![]() |
|||
|
|||
|
:
News
: Archives
|
|
The Definitive Multi-GPU Roundup |
|
|
Introduction Summer 2006 |
|||
|
|
|||
|
by Josh Walrath |
|||
|
The ATI Side ATI has taken a different approach with this generation of chips than they really have before. ATI has focused on performance and features and have sacrificed modest transistor counts and die sizes as a result. It is undeniable that ATI has the most feature packed and advanced design chips out there, but due to production and cost constraints, they are seemingly evenly matched with their NVIDIA counterparts. The ATI offerings for this article are the X1600 Pro, X1900 GT, and X1900 XTX/CrossFire. The X1600 Pro is based on the RV530 chip, while the three higher end cards are based on the RV580 chip. There is a significant difference between the X1900 GT and its XTX and CrossFire counterparts though which will be covered below. This generation of chips from ATI has many unique features compared to the competition. The X1x00 series of cards are all SM 3.0 capable, but have some nice added features. ATI has first of all uncoupled the texturing units from the pixel pipelines. This means that there are essentially four separate functional areas: vertex, texture, pixel shading, and render operation units (ROPS). For the RV530 and R580 chips ATI has designed a Ringbus Memory architecture which is 128 bits wide for RV530 and 256 bits wide for R580. It is also a programmable memory architecture, so its potential efficiency can be great. Branching performance was also considered key, so ATI designed the architecture with an Ultra-Threading Dispatch processor which breaks down the workload into small threads of 4x4 pixels. Each pixel pipeline also features a Branch Execution Unit. The ATI hardware is pretty much unmatched in branching performance, and the overall architecture is quite efficient. But again, because of these features, die sizes and transistor counts are greatly increased over competing parts. In terms of visual quality these products are unmatched. ATI has retained the same anti-aliasing features that were introduced with the Radeon 9700 Pro but has further tweaked the performance and functionality. It can do two samples per clock, with a maximum of 3 passes (for a total of 6 samples per pixel). It also features a programmable sample pattern, so the pattern is not rotated or ordered but rather described as “sparse”. ATI has enabled transparency supersampling, so alpha textures can be anti-aliased as well (much like NVIDIA’s Transparency AA). ATI has the added feature of enabling its anti-aliasing unit to work with HDR (FP16 values). NVIDIA does not support AA combined with HDR, but ATI can. It does take a significant performance hit, but it is an extra feature that can be quite useful. In the texture filtering department ATI is also unmatched. While using default settings ATI is very similar to NVIDIA in terms of bri-linear/angle-dependent texture filtering. This filtering does increase the performance with little image degradation, but is noticeable in many situations (such as textures that are set at non-horizontal or non-vertical angles looking blurry). ATI gives the user the option of enabling a high quality texture filtering setting, which features true trilinear filtering that is angle independent. This makes the scene about as pristine as it gets, and in some applications the subtle quality changes can have an overall stunning appeal. A good example would be in World of Warcraft, with its large, relatively flat environments, features a lot less distortion and shimmering on the ATI products than NVIDIA products. The X1600 Pro features features 4 ROPS, 4 texture address units, 12 pixel shader units, and 5 vertex shader units. The core runs at 500 MHz with 256 MB of 128 bit GDDR-2 running at 340 MHz giving it around 10.8 GB/sec of bandwidth. The RV530 is comprised of around 150 million transistors at around 152 mm square. Active cooling is included on most of these parts, but there are several fan-less variants available. The X1900 GT is based on the R580, but it is a partially disabled chip. It features 12 ROPS, 12 texture address units, 36 pixel shader units, and 8 vertex shader units. The core typically runs at 575 MHz with 256 MB of 256 bit GDDR-3 running at 600 MHz for an effective bandwidth of 38.4 GB/sec. Even though parts of the chip are disabled, the entire R580 chip is still intact. It still features the full complement of 380 million transistors packed onto 354 mm square of real estate. Active cooling is a must, but the single slot cooler does a fine job, but can be a bit noisy with stock products at full speed. The X1900 XTX and the X1900 CrossFire Edition are both based on the same R580 and board design, except the CrossFire Edition is clocked 25 MHz lower on the core, and a memory clock 50 MHz slower than on the X1900 XTX. The XTX version is clocked at a brisk 650 MHz core and 775 MHz 256 bit 512 MB GDDR-3 giving an effective bandwidth of 49.6 GB/sec. Double slot cooling is used on both cards, and at low speed the fan is very quiet, but once a 3D application is started then the fans spool up and can be heard well outside of the case (but far from being the dustbuster of the FX 5800 Ultra days).
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2006 PenStar Systems, LLC. |
|||