Re: Does This Kill The Mac Pro?

[Jeremy Garchow] "Alright, at this point I am going to need proof John-Michael. I think I fully understand where you are coming from, and I don't think the system operates in the way you are describing."

I think you're both right.

The PCIe bus was not intended to push data to the graphics card and receive fully rendered data back, but it is flexible enough to allow this to be done. I'd suggest that a 16x card may be important here.

I have heard that this was one of the big challenges that Assimilate ran into when porting SCRATCH to the Mac. On Windows, they use a NVIDIA daughtercard with SDI out, so the processed image data doesn't have to come back through the PCIe bus. On the Mac, they didn't have this option, so they have to push rendered frames back from the card to the computer. Guaranteeing realtime performance was apparently quite an engineering challenge.

Even if it's possible to render graphics on one card, pipe the rasterized frames over the PCIe bus in real time, and output them over another card, this would be lousy system design. You'd be building a large bottleneck into the system and soaking up several PCIe lanes for no reason at all.

Jeremy, check out the Thunderbolt process diagram PDF [link], page 3. It shows two ways that DisplayPort may be passed directly into the Thunderbolt controller (which I think is more or less what John-Michael is suggesting is necessary), preserving PCIe bandwidth for data instead of display.

Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog - What I'm thinking when my workstation's thinking
Creative Cow Forum Host: Live & Stage Events