What makes the Radeon 9000 important isn't a list of never-before-seen features or astronomical performance, but the DirectX 8.1 feature set that it brings to the mainstream desktop. Mass-market adoption of new versions of DirectX is what's needed before it's cost effective for developers to spend a lot of time taking advantage of advanced features like pixel shaders, and that makes the Radeon 9000 a very important graphics card for the industry as a whole.
The Radeon 9000 Pro is also a cheap graphics card. At the launch, ATI quoted a price of US$129, which is a good deal for a fully DirectX 8.1-compatible, multimonitor-supporting graphics card. Now, just weeks after the official launch, you can find Radeon 9000 Pros on Pricewatch for under $90. That's right, under $90. Heck, we thought $129 was a competitive price, but street prices hovering around $80 are insane.
Does the Radeon 9000 Pro have what it takes to push DirectX 8.1 into Joe Sixpack's computer? What exactly did ATI change from the Radeon 8500? We've got the scoop, and we've compared the Radeon 9000 Pro's performance against 12 of its closest competitors. Read on, through some of the biggest graphs you've ever seen, to find out more.
Snipping the Radeon 8500
Sacrifices inevitiably have to be made to compete at a lower price point. Chips aren't cheap to make, and you can squeeze a greater number of smaller dies onto a single silicon wafer. Both the Radeon 9000 and 9000 Pro are based on the RV250 core, a whittled-down version of the R200 chip that powers Radeon 8500. Here's how the RV250 differs from its predecessor:
- A slimmed-down 3D pipeline The RV250 retains the R200's four pixel pipelines, but each pipe now lays down one texture per pass instead of two. Just like the R200, the RV250 can "loop back" within the pipeline and lay down additional textures before writing the result to the framebuffer, only the RV250 can loop back six times where the R200 could only loop back three times. Theoretically, both chips can lay down six textures per pass; they just go about it in different ways. The R200 could deliver more textures per clock, but the RV250's ability to loop back six times may help it perform better when processing complex pixel shader effects in future games.
- One optimized vertex shader To achieve a smaller, simpler chip, the R200's dual vertex units just had to go. The RV250 features just one vertex unit, an optimized version of its predecessor that ATI has dubbed version 1.1.
ATI left all the DirectX 8.1 capabilities intact in the RV250, in part to give it a leg up on its GeForce4 MX competition, but also to push DirectX 8.1 into the mainstream market. (The GeForce4 MX chips don't have vertex or pixel shaders. They are essentially warmed-over GeForce2 MX chips.) It's definitely in ATI's best interest to ensure newer versions of DirectX get mass market support. Only then will the economics of game sales justify a developer taking advantage of the advanced features that last year's latest and greatest graphics cards brought to market. Graphics hardware will always be ahead of the software, but ATI and the other graphics chip makers can't afford to have software fall too far behind.
The RV250 isn't all about taking things out of the R200. ATI has actually added a couple of things that are worth mentioning.
- Better multimonitor support ATI's multimonitor software hasn't changed with the RV250. You can still run multiple monitors with different refresh rates, color depths, and screen resolutions. What has changed is how the chip deals with multiple displays in hardware. The original R200 GPU integrated a single RAMDAC for VGA output and a single TMDS transmitter interface to drive DVI-capable monitors like flat-panel LCDs. Output to a second VGA monitor, connected via a DVI-to-VGA adapter, had to be handled by a second, board-mounted RAMDAC, and TV output was handled by ATI's Rage Theatre chip.
Third-party manufacturers flooded the market with a lot of R200-based boards, and many cut out that second RAMDAC, presumably to save money. Cutting out the second RAMDAC castrated the R200's multiple monitor support, making it impossible to run two VGA monitors, even with a DVI-to-VGA adapter. For consumers, this was a nightmare; support for two VGA monitors didn't appear to be standardized for R200-based products in any way. ATI's own boards had the extra RAMDAC, but boards from Hercules using identical R200 GPUs lacked that all-important second RAMDAC.
The RV250 solves all the multimonitor uncertainty by integrating a single TMDS transmitter interface in addition to two 400MHz RAMDACs. Also, there's a TV encoder right on the GPU. Any graphics board using the RV250 chip, regardless of manufacturer, should have support for dual VGA monitors via a DVI-to-VGA adapter. The Radeon 9000 Pro's simpler and cheaper board design doesn't require a second RAMDAC or a Rage Theatre chip. Consumers should now know better what kind of multimonitor support they're getting, regardless of the brand of Radeon 9000 Pro they're purchasing.
- Video de-blocking It seems a shame to let a graphics card's GPU go largely unused unless you're using 3D applications, so ATI has decided to put pixel shaders to work on video streams. The RV250's version 1.4 pixel shaders aren't advanced enough to do the kind of real-time Photoshop-esque effects the Radeon 9700 supports, but the pixel shaders can be used to "de-block" video playback.
ATI says over 70% of the CPU load of video decoding is related to de-blocking. If so, it makes a lot of sense to offload that work to the GPU. The de-blocking possible on the RV250 actually looks quite good. It has the potential to make some lower quality video at least bearable to watch. Incidentally, the R9000 Pro's video de-blocking should be able to work with any block-based encoding format, including DivX.
The CPU resources saved by offloading de-blocking to the GPU might not seem like a big deal for powerful desktop computers, but it's something that sets up the RV250 core nicely for possible mobile implementations. Like NVIDIA's NV17 (found in the GeForce4 MX and GeForce4 Go), there's also an MPEG2 decoder built right into the chip, which makes a Mobility version sound even more attractive.