SLI in x16/x8

[Alex Ushigome]

Heyo all, 

 

Planning to buy a 3950x and 2x 2080 Ti, and I have noticed that the Gigabyte boards run in x16/x8 instead of x8/x8, and was wondering if there are benefits/issues with running SLI NVLink in lanes at different speeds. I'm looking to use OBS, and I've read that it needs more lanes to not stutter. Any information is greatly appreciated.

[Windows7ge]

There's no benefits but there's no issues either. What motherboard is it specifically btw?

[Mick Naughty]

So you’re gonna run sli, which means the cards will be in 8x/8x no matter what you do. 

 

Dont see how thats gonna effect obs overall. I assume you want to use the gpu to encode if that’s the case. 

[Alex Ushigome]

37 minutes ago, Windows7ge said:

There's no benefits but there's no issues either. What motherboard is it specifically btw?

X570 Aorus Xtreme, Master and Ultra, all 3 boards say 1 slot x16 and 1 slot x8.

25 minutes ago, Mick Naughty said:

So you’re gonna run sli, which means the cards will be in 8x/8x no matter what you do. 

 

Dont see how thats gonna effect obs overall. I assume you want to use the gpu to encode if that’s the case. 

I read that OBS needs the bandwidth, as it copies the frames from the GPU, may just have been rubbish information that I read.

[Mick Naughty]

Well I can’t really help in that regard. I was running sli on both of my pcs but using one to record the other. So I don’t think bandwidth ever because a factor. 

 

May be better off using shadowplay. 

[Windows7ge]

14 minutes ago, Alex Ushigome said:

I read that OBS needs the bandwidth, as it copies the frames from the GPU, may just have been rubbish information that I read.

Take what I say with a grain of salt but I do have high doubt that this would be the case. Most of the processing is done on the card the GPU itself doesn't actually need a ton of bandwidth for sharing information between itself and the CPU. An oversimplified example is it's kind of like you giving me a math problem to solve. I don't need to tell you every step I do along the way I just need to tell you the answer in the end. That uses a lot less bandwidth than say a PCI_e SSD where the sole purpose of the device IS to move large amounts of data between itself and the CPU/RAM.

 

PCI_e x8 Gen 3 is very close to 64Gbit. To copy the frame buffer and stream it should not require anywhere near that number. Not even close but again that's more of an educated guess so wait for someone to verify or deny what I've said.