I've done this with Max/MSP/Jitter, and we ended up doing something like this:
Use a high end machine and 2 triple Head to Go's to give you 6 outputs from a single GPU. This keeps all of your GL resources on the same GPU, which is important for throughput and being able to quickly reference the 'edges' of neighboring tiles and makes sure things are fast.* Due to texture size limitations we pre split the texture up from portions of the main image. I think our movie we were testing was 3k by ... 1024? or larger and max texture size was 2k on that system? We basically uploaded sub-portions to each output quad, with enough over-head to give us the maximum allowable overdraw/blending area. You can then run a pretty simple vignette shader to allow for the blending, and you are pretty much done. Its not too much work. For QC, you'd probably want to render the QCRenderer that is your 'scene' to a texure via a pbuffer or FBO, so you have an image to blend. We also were only dealing with video playback at the time, not live content, but the same ideas apply. Its not so bad. Try to figure out edge blending on a "wide" QC comp and a triple head, and you can quickly see its fairly doable without huge issues, and will be much simpler if you only need 3 outs. Maybe this is helpful? *of course with Cocoa/NSOpenGL you could have lower level access and use context sharing and upload to multiple gpus and manage this stuff, but it was not apparent in Jitter at the time how you'd go about doing that, so we just used 1 card to keep it simple stupid, we needed 6 outs. This was also on Windows which had its own share of headaches with jitter at the time. > Q1. > Whats the maximum number of DVI outputs I can reasonably pack into a single > computer these days and expect decent performance using QCRenderers? 4 x dual > DVI outputs (ie. 8?) - Can this figure be any higher? Any recommendations on > what cards I should look at? I believe Matrox's current drivers for OS X means you could have, my god, one triple head per DVI out on 4x GT 120s, which each has 2 outs, so thats.. 24 displays? Im willing to bet money you get issues going that high though. At least, I'd count on having issues :) > Q2. > Whats the maximum number of QCRenderers I can reasonably run on a single > machine? Or can I do something smart like have 1 renderer for the entire > composition and split it's output, post-processing the edge-blending and warp > distortion? I think not, but it's an idea. > lIke Chris said, I think this will be totally dependant on the context, and how many are drawing to the screen at once, if they are vsynced, etc. > Q3. > Would it be better to go for a distributed, multi-computer setup instead of > one-machine-does-it-all? I like QCV's approach of handling both effortlessly. > I think this will depend on the number of screens you want to drive in the end. > Q4. > Who on this list with the appropriate Cocoa/OpenGL/Wizardry skills might be > interested in being hired to help make something like this happen? At this > stage there's no guarantee this might happen, it's just a speculative idea > for now. However, drop me a line off-list if you think you'd be interested. > > Okay - that's all for now. Thanks for listening to my braindump! > > Best, > > > Ade. > > > > <ClayLogo.gif> > > Clay Interactive Ltd, Studio 3.1, 128 Hoxton Street, London, N1 6SH, UK. > http://www.clayinteractive.co.uk/ > > > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Quartzcomposer-dev mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/quartzcomposer-dev/doktorp%40mac.com > > This email sent to [email protected] _______________________________________________ Do not post admin requests to the list. They will be ignored. Quartzcomposer-dev mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com This email sent to [email protected]
