Hi
Are there any bottlenecks on PCIe v4 on higher resolutions past 1080p?
Is it better on PCIe v5?
Does anybody know if OBS will finally support multiple GPUs?
Like you game on one GPU and encode with the other?
Best 
LG can already do 1080p at PCIe Gen 3 x8, doing higher resolutions do not require PCIe gen5.
Yes, it has done so for years. Just select the GPU you wish to encode with in the settings.
Last time I tried that - it worked horribly and people from OBS forum were strongly advising against this practice.
Using an iGPU for encoding is one thing, but in case you have a secondary GPU in your system - most likely you will be using a chipset PCIe port, cuz your CPU port already occupied with another GPU. That means you will be sending a constant raw uncompressed video stream via DMI which may introduce a bottleneck.
In my case I had RX7900XTX as primary and wanted to use ARC A380 as AV1 encoder for 4k stream
@Hrybochok please pay attention to the category we are in.
-
LG already does this raw uncompressed transfer between two GPUs, it’s required due to how it functions.
-
The transfer is from system ram into the GPU, not GPU to GPU, as the frame LG captures from the guest is stored in system ram as a bounce buffer.
-
If the encoding GPU is an iGPU, then it accesses the ram directly using DMABUF, there is no additional transfer.
-
If the encoding is to be done on a dGPU and the iGPU is being used for display, again, there is no additional bandwith consumption on the PCIe bus when compared with a LG dGPU to dGPU configuration. The iGPU still accesses the captured frame already in ram directly, no additional copy.
I along with quite a few others in the LG community use a ARC series GPU for AV1 encoding to stream their guest VM via LG, at 4K. I personally stream it to restreamer (on the same host) that then again re-encodes for Twitch while passing the AV1 stream to YT.
2 Likes
Apologies, for some reason I thought that we are talking about generic dGPU to dGPU encoding in OBS without using Looking Glass