2019 MAC PRO Multiple output low frame rate

[wongkachun48]

Hardware:
Mac Pro(2019)
macOS Catalina 10.15.7
3.5Ghz intel Xeon W
48GB RAM
AMD Radeon Pro W5700X 16GB x2

Video have already change to DVX.

I'm now trying to have 8 NDI output(1080p per output) for my project, but the frame rate is surprisingly low just around 15fps.

The biggest thing concerns me is that my CPU only using 20%,and the GPU is 50%, it should handle better frame rate, but it didn't.

It's there anything I should take care of the system to get better frame rate?

[Arvol]

NDI output is coming from your CPU (it's GPU to CPU, but the CPU is doing the encoding). Your Xeon should be more than capable for 10 streams. (It's 250Mbps for a 3G stream btw).
Have you tried sending 10 NDI steams from any other software to see if it's a NDI issue or an Arena issue?
I was thinking about building a NDI stress test tool with benchmarking, but I won't be able to get around to that until around November when I'm finished traveling for the year.

Is your power management et to full power? I'm not sure if that's an OSX option or not?

[Arvol]

I had a free min:

https://drive.google.com/open?id=1MQA67 ... p=drive_fs

Requires TouchDesigner. No clue on how accurate this is btw, on a festival site with a few extra min. Could be way off? I'd monitor Wireshark and your system resource monitor for the most accurate results. Also, free version of TD limits your resolution to 720p so keep that in mind.

[wongkachun48]

Thanks for the reply Arvol, yes I have used other software to send out more than 8 NDI output,and they are all going smoothly.

And the power of the Mac is already set to highest performance as we speaking, but I'll do another check just to make sure.

Hope it's something that I miss check.

And I'll try your benchmarking tool later.

Really appreciate for your help.

[wongkachun48]

Update:
Even if I use HDMI output, the maximum screen my ARENA can use is 8screens(control screen included), and the frame rate is also around 15-17

[Zoltán]

Are you using a 10 Gig network?
8x NDI streams require a lot of bandwidth.

[wongkachun48]

Hello,Zoltán

Yes, I'm using a 10gig network. CAT6 cable, NDI box from Newtek.

[Zoltán]

My threadripper windows desktop can send 8x 1080p NDI outputs to 8x receivers (2 different computers) with an Nvidia 1080 Ti. Playing one dxv clip at 38 FPS.
1 Gig network 90% utilisation.
24% average CPU usage, 63% GPU (copy) usage.
Seems like a lot of texture copying to CPU, but no idea why that would be an issue on a mac.

[resolumejunkie]

To help provide another anecdotal piece of data: on an M1 Mac, with 8GB of unified memory, I was able to get (6) 720p24 NDI streams out of Resolume Arena (132,710,400 pixels per second). On an M1 Mac, with 16GB of unified memory, I was able to get (9) 720p24 NDI streams out of Resolume Arena (199,065,600 pixels per second), or (4) 1080p24 NDI streams out of Resolume Arena (199,065,600 pixels per second). Not sure what the bottleneck is on more performant systems. I would think a Mac Pro with a 3.6x more powerful GPU would provide more than a 25% performance boost. Something does seem amiss. Perhaps the information I have provided will assist in some way.

Lastly, I could also try using doing some benchmarking on the base version Mac Pro w 580x I have sitting here sometime as well, as a point of comparison. Although, not sure how soon I can do that.

[resolumejunkie]

GPU is 50%, it should handle better frame rate, but it didn't…anything I should take care of the system to get better frame rate?

When you say “GPU is 50%” …how is that calculated? As published here:
https://resolume.com/support/en/lots-of-outputs

“Single computer with two GPUs This works in most cases. Keep in mind… having two cards doesn't mean you can do twice as much. Quite the opposite…when using the outputs of the other card, the texture is first rendered on the rendering card. It then has to be transferred down the PCIe lanes of the rendering card, via the motherboard, up the PCIe lanes of the other card, and then sent out the ports. This will always result in a performance hit. The more cards you add, the bigger this hit will be.”

So… the first thing that comes to mind is that 50%. Is Activity Monitor showing the first W5700X at 100%, the second W5700X at 0%, and averaging them together for “50%”? What do the individual GPU performance / utilization graphs show in Activity Monitor when using 8 1080p HDMI outputs?