ATi Radeon X700 XT - ATi's Answer To The GeForce 6600
Architecture Overview & The Lineup
Built on TSMC's new .11 micron process, the Radeon X700 VPU (aka R410) certainly seems to be an efficiently designed graphics core, at least on the outside looking in at its relatively tiny die size. We'll pop the hood here quickly now and cover what's on the inside, as well.
 |
|
Look familiar? The above chip-level block diagram certainly should, if you've seen our recent X800 launch article. If you haven't been there yet, shame on you. A thousand lashes with a beat-up ribbon cable for you....but alas we digress. The architectural layout and main building blocks of this chip are strikingly similar to those of an X800 VPU (either 12-pipe X800 Pro or 16-pipe X800 XT) and in fact they are, with the small (literally, very small) exception that this mid-range machine is built on .11-micron technology versus the X800's .13-micron fabrication.
Eight Pixel Pipes backed up by a Vertex Six-Gun:
In addition, as you can obviously see, this new ATi VPU is an eight-pixel-pipeline machine, but unlike the Radeon 9800, it has six vertex engines, which is the same number of vertex units an X800 series VPU has. This is actually two more than you'll find on a Radeon 9800 VPU and three more than are in the new GeForce 6600. For the end user, this means that high-polygon-count renderings are going to run faster on this architecture in general because this six-vertex engine design will be able to drive a higher number of polygon transformations per second. ATi claims the X700 architecture has two times the vertex engine power of a GeForce 6600. Mathematically, the numbers seem to support that claim, but the benchmarks will tell the real story. When the action is heavy with lots of scenery detail and character animation, the X700 should have a considerable advantage against others in its class.
Other than that, it's business as usual inside this chip, for ATi's new generation architecture. Think of the X700 VPU as an eight-pixel-pipe version of the Radeon X800 but built on a power- and heat-saving .11-micron process.
|
|
ATi's X700 lineup is pretty straightforward... or is it? Here we see your basic high-end, mid- and low-end VPU clock speed scale, along with associated memory interface speeds that scale lower as you reach the $149 price point. However, there are two $199 cards here, and one of them has 256MB of memory but is coupled with lower speed core and memory clocks. Then there is a 128MB card that has the highest speed core at 475MHz and highest speed memory at 1.05GHz. Here we simply have to ask the question... what's up with that?
We understand the strategy to offer two $199 cards, which seems to be the psychological sweet spot for retail (and fly right in the face of an NVIDIA GeForce 6600GT price point), but why on earth not offer even a fourth SKU that has it all: clock speed and memory density? A strategy could have even been to offer a $179 128MB "Pro" speed card and make the XT a 256MB card at $229. However, to not offer the best of both worlds, that being a full 256MB frame buffer and top speed core and memory clocks together, is just begging the consumer to buy the Pro and overclock it. Or could there be something else amiss? We won't speculate too much here, but it's no secret that Radeon X800 XT PE (Premium Edition) cards are scarcer than hen's teeth in the retail channel. Now, the branding on the top-end X700 XT doesn't suggest it is a "premium" selected card, so we are hopeful that ATi delivers solid volumes of this flavor of the card and that it's not just a cherry-picked benchmark runner.
Our first impressions suggest that is not the case, as you'll see in the pages ahead. The card also has some headroom, as it does overclock pretty well. Perhaps it was a memory loading issue off the X700 memory controller. The more memory that is added to the card design, the tougher it can be for clocking inside the VPU to meet and drive timings at the higher-end memory speeds. So perhaps a 256MB, 1.05GHz card just wasn't possible? In any event, this is all pure speculation at this point, but these questions will all be answered in the weeks and months ahead, as both versions of the cards hit the channel and gamers try their hands at overclocking that 256MB version.
