ATI Radeon X1900 Series Preview

Reviews: Hardware

Motherboards | Power Supplies | Memory | Systems | Video Cards

Author:
Editor:
Sponsor:
Published:

Kurtis Kronk
Brian
ATI
Jan. 24, 2006

Specifications

<< Previous

Next >>

Introducing the X1900 Series (R580)

The R520's late launch was something that ATI caught a lot of flack for. There were months of delays on ATI's new cards while NVIDIA launched the 7800 GTX at full volume, followed by the 7800 GT. While ATI was late, when the X1800 XT finally made it to the market, it was very competitive with NVIDIA's 7800 GTX.

It may seem like it's too soon for a refresh from ATI since they only launched the X1000 series in October (3 months ago), but obviously it wouldn't be in their best interest in the long run to delay future product lines following the R520 delays. Today marks the introduction of the X1900 series, code name: R580. With this launch it looks like ATI is "back" in the game (as if they really went anywhere). Word from ATI is that "thousands upon thousands of X1900 series cards have shipped" to retailers, so we should find the cards in stores almost immediately.

Architecture & New Features

In our ATI Radeon X-1000 Series Preview, we discussed architecture in-depth. The X1900 cards are a refresh product based upon that same architecture. The biggest difference with the X1900 is that it has more pipelines as well as slight increase in core and memory clock speeds. The only other new thing is the new texture sampling feature known as Fetch4. Basically it works by exploiting the fact that textures are composed of color values with 4 components (reg, green, blue, and alpha). It samples and filters all four components from one texture address at once. When used in conjunction with textures that have single-component values, such as shadow maps, it instead takes four values from adjacent texture addresses to be sampled simultaneously. ATI's claim is that with their Ultra-threading technology providing fast flow control and with Fetch4 providing fast texture lookups, the X1900 can render soft shadows at speeds approaching those of traditional hard-edged shadow mapping techniques. Our Splinter Cell: Chaos Theory benchmark will be a good example where this would be useful, so we'll see what kind of implication this technology actually has.

How many pipelines were added, you might ask? A few. More precisely, three times as many. In case you don't recall, the X1800 XT had a core clock speed of 625MHz, 512MB of GDDR3 memory clocked at 1.50GHz, and a total of 16 pixel pipelines. The X1900 XTX has a core clock of 650MHz, 512MB of GDDR3 memory clocked at 1.55GHz, and 48 pixel pipelines. Of course, it's also a tad faster in terms of clock speeds than the X1800 XT. Whether or not these changes make a substantial difference in today's games, we shall see.

Click here for the full-size version of this image
Since I've already written all about the X1000 series architecture, and the X1900 is based on the same architecture, I've yanked the X1000 architecture information from my X1000 series preview article. Please note that the images below refer to X1800, X1600 and X1300 architecture and are being presented for the sake of elaboration only.

The X1000 family is touted by ATI as a "no-compromise design," consisting of a "ruthlessly efficient, ultra-threaded" SM 3.0 core, 512-bit Ring Bus memory controller, AVIVO video display technology, and 90nm technology from top to bottom.

The X1000's shading architecture allows for full-time 32-bit floating point precision, fast dynamic branching and also makes it the first capable of doing High Dynamic Range with Anti-Aliasing. When ATI talks about having an Ultra-Threaded core, what that actually refers to is the size and number of threads. On the X1000 cards you have a small thread size of 16 pixels (4x4) whereas NVIDIA's 7800GTX has a thread size of 64 pixels (8x8), which is obviously a bit larger. ATI argues that their smaller thread size is more efficient, particularly when dealing with dynamic branching.

Click here for the full-size version of this image
The 512-bit Ring Bus memory controller supports the fastest available rendering engines, has a new cache design, improved Hyper Z with better compression and hidden surface removal (removal of 60% more hidden pixels over X850), and programmable arbitration logic which maximizes memory efficiency and can be upgraded via software. There are actually two internal 256-bit rings running in opposite directions to minimize latency which help to reduce routing complexity as well as permitting higher clock speeds. There is one ring stop per pair of memory channels which are linked directly to the memory interfaces. The X1800 has 8x32-bit memory channels compared to the X850's 4x64-bit. As mentioned earlier, the newly improved Z Buffer comes into play when you run into scenes with a large amount of overlapping objects to help reduce overdraw.

Click here for the full-size version of this image

One exciting thing about the X1000 series is that ATI is promising to open the platform by documenting the GPU architecture for DPP and they will seed developers with tools and prototype applications. Something that was discussed at the Tech Days event was that modern GPUs are powerful processors which can be utilized for non-graphics applications such as physics simulation, medical imaging, speech recognition, protein folding, matrix/vector operations, FFT, Ray Tracing, Sort/Search, etc. By opening the platform to developers, it will be possible to create more non-game applications to take advantage of this powerful piece of hardware.

AVIVO is a new technology which, among other things, allows for "universal connectivity" - ATI actually went so far as to say "You got a cable, we can support it." I'm betting I can find a cable that isn't supported, but I suppose they didn't mean it literally. With AVIVO you can encode H.264, VC-1, WMV9, WMV9 PMC, MPEG-2, MPEG-4, and DivX, and you can decode H.264, MPEG-2, MPEG-4, VC-1, and WMV9. All of the X1000 cards are equipped with a Xilleon TV Encoder and will have a dual-link DVI (X1800 has two dual-link DVIs), 10-bit/16-bit DVI output, as well as HDCP and HDMI support.

Speaking of dual-link DVI, I'm reminded of the already infamous 1600x1200 @ 60Hz issue with a pair of X850s in Crossfire. This issue has in fact been addressed with the X1000 series, thanks to a new compositing engine. It turns out that dual link DVI was required for the amount of bandwidth needed to play at high resolutions with high refresh rates. Just for your reference, you will be able to run a pair of X1800s at 2048x1536 @ 70 Hz. Of course, NVIDIA didn't have to worry about that since they went with an internal connector for their dual graphics solution.

AVIVO can also map any input to 10-bit integer, including 8 and 10-bit integer, 16-bit integer (INT16) and 16-bit floating point (FP16). FP16 output via dual-link DVI means HDR on capable displays. You can send FP16 or INT16 data.

We'd like to thank Directron.com for donating the ASUS A8N-SLI Premium Socket 939 Motherboard which we used for testing.

<< Previous

Page 2 of 10

Next >>

Page 1: Introduction
Page 2: Introducing the X1900 Series (R580) / Architecture & New Features
Page 3: The Card (X1900 XTX)
Page 4: Test Setup
Page 5: Benchmarks - Half-Life 2
Page 6: Benchmarks - Splinter Cell: Chaos Theory
Page 7: Benchmarks - F.E.A.R. First Encounter Assault Recon
Page 8: Benchmarks - Quake 4
Page 9: Benchmarks - Need for Speed Most Wanted
Page 10: Closing Thoughts

Subscribe to Video Cards [more info]

Add Comment

3 User Comments

1 - Posted by J2T on May 31, 2007 - 5:53 am

??? time warp?

2 - Posted by Kurtis on May 31, 2007 - 12:17 pm

i'm sure there was a spam post, then the spam was deleted, leaving a bumped thread. =]

3 - Posted by J2T on June 1, 2007 - 6:07 pm

haha, excuses excuses :)

Add Comment

To add a comment without being a member, you may omit the password field, but you must enter your name (or nickname) along with your comment. * Denotes required fields.