crayzcrinkle
Member
Hi there, this is something I think would be a very good idea for obs...
We all know you can now set the exact GPU number in settings which is used to encode (e.g. in CUDA) a video. However, from testing, I have noticed that OBS will still use an auto-selected "main" graphics card to composite the frame (e.g. what is seen in the live view). This leads to both GPUs being used by OBS related procedures. This is especially important in the case of VR streamers where every % counts in maintaining a smooth stream. Especially in view of framerate of broadcast.
I have a Vega 64 main GPU and a GTX 750 TI (2nd , encoding) CPU setup.
When encoding a "intro screen" looping animation @ 60fps, the 2nd card is used (according to MSI afterburner) 3-4%. During the record, the Vega is still 15-27% used. This usage is due to OBS forcing composition of the scene elements on the main card, not the 2nd gpu. Closing obs during the testing, lead to the main GPU usage dropping off from 15-27% to 0-5%, a significant impact.
The reason this is so important is that 2nd card is hardly used. A GTX 750 is not as powerful by any means as a Vega 64, but could probably take it with some fine tuning of settings. Reducing the video from 60fps to 30 fps reduces the main card usage down to 4%-15% but for VR, this simply isn't feasable, as you need as least 60fps to show fast motion and movements clearly.
Therefore I propose that we should be able to move not just the video encoding during stream/record to a second card, but also the entire composition process in OBS.
We all know you can now set the exact GPU number in settings which is used to encode (e.g. in CUDA) a video. However, from testing, I have noticed that OBS will still use an auto-selected "main" graphics card to composite the frame (e.g. what is seen in the live view). This leads to both GPUs being used by OBS related procedures. This is especially important in the case of VR streamers where every % counts in maintaining a smooth stream. Especially in view of framerate of broadcast.
I have a Vega 64 main GPU and a GTX 750 TI (2nd , encoding) CPU setup.
When encoding a "intro screen" looping animation @ 60fps, the 2nd card is used (according to MSI afterburner) 3-4%. During the record, the Vega is still 15-27% used. This usage is due to OBS forcing composition of the scene elements on the main card, not the 2nd gpu. Closing obs during the testing, lead to the main GPU usage dropping off from 15-27% to 0-5%, a significant impact.
The reason this is so important is that 2nd card is hardly used. A GTX 750 is not as powerful by any means as a Vega 64, but could probably take it with some fine tuning of settings. Reducing the video from 60fps to 30 fps reduces the main card usage down to 4%-15% but for VR, this simply isn't feasable, as you need as least 60fps to show fast motion and movements clearly.
Therefore I propose that we should be able to move not just the video encoding during stream/record to a second card, but also the entire composition process in OBS.