ AF is free.