Introduction

With 320 stream processors, a 512-bit memory interface, and over 700 million transistors, AMD’s Radeon HD 2900 XT is an exciting product to talk about. Even the most die-hard NVIDIA fan must admit, on paper it boasts some pretty impressive specs.

High-end cards like the Radeon HD 2900 XT and its nearest competitor, the GeForce 8800 GTS from NVIDIA, draw all the buzz and headlines from press and end user’s alike, but it’s actually the cards that sit just below the high-end that do all the volume. After all, forking over $400 or more for a graphics card isn’t something that a lot of people actually do -- it’s a lot like reading about the high-end sports car or convertible, but actually buying the more practical, less expensive sports coupe or sedan that it shares some traits with. A quick glance at the latest hardware survey from Valve reinforces this. The top GPUs among Steam users are mainstream parts like the GeForce 6600/7600, and Radeon 9600 from ATI.

Because of this, it’s important that AMD and NVIDIA have competitive products in this segment. This is where market share gains can be won and lost.


After announcing the Radeon HD 2000 series a little over a month ago, today AMD-ATI and their board partners are pulling the wraps off their mainstream and value desktop GPU lineup, namely the Radeon HD 2600 XT and 2600 Pro, and the Radeon HD 2400 XT and 2400 Pro. Due to time constraints, today we’re going to focus on the former cards in ATI’s mainstream lineup, the Radeon HD 2600s. Let’s take a look at the specs:

Mainstream Graphics Comparison
	Radeon HD 2600 XT (GDDR4)	Radeon HD 2600 XT (GDDR3)	Radeon HD 2600 Pro (DDR2)	GeForce 8600 GTS	GeForce 8600 GT
Stream Processing Units	120	120	120	32	32
Shader Clock Speed	800MHz	800MHz	600MHz	1.45GHz	1.19GHz
Core Clock Speed	800MHz	800MHz	600MHz	675MHz	540MHz
Texture Units	8	8	8	16	16
Render Back-Ends	4	4	4	8	8
Memory Frame Buffer Size	256MB*	256MB*	256MB*	256MB*	256MB*
Memory Interface Width	128-bit	128-bit	128-bit	128-bit	128-bit
Memory Clock Speed	1100MHz	800MHz	500MHz	1.0GHz	700MHz
Memory Bandwidth (GB/sec)	35.2	25.6	16	32GB/sec	22.4GB/sec
# of Transistors	180M	180M	180M	289M	289M
Manufacturing Process	65-nm	65-nm	65-nm	80-nm	80-nm
MSRP	$149	$119	$89-$99	$199	$149

The Radeon HD 2600 Series

At the top of AMD’s Radeon HD 2600 lineup is the Radeon HD 2600 XT. This card is intended to take on NVIDIA’s GeForce 8600 GT. We included an asterisk in the memory size field because honestly, both AMD and NVIDIA leave this spec up to their board partners. One cool feature that isn’t listed in the specs above though is the Radeon HD’s support for HDMI audio output; a separate pass-through cable for carrying audio isn’t necessary as it is with HDMI-equipped GeForce 8600 cards. AMD and their board partners will be bundling HDMI adapters with their cards which are capable of delivering full audio and video via the standard DVI output on the back of the Radeon card.

In addition, the Radeon HD 2600 series supports AMD’s Avivo HD technology. In a nutshell, Avivo HD provides dedicated hardware decode acceleration for playing back high definition Blu-ray and HD-DVD movies. AMD is quick to point out that their Radeon HD cards support both VC-1 and H.264 Codecs, whereas NVIDIA is limited to just H.264.

Another interesting, yet untapped feature present inside the Radeon HD 2600 series is AMD’s tessellation unit. This is the same unit found in the Radeon HD 2900 XT and Xbox 360. The idea here is that game developers can use the tessellation unit to create highly detailed scenes/game characters with much fewer triangles than would otherwise be necessary. This in turn frees up memory bandwidth, resulting in improved performance.

Looking over the specs, you can see that the key design principles found in the Radeon HD 2900 XT carry over into the 2600 line. In other words, the GPU is a very forward-looking design that’s heavy on stream processors (120 in the case of the Radeon cards, versus just 32 in the GeForce 8600s), but light on more traditional aspects of the rendering pipeline like texturing and ROPs. It will be interesting to see how this plays out in our benchmarks…