NDI Computational Overhead & Limits vs HDMI & Friends?

Just let it all out, buddy. You're among friends here.
Post Reply
CaseyJ
Met Resolume in a bar the other day
Posts: 11
Joined: Sat Nov 16, 2019 06:52

NDI Computational Overhead & Limits vs HDMI & Friends?

Post by CaseyJ »

Getting really pumped on being stoked about the NDI protocol from NewTek.

From what I gather Syphon and Spout are still the most efficient way to send video between applications on the same computer.

But! I am curious about what kind of limits there are for sending video over the network using NDI?

For instance, I love maxing out computers to see how many video outs I can reliably get. You know, multiple graphics cards, weird USB-C -> HDMI Dongles, and of course the famous Matrox TripleHead2Go's.

But! What if I could best my current records (x7 1080p@30 videos outs from a NUC Hades Canyon - running at 30 FPS with 3 layers clean BTW!!) and get say 10, or above, by using the NDI protocol? It seems like there is about 100 Mbps of overhead from each HDMI stream in NDI. So on a 1 Gb network and card we should be able to connect to at least x9 1080p@30 screens/projectors/etc right?

Now to the real meat of it... Without buying all of the gear is it possible to test this? Say send NDI signals but not actually have any of those decoders downstream? To test out the limits right? Can your GPU/CPU handles more video outs, more efficiently, than just native HDMI & DisplayPort IO by themselves?

Wow! If so that would be revolutionary! Finally a simpler cable that can do a lot more!

Just checking in... thank you for keeping this tool on the new new.

User avatar
SystemD
Hasn't felt like this about software in a long time
Posts: 113
Joined: Sun Jan 25, 2009 19:57
Location: Caen, France

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by SystemD »


CaseyJ
Met Resolume in a bar the other day
Posts: 11
Joined: Sat Nov 16, 2019 06:52

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by CaseyJ »

Yes exactly. Looks like you can have up to 8 in a 1Gig network.

But is that more efficient than a GPU video out?

What about adding another 1Gig network?

How far can we really push it?

Zoltán
Team Resolume
Posts: 4619
Joined: Thu Jan 09, 2014 13:08
Location: Székesfehérvár, Hungary

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by Zoltán »

With NDI it's best to have a single network connection.
8x 1080p@30 should be fine on a single 1GbE connection, but your machine also needs to be able to encode these streams in real-time. That's done on the CPU, and you still need to get the texture from the GPU to the RAM/CPU.

I don't think I can see that happening nicely on a NUC with 8 streams.
You can test this by opening Resolume on the sender, setting up the NDI outputs in the Advanced output. Then on another networked machines, load the streams into Resolume and play them. Keep an eye on the CPU usage on the sender, that could be your first bottleneck.
Theatre sound engineer, photographer, programmer, aka ravensc.
Control Your show with ”Enter” - multiple Resolume servers at once - SMPTE/MTC column launch
try for free: http://programs.palffyzoltan.hu
VJ footage: https://goo.gl/58oHy7

User avatar
SystemD
Hasn't felt like this about software in a long time
Posts: 113
Joined: Sun Jan 25, 2009 19:57
Location: Caen, France

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by SystemD »

I've made a little test with a rasperry Pi 4 and Dicaffeine, looks good : https://dicaffeine.com/

CaseyJ
Met Resolume in a bar the other day
Posts: 11
Joined: Sat Nov 16, 2019 06:52

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by CaseyJ »

Thanks Zoltan!

I was wondering if I could test this out. Do the computers have to actually receive the frames to simulate the load? Or can I get the same thing by just sending out the frames into nothing on my NUC as a test?

Lets say we have something more modest like 2 or 3 screens. Same 1080p@30. What is more computationally efficient? NDI or the native graphics I/O like the HDMI ports on the side of the computer?

When you graph both NDI vs HDMI Ports do the lines stay linear or is there some advantageous trade off at the low or high end for either? Does that line of questioning make any sense?

Zoltán
Team Resolume
Posts: 4619
Joined: Thu Jan 09, 2014 13:08
Location: Székesfehérvár, Hungary

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by Zoltán »

I was wondering if I could test this out. Do the computers have to actually receive the frames to simulate the load?
Yes ;)
CaseyJ wrote:
Fri Jun 05, 2020 02:36
Lets say we have something more modest like 2 or 3 screens. Same 1080p@30. What is more computationally efficient? NDI or the native graphics I/O like the HDMI ports on the side of the computer?
Direct output on the renderer card is the most efficient. Anything else comes with an overhead.
CaseyJ wrote:
Fri Jun 05, 2020 02:36
When you graph both NDI vs HDMI Ports do the lines stay linear or is there some advantageous trade off at the low or high end for either? Does that line of questioning make any sense?
Comparing NDI with HDMI is like comparing turtles to carrots. If you can go HDMI to your display without converters, then I'd recommend that.
Theatre sound engineer, photographer, programmer, aka ravensc.
Control Your show with ”Enter” - multiple Resolume servers at once - SMPTE/MTC column launch
try for free: http://programs.palffyzoltan.hu
VJ footage: https://goo.gl/58oHy7

CaseyJ
Met Resolume in a bar the other day
Posts: 11
Joined: Sat Nov 16, 2019 06:52

Re: NDI Computational Overhead & Limits vs HDMI & Friends?

Post by CaseyJ »

Thank you very much this has been an informative discussion and I appreciate the conversation!

Post Reply