Thank you very much for your opinion. My case may be a little different. My case is that I have 4 cameras (may be more) connecting NDI devices, OBS gets the 4 NDI signals and dynamically assign the 4 cameras to different combination of multi view scenes, user can choose different multi view scenes to be shown. So it may not be suitable to be done by physical capture.
My concern is that the 4 cameras are capturing the same object in different angles, but they get unsynchronized gradually shown on a OBS multi view scene.
I was thinking more about the code that handles NDI, having a single bug that causes both problems. The more ways the same thing manifests itself, the more detail there is to track it down with. But there's also the possibility of it being two separate problems.
Anyway, what you said just now, does give me an idea. There are multiple-input capture cards that inherently don't have a mutual de-sync problem. And the overall latency is also low enough to use for a live display through OBS. The one thing you have to be extra careful about, especially if you're sorting by price, is that it's cheaper for the manufacturer to use a single converter and a 4-input switch, instead of 4 separate converters. A single converter can only show one input at a time, but they're often sold as 4-input with no mention of that. A big red flag is if it focuses on security more than broadcast. Different use, different priorities, same search terms.
I used to have good luck with this, but it seems recently that they have problems with both quality control and compatibility with Linux, despite having a Linux driver on their website that says it's supposed to work:
Specifications Host interface: PCIe 2.0 X4 Transmission bandwidth: Up to 20Gbps Input interface: SDI*4 Compatible development interface: Directsound, Directshow (Windows) Output color space: YUY2 Capture resolution: Resolution within 1080p De-interlacing: Hardware de-interlacing + 3D noise...
www.acasis.com
Specification Size: 100*86mm Host interface: PCIe 2.0 X4 Input interface: HDMI*4 Output color space: YUY2 Video encoding format: H.264 Main profile (Hardware Compression) Capture resolution: Resolution within 1920*1080p 60Hz Supported resolution: 1920*1080p 60/50fps, 1920*1080p 30/25/24fps...
www.acasis.com
The first link is 4x SDI, and what I actually have. My two PTZ cameras have SDI out, and they run straight into the card. My other two sources convert from HDMI to SDI: one much farther away than HDMI will go by itself, and one right next to it because there isn't a 3x SDI + 1x HDMI option.
The second link is a 4x HDMI version of the same thing.
There's also this, which I haven't tried yet and looks suspiciously similar:
Buy Unisheen 4-Channel Quad SDI PCIe Video Capture Card,Ultra-Low Latency for Multi-Channel Live Streaming, Video Conferencing, Teaching, and Broadcasting: Internal TV Tuner & Capture Cards - Amazon.com ✓ FREE DELIVERY possible on eligible purchases
www.amazon.com
Buy UNISHEEN 4 Channel 3G SDI HDMI Video Capture PCIeX4 Card 1080p - Work in 4 Channel Simultaneously, Support Linux/Win11, Quad HDMI SDI PCIe Video Capture Card: Internal TV Tuner & Capture Cards - Amazon.com ✓ FREE DELIVERY possible on eligible purchases
www.amazon.com
Buy UNISHEEN Quad HDMI PCIe Video Capture Card - Work in 4 Channels Simultaneously, Support Linux/Win11, Built-in Capture Card: Internal TV Tuner & Capture Cards - Amazon.com ✓ FREE DELIVERY possible on eligible purchases
www.amazon.com
This one has a third option for 2x SDI + 2x HDMI, which might be useful in some cases. I wouldn't mind seeing the other two possibilities as well, but variation costs money to make and keep track of, so I understand if that never happens.
Both brands say they support multiple simultaneous cards, so you're not limited to just 4 inputs. I think the ACASIS one says you can put up to three cards on the same motherboard for a total of 12 inputs, provided of course that you have the slots for it. But be careful with that too. Just because it fits, doesn't necessarily mean it'll work. Each slot still has to be fast enough, which I found out the hard way and ended up grinding my case to use the one slot that is.
Anyway, assuming you end up with a bunch of simultaneous low-latency physical captures that work reliably, what's the difference between that and a bunch of NDI feeds? If the auto-switch or auto-composite logic in OBS happens after it gets a picture, then you can swap the sources and everything still works, right?