11 Jun
2013
11 Jun
'13
10:19 p.m.
Am 11.06.2013 um 16:52 schrieb J-sun Bailey:
I'm new to using DVD-O-MATIC, it's a great
product, I'm running it on an older system:
Windows 7 32bit
Intel P8600 2.4 GHz CPU
4 GB RAM
It's taking between 2-3 hours to convert a 15-30 minute movie/short, Is this average,
I'm looking at upgrading to a quad core i7 processor with 32GB of RAM running Windows
7 64bit. I'm guessing this should speed up my conversion time quite a bit, but curious
what others are seeing before I invest the money in this.
Your numbers are not too specific, but they compute to something like 2-4frames per
second. Yes, that is 'normal', not even bad for an old machine. Have you optimized
your thread setting (I think it should be set to '2')?
Do you have some spare machines sitting around?
We had a few discussions regarding encoding speed and benchmarking. I think no one posted
numbers for an i7 so far.
Two quotes:
---
My quad core i5-3550(a)3.3GHz converting a dvd-clip with
no extra filters clocks in at 3-3.5fps. LinuxMint 64-bit.
Adding my HTPC as an extra encoding server converts the same clip at 5-6fps. It runs
adual core i3-2100(a)3.1GHz with HT, so 4 threads. Win7 64-bit.
This seems to suggest that HT helps with DVD-o-matic. But I'm not certain. Is it
better to run your CPU with HT off and use only 2 threads, vs having HT on running 4
threads. My gcc-compiles at work do not care for HT.
---
AMD dual octo-core (true eight physical cores - not bragging, just clarifying) at 4.0 GHz
and using 15 cores under Win7 x64 nets me ~8 to 9ish fps consistently. I keep considering
making my notebooks (two quad-cores and a dual-core) join in, but my single machine does
okay.
As for HT or not HT, despite the fact that they're a single, dual or quad physical
core processor, even with the overhead subtracted for treating it as a dual, quad or octo
virtual core processor still adds 30% performance, which though not a substitution for
physical cores, are a welcome thing, I imagine.
---
Bear in mind that the j2c encoding is the most CPU intensive task, and that turned out to
be highly dependent on image complexity:
---
I just did a test comparison between my frame count test video (white numbers on black)
and a (static) synthetical colour noise pattern (noise at the single pixel level). Both
MP4, 1998*1080 24fps with stereo audio, all transcoding parameters identical, no
additional processing/filters enabled.
While the very simple frame count images compress at up to 3.6 fps in DVD-o-matic, with
the noise pattern it drops to 0.7fps on the same machine. That is a 1:5 difference between
best and worst case content!
---
A quick test encoding of the SINTEL Trailer shows that the encoding speed runs up and down
between 1.1 and 4 fps - so it is unusable as a benchmark. At least if you want to
benchmark the fps display during the encoding and not the total encoding time (which is
12min24s = 1,68fps total average).
---
- Carsten