3 ms·
The bottleneck is not on the CPU. I'm afraid this company may have wasted their time trying to reinvent WebRTC. If you really want to get realtime video, I thin
by zelly 6y ago
The bottleneck is not on the CPU. I'm afraid this company may have wasted their time trying to reinvent WebRTC. If you really want to get realtime video, I think the best approach is a custom codec on CUDA or better yet custom hardware (FPGA). You can only go so far on general purpose hardware before you hit a wall and get Zoom/WebEx quality.
- busrf 6y agoCan you recommend some resources for the current state of the art for low latency video? Somebody else in the comments posted https://github.com/CESNET/UltraGrid/wiki https://github.com/CESNET/UltraGrid/wiki, but I’m curious to learn more.
- ronyfadel 6y agoIs or is not? I’m confused: if the bottleneck is not on the CPU what does CUDA solve?
- zelly 6y agoThe bottleneck is the video encoding/decoding/rendering, which is done on GPUs to begin with. Of course if it were done on CPU instead, then it would be significantly worse, but that's not where we're starting from. Improving stuff on the host side by, say, rewriting WebRTC in Rust won't improve the latency of your video by much or at all.