CB: Yes. Absolutely. Apple's entire windowing system depends on OpenGL. That's an industry open standard. Microsoft depends on its proprietary DirectX technology.

AFR: With Vista...

CB: Yes.

AFR: Right, Apple is a key stakeholder in OpenGL. Without it the user
experience of Mac OS X wouldn't be the same.

CB: The thing is Mac users are always concerned they are not getting the best stuff. They want the same as their PC brothers. Over the past few years, since the ATI Radeon 9600, the ATI Mac drivers have combined Apple specific code to support the advanced features and fast paths of MacOS X together with code leveraging the best parts of ATI's PC OpenGL drivers.

For example, at AMD we have an entire group dedicated to writing a really good compiler for vertex and pixel shaders. That same compiler that is used for the PC graphics card development is used for the Mac cards. We use the same code. When hardware became programmable we started being able to do this. It changed everything -- we could expose programming languages as the interface to the hardware. The driver then needed to contain a compiler.

AFR: So all your AMD (ATI) graphics hardware is fully programmable?

CB: Yes, since the Radeon 9600 all our hardware has supported vertex and pixel shaders. The old hardware, for instance, had the idea of a light. It had 'fixed function' circuitry devoted exclusively to lighting calculations. With the new hardware you don't talk about lights. Instead you express the math done in those calculations as a vertex or pixel shader, which gets executed by general purpose ALU hardware on our chips.

The Latest Cards

AFR: Okay let's get to the main question. Why should I buy the ATI
Radeon X1900 GT card? What's good about this card? Who is it for?

CB: Because it's great! [...laughs]

AFR: I'm sure....

CB: The Radeon X1900 GT is the chip in our newest most powerful retail card for the Mac market, which is actually called the ATI Radeon G5 Mac Edition. It is targeted specifically to owners of the last generation of Power Mac G5 towers, the machines with PCIe connections. There is also a sister card, the Radeon X1900 XT, that that has additional graphics muscle, which users can select as an upgrade for the Mac Pro.

AFR: This is the generation of Power Mac G5 with its own dual-core G5 chip from IBM. It came in dual and quad models. I'm just being clear.

CB: Yes. These machines shipped with a default Nvidia GeForce 6600 card. For Pro users who purchased these machines with this card our Radeon X1900 G5 Mac Edition offers a dramatic improvement in 3D graphics performance.

AFR: There was an Nvidia GeForce 7800 card as well as a BTO option.

CB: Yes, but that card is not available as an after-market option for
users who purchased the default Apple config. Our Radeon X1900 G5 card fits the bill nicely for users who want more graphics power. Our
testing showed that the Radeon X1900 G5 Mac Edition was up to 2x-5x faster than the Nvidia GeForce 6600 card that shipped as the default in the PCIe G5s. Up to 2x faster for vertex geometry bound cases, and up to 5x faster for pixel bound cases, such as Core Image and high-resolution shader heavy games.

AFR: So what is the difference between this card and the one that ships as an option in the middle range for the new Mac Pros?

CB: The Radeon X1900 G5 Mac Edition is for PCIe-based Power Mac G5s only, hence the name. It's core processor speed is 500 MHz while the Radeon X1900 XT is 600 MHz. Likewise, its memory speed is 1.2 GHz while the XT card is faster at 1.45Ghz. Additionally, the pixel shader processors number 36 on the GT Mac Edition while the XT has 48. The vertex shaders are the same at 8. So overall, the XT will have around 30% faster vertex and pixel performance than the GT.

AFR: How is that breaking out specifically in number of pipelines?

CB: "Pipelines" is actually becoming something of an outmoded term. The notion of a top to bottom rendering pipeline has been replaced by a shader engine with independent pixel shader processors that are decoupled from the rendering units - the functional elements that were traditionally described as pipelines. The X1900 XT has 16 of what might be called "extreme pipelines", with 3 parallel math execution units for every rendering unit for a total of 48 so-called "pixel shader processors ". The GT card has 12 sets of these math units for a total of 36.

AFR: Three operations per pipe per cycle.

CB: Actually these shader processors are capable of more than one
operation per clock cycle, depending on the type of operation, but
generally speaking we're talking about 48 or 36 shader operations per clock cycle respectively. This is a huge leap compared with previous generation GPU's where there was traditionally one pixel shader unit per pipeline.

AFR: What matters most in graphics speed, processor or memory?

next page: | 1 | 2 | 3 | 4 |