Introduction
ATI is launching today their new series of graphic chips based on a new architecture, which we will be explaining its main features in this article. This new series has basically three families, each one for a distinct market segment: Radeon X1300 for the entry-level market, Radeon X1600 for the mid-range market and Radeon X1800 for the high-end market. In each series there are several models being launched and we will explain each one of them on the following pages.
For now, let’s summarize what’s new on Radeon X1000 series:
- Shader 3.0 model (i.e. DirectX 9.0c – at last).
- Ring bus memory controller on Radeon X1800 and X1600 series.
- 64-bit High Dynamic Range (HDR) with anti-aliasing at the same time.
- New anti-aliasing (adaptive anti-aliasing) and anisotropic filter (high quality anisotropic filter a.k.a. HQ AF) modes.
- Full time 32-bit floating point precision.
- The high-end model Radeon X1800 XT can achieve a floating point performance (i.e. math performance) of 83 Gflops (billion of math calculations per second). It’s interesting to keep in mind that nVidia’s GeForce 7800 GTX has a 165 Gflops performance, but Radeon X850 XT Platinum Edition has a math performance of 66 Gflops and GeForce 6800 Ultra, 54 Gflops.
- 2D video processing enhancements, called Avivo (competes with PureVideo, which is available on GeForce series 6 and 7 from from nVidia). Basically what both Avivo and PureVideo do is to make 2D video processing (MPEG2 decoding, for example) to be done by the graphics chip instead of the CPU, increasing the performance of your system, since it will free your CPU to be doing something else.
- 90 nm manufacturing process.
- CrossFire support for all models.
- ATI will document its architecture, i.e. anyone interested in learning how it works inside will have access.
High Dynamic Range (HDR)
The problem of lightning is that in the real world lights have unlimited lightning, and the human eye has a perception of 14 dB (10^14:1), but the video card using a standard 32-bit integer buffer is only capable of reproducing 2.4 dB (255:1), because it uses only 8 bits to store each video component (R, G, B and alpha).
Radeon X1000 series uses a 64-bit floating point register for HDR, using 16-bit for each video component. It is really interesting to note that GeForce 7 series from nVidia uses a 128-bit floating point register for HDR.
On the other hand, Radeon X1000 series allows the use of HDR and multi sampling anti-aliasing (MSAA) at the same time.
New Anti-Aliasing and Anisotropic Filter Modes
Radeon X1000 series provides a new AA mode, called adaptive AA. On Figure 1 you can see the difference between adaptive AA and standard multi sampling AA (MSAA).
click to enlarge
Figure 1: Multi sampling anti-aliasing vs. adaptive anti-aliasing.
As for anisotropic filtering, Radeon X1000 series introduces a new 16-sample high quality mode (16x HQ). This kind of filtering enhances the quality for images that are not on the conventional plan. To better illustrate this feature, remember Star Wars opening titles. As the letters go to the top of the screen, they become fuzzy and hard to read. Anisotropic filtering is used to get a sharper image and to not become fuzzy like in this example. This filtering can be usually configured as the number of samples used on the filtering process. The highest the number of samples, the sharper the game image will be, but the game will be also slower.
Radeon X1800 Family
Architecture
Radeon X1800 family architecture can be seen on Figure 2.
click to enlarge
Figure 2: Radeon X1800 family architecture.
In summary it has:
- 16 pixel shaders processors;
- 8 vertex shaders processors;
- 16 texture address units;
- 16 texture units;
- 16 render back-end units;
- 16 z compare units.
On Figure 3 you can see how a pixel shader unit looks like inside.
click to enlarge
Figure 3: Inside a Radeon X1800 pixel shader unit.
Chips from Radeon X1800 family use a 512-bit ring bus memory controller. Actually this controller uses two separated 256-bit datapaths, running in opposite directions. The memory controller architecture can be seen on Figure 4. Obviously the idea behind enhancing the memory controller is to make it faster to deliver data. It does that by prioritizing memory requests according to the client that asked for it. According to ATI, this change in architecture benefits basically high-bandwidth situations, like when you have your video settings at resolutions from 1600 x 1200 up, anti-aliasing at 4x and up and anisotropic filtering at 8x and up.
click to enlarge
Figure 4: Ring bus memory controller.
Models
Two Radeon X1800 models were announced: Radeon X1800 XT (formerly known by its codename, R520) and Radeon X1800 XL. While Radeon X1800 XL is being launched today, Radeon X1800 XT will be only released on November 5th, 2005.
click to enlarge
Figure 5: Radeon X1800 XT.
