(Page 1 of 1 pages for this article )

Filed under:

New ProRes info - Data Rate Chart

Apple posted a handy white paper on the new ProRes codecs, and there’s some good stuff in there.

I went digging through the ProRes white paper that Apple posted with the release of Final Cut Pro 7, and found some useful info (which I’m betting they won’t mind m highlighting here). Read on.

First, I got it wrong originally about the ProRes4444 codec - alpha channels are supported up to 16 bits instead of the 12 I thought (and may have posted elsewhere). They are losslessly encoded, so they will match EXACTLY from the uncompressed alpha source. Not visually losslessly, mathematically losslessly. No argument - it matches. That ProRes4444 handles up to 12 bits is great - a minor point of quibblage - since it is compressed, it can be difficult to determine whether all 12 bits are accurately recorded/encoded since there is some loss of precision in a lossless codec. In practice, for shot-in-the-real-world material, this won’t matter, since every camera that generates 12 bits has more noise than signal in those last bits. Put another way - there is so much noise in the signal, recording 12 bits of it really doesn’t give you a meaningful benefit. (MANIPULATING those bits in a 12 bit or greater color sandbox, however, IS important!!!!)

Secondly, this useful chart to give you an idea of how many Mb/sec or GB/hour your footage takes up. It was stashed back in the Appendix, but considering I was about to test and generate this info, this just saved me a day:

continued on the next page:

Note that these datarates are in megaBITS, not megaBYTES,  per second - 480p24 is 10 Mb/sec=1.25 MB/sec - that’s about a third of DV’s data rate, so only 4 GB per hour.

ProRes422 Proxy for 1080p24 is about 4.5 MB/sec - just a bit more than DV.

When you’re ready for full size, ProRes4444 (without alpha) is about 33 MB/sec for 1080p24 - VERY doable on a single modern hard drive.

Wanna do 2K files? 2Kx1080 at 24p for ProRes4444 is about 38 MB/sec, 2Kx1556 (anamorphic film scan sized) is about 57 MB/sec - still doable if you have a new fast drive - but only single stream, no realtime cross dissolves without a RAID.

Secondly, how much does it crunch the image down? This chart shows it pretty clearly:

There’s further info about the PSNR of ProRes422HQ (Peak Signal to Noise Ratio, that means how good is the quality/fidelity) compared to D5 and DNxHD, showing (according to Apple) that ProResHQ is better than D5 quality - something our own testing at DFC verified last year.

Anyway, there’s lots of good info in there - go read it if you are a codec geek like me.

I do wish they’d showed a PSNR result of uncompressed vs ProRes4444, and some samples of how much the alpha channel boosts the throughput - which shouldn’t be too terribly much if it is a typical alpha channel. Tests will confirm, though.

Argh - draft got eaten by timeout, now I gotta get back to work. But there’s good stuff in there - go read!

-mike

(Page 1 of 1 pages for this article )

Another interesting point in the white paper is when it describes how many streams of prores 4444 it can handle in FCP (5 with a proper raid). It says in small print that it was tested on the latest build of snow leopard.

I hope this is a hint that the new version of quicktime that ships with snow leopard is faster (gpu accelerated?) and also makes apples pro apps faster at the same time.

Maybe a thought for those who were disappointed that fcp7 came before snow leopard, and therefore not benefiting from the new opencl gpu accelerated api that will come with the new os..

Posted by .(JavaScript must be enabled to view this email address)  on  08/14  at  03:26 AM

---

This is a WP from Apple, no? What is the merit?

Posted by .(JavaScript must be enabled to view this email address)  on  08/14  at  02:50 PM

---

Christian - I specifically asked about Snow Leopard, since a lot of people presumed this might be a Snow Leopard contingent update.

The product manager specifically said no new FEATURES would be enabled by Snow Leopard, but that PERFORMANCE may change.

The trick would be to run their tests under Leopard and see if there was a difference. But if they are running on pre-release Snow Leopard, that implies that is the best place to test it, no?

So yeah, Snow Leopard should be optimal for performance.

I DID ask if that would be the recommended platform for running this and they emphatically said yes.

-mike

Posted by Mike Curtis  on  08/14  at  03:09 PM

---

Name:

Email:

Location:

URL:

Smileys

Remember my personal information

Notify me of follow-up comments?

Submit the word you see below: