Please can you explain “amount of lines on 7900x is doubled”, 7900x has 44 lanes, can you use x16 on all pcie slots on your board?.

I have designare ex and if i add second m2 slot i loose one pcie express x4. Or if add third m2, i loose some sata ports. All are shared on gygabyte, is different on your asus?.

Actually there is a magical way, it's called a PLX chip. Martin referred to it in his z170 builds, where they bought a motherboard with plx chips to double the lanes or their 7700k to allow for more bandwidth. Just google it, it's not hocus pocus, it actually works fine and allows for more I/O in vMix :)

Martin also mentioned these, in this thread for example:

https://forums.vmix.com/...et-Issues-and-Discussion

To give you an idea: the blackmagic 8k capture card, which has 4 12G sdi inputs, only uses 8 gen3 lanes. This means you could easily stash 6 of them in an x299 board with plx chips, and have enough lanes left for your gpu and ssd's, etc..

Here are the specs of the bm card:
https://www.blackmagicde...klink/techspecs/W-DLK-34

And here you can read about the x299 ws sage board, which I use in x299 builds nowadays:

https://www.asus.com/us/...s/WS-X299-SAGE/overview/

The manual says you can run one slot in x16, while all the other 6 are occupied with x8 cards.

The bm quad 2, which will provide you with 8 HD sdi inputs, will only use 8 gen2 lanes. Gen2 lanes are only 8GB/s, while gen 3 lanes are 16 GB/s. In theory, if you would use exotic hardware to split your gen3 lanes in half you could stash 12 of these cards into a single system, which would provide you with 96 inputs.

The yuan 4x 4k p30 capture card only uses x4 gen.2 lanes. In theory, again with even more exotic hardware, you could split your lanes and take in almost 200 of these inputs into a single system.

I know theory is far from reality, I think going over 30 ish inputs will probably run you into troubles, I personaly never tested any system with over 25 (50i) in/outputs running simultaniously. Also when you are talking about this much bandwidth going over your pcie lanes, you should consider the bandwidth of your gpu, which is still limited to x 16 lanes, etc..

When we talk about using dual socket systems, Martin ever said that it was introducing a lot of extra latency and was not worth the effort. Same for using SLI or dual graphics cards. I would ever like to try it though, just to see what we're talking about. Hence I'm also happy that someone tried a threadripper system and published the results. For now I just don't think it's worth it (yet), when comparing to an x299 system.

So its not doubling lanes but slots. And since most cards use 4 or 8 lanes you get to use plenty of slots.

That sounds a lot more realistic.

Yeah, did some reading up on PLX, and its not multiplying any lanes. Its just doing traffic management.

Basically it uses existing lanes to funnel multiple PCIE devices into fewer lanes. You are still limited by the PCIE bandwith available on your CPU, but PLX allows you to connect more devices and by balancing properly you can get them all to work. In some specific setups you can get multiple 16 lane devices to work if their bandwith usage can be managed properly.

So in our case, one can get a motherboard with PLX controller to get more PciE slots to use the same x16 lanes, especially when we use x4 and x8 capture cards.
There are also stand alone PLX controllers that allow you to do the same on any motherboard.

I've seen Linus do something crazy with like 16 GPUs for virtual machines on a dual socket motherboard.
So no, this is not a trick specific to intel, its pretty much about PciE in general. This however does offer an interesting option to Threadripper builds, as the 2990WX has 64 PCIe lanes, and technically you can get a whole load of capture cards on that with a PLX setup.

Haha, if I have something down there Ill sure hook up.
Yesterday we did a test of our online show, and ran about 16 Vcalls and 6 other NDI sources at 20ms render time stable, on my 2990WX, with plenty of CPU to spare (avg 8% use). So for those workflows it feels great. I use a lot of composed multiview scenes in this show.

One thing of note for the TR builds is that memory speed directly affects CPU performance. This is because the speed of the infinity fabric is linked to memory clock. Thus a TR at 3600Mhz performs miles better then non OC speeds. So when building make sure to get the highest memory speed you can get that is certified for your motherboard.

Not getting any dropped frames in this setup. In my default preset (which is around 20 inputs) my render time stays under 10 ms tho. So its just the excessive use of multiviews and many many input sources that pumps it up. Basically, there's currently close to 100 inputs running. Of which around 22 are live sources. Im also using all 4 NDI outputs with their own scene in it.

This is a non issue for this show as long as there's no frame drops or audio glitches.

On a note about X299, both the WS X299 SAGE/10G and the not yet released GIGABYTE X299-WU8 have dual Broadcom PLX PEX8747 chips, allowing for:

x16/x16/x16/x16 or x16/x8/x8/x8/x8/x8/x8

That's insane actually.. I run at 6-10 ms on an 7900x machine, with 16 sdi inputs, 2 sdi outputs, 2 NDI outputs and 7 playlists with videos and pictures.

Greetings

I know this is an old thread.

But I ask basic question.

How are NDIs generated from DeckLink Card?

Thanks