[Cin] Render farm - client nvenc and main none, can this work?

[Phyllis Smith]

Discussed this with GG and he provided the following tips based on a video
that plays in 1 hour running at full speed.  In his final render on an AMD
desktop computer with many threads, he breaks this up into 20 threads using
a local render farm (just the 1 computer) so it only takes about 5 minutes
to render the whole thing  -- in your case on your Ryzen desktop with 16
cores a smaller number would be good.  such as 16 maybe.  One thing he said
to be sure to do to optimize this is to "Reset rates" in the Render Farm
menu so that the jobs are evenly distributed.  In deciding how many threads
to use, he suggests looking at the output of the linux command "top" to see
how much cpu is being actually utilized on a typical render.

Recently however the render times have been becoming a problem. Also, I now
> have a second machine (desktop) with an AMD Ryzen 5 3600 and an nVidia
> 1660. I've set up renderfarming and it seems to work fine (my one and a
> half hour render is down to half an hour). Now I'm wondering whether that
> nice graphics card can be used (without me having to actually sit at the
> desktop, I'd much rather work/edit on my ultrabook with all its attached
> monitors) for editing.
>
I am not quite sure if I understand this correctly, but I would do all of
your editing on your laptop first and when time to render just move the
project to the desktop and do the final render there using the render farm.

>
> It seems whether nvenc gets used depends on whether I select h264-nvenc
> (or h265-nvenc) on the main machine, but obviously I can't do that on this
> particular main machine.
>


>  Is there a way to get this to work, or am I forced to choose between
> software renderfarm and single-machine nvenc (in my specific situation, I'm
> assuming that if I had two nvenc-capable machines this would work fine). Or
> is there some cuda-related speed up I could use?
>
Although I can not test whether a software render farm or a double-machine
nvenc is faster, gg thinks that a localhost render farm would be faster
than using 2 nvenc graphics card enabled computers because there would be
contention on the path to the graphics cards.  It could be slower to have 8
threads using 2 cards than having 16 software threads.  There does not
appear to be a cuda-related speed up available as far as I know.

Perhaps, I have confused this answer.  If so, please reply.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20201022/d458ee83/attachment.htm>