Encoding server does not seem to make much difference :(

[funkytwig]

I built a new desktop and did a little testing, in general CPU tests I got around a 30% improvement over my laptop. I have made a DCP and it was a bit faster but can't remember exactly.

So I thought ide use my laptop as an encoding server (both have 2.5G ethernet plugged into a 2.5G ethernet switch).

The thing is the server is only showing around 3.5ftp, which is a lot less than I thought it would get (the laptop used to do, from what I can remember, at least 15ftp when it was the main encoder).

Any ideas, here is a bit more info about the computers:
Main: AND Ryzan 7 7700, 32GB Ram, 2.5G Network, CPU around 90%
Server: AND Ryzan 9 5900HS, 32GB Ram, 2.5G Network, CPU around 20%

[Kewl]

Things to check: number of cores used on client and server, bandwidth of network. Are the client and server connected to a switch or router? Maybe there's optimization to be done there. I don't know if it's possible on Windows, but on macOS, connecting directly, using Ethernet, two computers will create an ad hoc network that can be faster than connecting to a switch or router.

In my experience, if the network gets saturated (the send throughput from client to server being at the limit), the FPS declines. So, sometimes, the answer is to use less cores on the server side to desaturate the network.

[Carsten]

You should definitely get much better results from these machines paired up.

As Kewl says, first check the number of encoding threads in prefs on both machines (the number of hw threads to use by the encoding server is set in DCP-o-matic main on the notebook, and the number of encoding threads used locally by DCP-o-matic main on the workstation).

You can verify these settings by using the ->Tools->Encoding servers window on the workstation, and the encoding server status display.


Both machines should have a least 16 threads set, maybe 20-24 will give some small improvement. Don't overdo it.

[funkytwig]

Thanks, both are connected to the same switch and have 16 threads.

[Kewl]

OK, I would find a utility to monitor the network throughput to the server and optimize the number of threads on the server. Start with, say, four threads and raise the number while checking the FPS to find the sweet spot.

[Carsten]

Are you testing it with 2k or 4k DCPs?


I suggest you try the 'official' SINTEL Benchmark so we can compare results.

https://dcpomatic.com/benchmarks/