Need to consolidate two Nimble servers I have into one. Wondering if I should go for more CPU or GPU? Here are my transcoding scenarios:
EACH server currently does the following:
Incoming RTMP 1080P
ABR transcode for 1080, 720 and 360p h.264 HLS
Fallback transcode (still image slate in case stream goes down)
Incoming emergency RTMP preparation (to override original rtmp)
2-hour ABR DVR
Outgoing to CDN
So the new server would do double that.
Currently each server is a single Xeon E5-2630 v3, and the CPU usages runs at about 35%. I tried to put them both onto one machine, but instead of doubling CPU usage, it jumped to 98%.
1) Does that CPU usage jump seem wrong when I move everything onto the one server?
2) As I am consolidating onto one server, would it be better to just do a dual (newer) Xeon, or should I get a GPU?
3) If GPU, my options are: Tesla P4, Tesla P40, Tesla P100 or Titan V. Which would be better-suited?
Thanks for the advice!
1. I cannot comment on your CPU usage, as close inspection and monitoring of the system's load, the number of incoming, outgoing, transcoded streams and client connection number 'before and after' is needed. Have you analyzed, what exact Nimble's threads lead to CPU exhaustion, was it Nimble at all?
2. GPU Transcoding will always be more efficient than CPU Transcoding.
3. Unfortunately, we short of resources to benchmark GPU to recommend to you.
We have tested Tesla M60 and the results can be used as reference. Please find it at the following documentation page:
Transcoder allows using the NVENC-only pipeline on Ubuntu 18.04 now and it should bring a huge improvement to CPU usage as filters (e.g. scale_npp) can be processed on GPU. Please find more information on the following documentation page.
Please ask your questions using our support page (https://wmspanel.com/help), as forum may reveal some of your company's sensitive information.
and required to achieve the purposes illustrated in the
If you want to know more or withdraw your consent to all or some of the cookies, please
refer to the