Touch setup 4 HD Outputs & 4 Kinect inputs

Hi All!
I am about to setup a shopping list for an installation using 4 kinects,

Technical Aspects:
-our output is to 4 projectors, 2 by 2 on 2 walls of 18*40 meters each.
-We would like to use 4 kinect v2 sensors
-run a fairly intense space scene using particles, object instancing feedback comp effects etc…

We are now doing the shopping list for hardware and we have a few options:

1:
One 2x1080 GTX (with SLI) , 4.16Ghz 4-core i7, 32G RAM workstation who will run the project + output to 4 HD Projectors.
Four Mini HP EliteDesk Computers who will sent both kinect depth TOP and CHOP over network to the main computer.

2:
Two main computers with ONE GTX1080/TITANX who would run the same project with clock sync and render different Outputs who would have ONE kinect plugged directly and one over network using touchin nodes, each will render only 2xHD & will output to 2 projects.
Two Mini HP Elite Desk who will only process the kinect data and send over network.

Option 3 is to send one HDMI signal of 4K and use 2xDouble head video splitters for our projectors, if better than plugging directly to graphic card…

Which way to go about hardware in this case is preferred?
/
Do you have any other recommendations for hardware setup who might be useful to us?

Thanks in advance,
Ronen.

I would go with 1, but only get a single GPU. SLI will not give you any benefits and just complicates the setup.
If you went with two you’d need Quadro cards since there is no way to sync Geforces together.

Hard to say without knowing what kind of content you’re going to make and how efficiently you can make it.

My suggestion is to get:

Main system

  • fastest i7 quad core you can afford
  • 32gb is fine (honestly prob fine with 16gb too)
  • 1x small SSD for the OS
  • 1x large PCIe NVMe drive for all your media
  • 1x Quadro M6000 or M5000 (the P series is out but i’ve heard its an architecture change the nullifies ability to use Sync cards currently, so probably best to steer away from that for now unless you have time to test and ability to return for an M series if something isn’t working right)

Kinect computers

  • fast i5 or i7 (Windows has some “minimum” specs they recommend which I’ve heard you can get away with quite a bit lower)
  • small SSD just to make the computer run a bit snappier when working
  • if you can get one that has even a weak geforce chip like the 950M or something it will allow you to offload some of the sensor processing onto these machines and then you’re just sending processed chop channels back to the main computer

I dont know much about the HP small boxes, but Zotac has quite a few nice ones as well as the Gigabyte Brix.

I don’t know how proficient you are with TouchDesigner, but I recommend to most beginner - experienced intermediate developers that they’re better of cutting features before moving to sync systems or multi-system render setups. Keeping the timings locked in can add a bit of overhead that you probably won’t have bandwidth to deal with unless you’re pretty experienced in dealing with everything else you need to in TouchDesigner.

Hey Guys,

Thanks for your answers, useful info!

Malcolm,
If Quadro’s are the only cards that can be SLI’yd and really work together as one unit for touch, than Is it possible to run 2 instances of touch on the same computer using different GPU’s? and if so, how?

Elbuz,
The Content is basically creating space environment, at the moment i use hundreds of instances, 2K particles and render twice, one doubleHD from a front camera, and the other half same res with the rear camera, there for splitting the scene into 2 renders looks like something logic to do, unless this not use the GPU at all in touch?

This is my output & network at the moment (see attached), running @ 40-50fps on a single 1080. but I want to know my limitations now as for how much more stuff i can do, and if 27 players from 4 different kinects would affect this 3D scene in some way at the same time still have some room for the GPU and run @ high fps…

I like the idea of doing some processes on the mini computers who sensed the kinect processed CHOP info but i also need the depth top for silhouettes.

What will we benefit from if we switch to a single quadro card vs 2 1080’s (either on separate machines or, on the same one)?

Thanks again,
Ronen.


over1.tiff (16.4 MB)

Quadros in SLI will also not give any performance benefit. The only ‘SLI’ benefit you can get from Quadros is the ability to SLI Mosaic them which gets you more outputs. However that is slower than a single Quadro as they are not dividing the workload, but rather both doing the same work + extra overhead which gives you a slower result. With the DataPath Fx4 devices available now it seems much less useful a case.

You can have two Quadros in a system though running two different instances of TD. That is done using GPU affinity which is outlined in the wiki http://www.derivative.ca/wiki088/index.php?title=Using_Multiple_Graphic_Cards

Awesome!

Thanks Malcolm.

This sounds like a mega project. :smiley: Please share or post some of the results when you have them, would love to see what you’re up to!

2x 1080 wont do anything in 1 machine. 2x 1080 in 2x machines will give you 2 machines you need to keep in frame sync, which you can’t get down to less than a frame with just GTX cards. Also you wouldn’t be able to frame sync 2 particle systems on 2 separate systems since frame sync generally means everything needs to be more or less deterministic.

The best way to test your overhead is to push your machine till it starts dropping performance. Everything other than that is just an educated guess. In your network if you see any SOPs cooking (check the wires to see if they’re animating between nodes), you’ll probably run into scaling issues, otherwise you should be ok for the rest of what I read quickly of your description. Honestly really comes down to how you do it. In TouchDesigner, you can do the same thing 10 different ways but only 1-2 of those ways will scale efficiently for real-time performance.