18 Sep
2014
18 Sep
'14
11:07 p.m.
Am 19.09.2014 um 00:42 schrieb LESLIE HARTMIER:
I certainly get the whole commercial thing, but
it's far more fun to do it our way! In the end, we have commercial options that we
don't want to use. Our TMS of choice even contains a DCP creator (CineDigital
Manager), but we liked DVD-o-matic‎, and we like DCP-o-matic, so we use that!
Thanks to both Carl and Carsten for the replies. See ya on Film-Tech (and on this mailing
list!)
Alright, in that case I would opt for a network of e.g. 5820 machines, overclocked. I
never quite understood why a dual 8 core Xeon running 48 threads maxes out at 16fps. They
guys contributing to the benchmark list confirmed all cores were running near 100%.
Results will vary a bit depending on the type of source you will start with. e.g.
uncompressed single frame sequences will benefit more from a fast disc (SSD) than
precompressed MP4, DNxHD/ProRes, etc. Aside from the mass storage reading throughput, I
also think uncompressed source will be beneficial for very high thread number encoding, as
uncompressed source will probably have the lowest processing need in the pre-processing
stage.
Also of course, the more threads you use, the more RAM you need.
I would love to see the 20fps line being taken, by any single machine or networked config.
Getting near realtime would be nice.
In my network test, a single outdated W3530 was able to process/feed 17fps. I guess with
a modern six-core i7, the master process would not be the bottleneck. I could see some
network saturation from 15-16fps. You may still benefit from 1GigE until 20fps I would
assume, as long as Carl does not switch to 12Bit. I know Carl is thinking about sending
compressed bitmaps to render nodes in order to ease network load, but that compression
will also take time. Hard to say where this will come out. Always interested to learn
about your results.
- Carsten