[Maurice Calvert]

I don’t know how to thank you enough, I was getting to the end of my tether.

The mistake I made was not thinking about the amount of data. A quick calculation led me to this, which I hope others may find useful:

The video I am working on is 102 minutes, which from the above means that the raw data shot was 338Gbytes if it were truly uncompressed.

I converted my 19Gbytes of source with Neoscene, which produced 123Gbytes of (still quite compressed) data. This confirms Neoscene’s estimate of better than 3-1 compression ratio. On my 8-core machine, it took 75 minutes, with about 6 of the 8 cores busy on average.

Let us assume that 60% of the work was decoding AVCHD, 30% encoding Neoscene (it’s suppposed to be relatively simple) and 10% overhead for I/O.

60% of 75 minutes = 45 minutes to decode 102 minutes = 26 seconds of CPU per minute decoded.
6 cores at 2.5Ghz = 15Ghz. 15Ghz * 26 / 60 = 6.5Ghz to decode in real-time.

Because of its structure, AVCHD cannot be decoded in parallel in real-time (see P.S. below), thus to preview AVCHD in real time, you need a processor with 6.5Ghz cores, which AFAIK is currently impossible due to the laws of physics.

All this goes to confirm what many have discovered the hard way, and despite fanciful advertising by the software makers, it just can’t be done.

P.S. I appear to contradict myself, as Cineform decodes in parallel. It does so by cutting the video into pieces (of X seconds), decoding the pieces in parallel, and re-assembling them afterwards. Obviously, you can’t do this for real-time preview.

Needless to say, I stand to be corrected…

HTH and many thanks again!

[Maurice Calvert]

Thanks.

I didn’t want bore with all my hardware. I use an 80Gb SSD (it’s not a disk, it’s memory that behaves like a disk, with a seek time of 0 and a huge data rate), as a work drive dedicated to the source videos. The render output is directed to a separate WD Caviar 15’000 rpm SATA drive. The operating system is yet elsewhere, and with 16Gb, I’ve disabled all paging, no need.

In a nutshell, I’m praying it’s not the disks, as there’s little else I can do to improve things >;-)

[Maurice Calvert]

Thanks for the quick reply.

I’ve indeed tried rendering to RAM. I set aside 10Gb and it took 3 minutes to fill it with 1 minute of video.

The suprising part was that viewing from the rendered RAM was just as choppy (I was of course careful not to change the preview window size). I conclude that the chopiness is not due to rendering, but what could if be?

I’m sure something’s glaring me in the face, but I can’t see it >;-)

Any ideas?