The main features for these chips are shown on the table below.
|
Model |
Clock |
Memory Clock |
Suggested Price |
Availability |
|
Radeon X1800 XT |
625 MHz |
1.5 GHz |
USD 499 (256 MB) and USD 549 (512 MB) |
Nov 05th, 2005 |
|
Radeon X1800 XL |
500 MHz |
1 GHz |
USD 449 (256 MB) |
Oct 05th, 2005 |
Radeon X1600 Family
Architecture
Radeon X1600 family is the new mid-range family from ATI, but it will only arrive the market on the last day of November – so you won’t see these video card before december!
Radeon X1600 family architecture can be seen on Figure 6.
click to enlarge
Figure 6: Radeon X1600 family architecture.
In summary it has:
- 12 pixel shaders processors;
- 5 vertex shaders processors;
- 4 texture units;
- 4 render back-end units;
- 8 z compare units.
Like Radeon X1800, Radeon X1600 series also uses the new ring bus memory controller. However, instead of using two 256-bit datapaths like on Radeon X1800 series, it uses two 128-bit datapaths.
Models
Two Radeon X1600 models were announced: Radeon X1600 XT and Radeon X1600 Pro. The main features of these two chips you can check on the table below.
click to enlarge
Figure 7: Radeon X1600 XT.
|
Model |
Clock |
Memory Clock |
Suggested Price |
Availability |
|
Radeon X1600 XT |
590 MHz |
1.38 GHz |
USD 199 (128 MB) and USD 249 (256 MB) |
Nov 30th, 2005 |
|
Radeon X1600 Pro |
500 MHz |
780 MHz |
USD 149 (128 MB) and USD 199 (256 MB) |
Nov 30th, 2005 |
Radeon X1300 Family
Architecture
The architecture of the entry-level Radeon X1300 family can be seen on Figure 8.
click to enlarge
Figure 8: Radeon X1300 family architecture.
In summary it has:
- 4 pixel shaders processors;
- 2 vertex shaders processors;
- 2 texture units;
- 2 render back-end units;
- 4 z compare units.
Models
Theoretically all Radeon X1300 models should reach the market today. Three models were announced: Radeon X1300 Pro, Radeon X1300 and Radeon X1300 HyperMemory. Also, a sub-25 W low profile version was mentioned, but we still don’t know where this model fits. The main features for these chips are shown on the table below.
click to enlarge
Figure 9: Radeon X1300 Pro.
|
Model |
Clock |
Memory Clock |
Suggested Price |
Availability |
| Radeon X1300 Pro | 600 MHz | 800 MHz | USD 149 (256 MB) | Oct 05th, 2005 |
| Radeon X1300 | 450 MHz | 500 MHz | USD 99 (128 MB) and USD 129 (256 MB) | Oct 05th, 2005 |
| Radeon X1300 H.M. | 450 MHz | 1 GHz | USD 79 (32 MB) | Oct 05th, 2005 |
Avivo
Avivo is ATI’s technology for 2D video enhancement, for example, MPEG2 decoding. With a graphics chip with this function, the decoding is done by the graphics chip and not by the system CPU anymore, increasing the system performance, since the CPU will have more free time to do other things at the same time.
On Figure 10 you can see all stages involved on ATI’s Avivo solution. The first two stages – capture and encode – are present inside an external video capture chip, like ATI Rage Theater. If your Radeon X1000 series video card doesn’t have this chip, it won’t have these two stages. Decode, post processing and display stages are embedded in all graphics chips from Radeon X1000 series.
click to enlarge
Figure 10: Avivo pipeline.
Radeon X1000 series chips can decode formats like H.264, MPEG-2, MPEG-4, VC-1 and WMV9.
The post processing stage includes a de-interlacing engine. Videos originally targeted to TVs are interlaced, because that's they way TVs work. In interlacing, each video frame has only half of the total lines available. Video monitors used by computers uses non-interlaced scanning, which is capable of showing all line available per frame, so when reproducing this kind of video on your computer, you can see it doesn't have the best possible quality. A de-interlacing engine creates the missing lines from each video frame, thus improving 2D video quality.
The display stage features gamma correction, color correction, scaling/compression and dithering. In other words, it will correct the image to the best quality. It is interesting to note that this stage works internally using 10-bit color quality. If the digital display used with the video card has 8-bit or 6-bit color quality, the dithering engine will make the proper color conversion.
The display stage also features ATI’s Xilleon processor embedded. Xilleon is a digital TV decoder chip and, according to ATI, 80% of digital TVs in the US use this chip.
click to enlarge
Figure 11: Avivo display stage.
Originally at http://www.hardwaresecrets.com/article/227
© 2004-9, Hardware Secrets, LLC. All Rights Reserved.
Total or partial reproduction of the contents of this site, as well as that of the texts available for downloading, be this in the electronic media, in print, or any other form of distribution, is expressly forbidden. Those who do not comply with these copyright laws will be indicted and punished according to the International Copyrights Law.
We do not take responsibility for material damage of any kind caused by the use of information contained in Hardware Secrets.