Can NVLink share Spout across 2 GPU's?

We currently use TD to record 8 video inputs to an internal NVME drive, and then pass those on to Resolume Arena to do live-switching for both our video wall and live-stream output. If both Arena and TD run on the same card, we can use Spout, but the GPU hits its ceiling and frames drop.

Is it possible to shared Spout across the cards if we were to utilize an NVLink bridge and assign the programs to separate cards?

What codec are you recording the video inputs into?
Spout uses DirectX under the hood, so in theory it could be used to do multi-GPU work, but Spout does not currently support this.

I was a bit inaccurate in my description. We simultaneously send each SDI VideoDeviceIn source to both a moviefileout as well as either a SpoutOut or NDIout. Each input flow is running within its own touchengine, and while we have some questions about potential bottlenecks and system optomization, those will probably need their own topic.

The touchengines all look like this:

There’s no specification within the VideoDeviceIn for codec, I’m not sure if it applies or not?

Thanks

Honestly this is a very GPU light network, and the overhead involved with cross-GPU transfers (which is very high) would likely be worse than what you are currently doing here.
Are the TouchEngine’s running within Resolume in this case? Or are you running TouchDesigner instances?

I didn’t even think to search if Resolume had touchengine support. Perhaps @djevo1 's FFGL plugin would allow Resolume to record all 8 video sources?

We have all 8 touchengines running in one TD instance. each engine has a corresponding “arm” button, and references both a text dat for filename and a “record” button to start recording. Would it maybe run better a different way?

But really, if we could run everything inside TD that would be the dream; Resolume likes to run both GPUs (rtxa5000’s) and every CPU(TR Pro 5965wx) with no ability to manage resources - like TD’s GPU affinity and touch engines - among other issues and limitations.

Maybe the system just can’t handle the load? We tried to spec the workstation to do everything we’re trying and more, and with a little latency as possible (ideally not separate machine)…
Asus WRX80E
256GB Memory
2 x RTX A5000
Threadripper Pro 5965WX
Recording to a Samsung 990 NVME