How fast is REDCINE?

R

Rory Hinds

Member
How fast will the REDCine software be?

I've worked with Andromeda Sculptor software which processes the files captured from the DVX, which I assume will be similar to working with RED as you'll need to process the RED files into whatever codec you want to work with.

Sculptor is cool but very very slow and I'm wondering how fast REDCine will be? Working with Sculptor can take days to process your footage (if you've shot loads) and on projects where you need a quick turn around Sculptor just can't be used, thus you need to shoot with another camera to get the footage fast.

How are the RED team dealing with the speed of processing the RED files?

Regards
Rory
 
How fast? Very.

1) you can play your REDCODE files in real time - at what resolution depends on your processor specs.

2) colour transforms are GPU based, and hence very fast indeed. I'd expect realtime performance for a lot of configurations, but again, more details as we get nearer release.

3) full quality demosaic of the full 4k will be a touch slower. It's got to be as we're using a complex / nice demosaic. However, hardware is getting rapidly faster and we'll be looking into how to best optimize the demosaic algorithm.

Hope that helps,

Graeme
 
This is wonderful news, I too was worried about the speed of the software. Glad to hear that you guys are taking advantage of GPU acceleration, that should definitely speed things up.
 
I hope you're going to optimize for the dual quad cores coming this month. Let's hope Apple leads the way with a blu-ray burner too.

Graeme, so does redcode just plug into an Avid and replace its Avid DNxHD? By the way is DNxHD a wavelet algorithm too? (It seems to do a lot quickly).

And on Final Cut, I assume it will be just like a HDcam codec that plugs right into quicktime and edits. Is that right?

Also, from what I understand it scales very well... so when Im editing am I watching 1080p in real time with FX? Or Faster performance with 720p? Then, do I render everything in 4k and then downconvert to 1080p for blu-ray when Im done?
 
Yes, we'll be looking at multicore / processor optimization also.

I don't know what DNxHD is, but I'd hazzard a guess that it's DCT based. I wonder if any Avid gurus can tell us more about it?

REDCODE is "just" a "normal" Quicktime codec :-) Should plug into practically anything.

If you're just mastering to a 1080p format, I'd probably just get REDCINE to convert the 4k to 1080p uncompressed and go from that. However, nothing to stop you doing other workflows. REDCINE / RED in general is about facilitating choice.

Graeme
 
...or cook out a DV or DVCPRO HD for quick, responsive, RT enabled offline edit, then when ready for the online, recook your selects to the uncompressed, then use Media Manager to recreate a high res timeline and relink the pfhat pfhiles.

I do a similar workflow for uncompressed telecine projects.

-mike
 
Last edited:
Mike's post brings up a question I've been meaning to ask... say your ultimate aim is to produce a 2K master but you want to work off-line in DVCPRO HD (BTW, I really get a kick of thinking of these formats as "off-line"), will REDCINE be able to apply a slight scale as it's making the proxy, so that the full image area is placed within the bounds of DVCPRO HD 1080 or 720?

I'd hate my proxy to be a "crop" and not see what's going on in the above/below cut-off areas, yet at other times you might want it to be a crop if the 16:9 you're producing is a final. Or you want to tilt up on the image area, or down to hide a mic in frame...

In general, it would be nice for REDCINE to be flexible in cropping and scaling, beyond simple every-day presets, because who knows what kind of strange thing you might need, such as the above examples?

Regards,
 
I should add, that 1) and 2) and 3) above go together, so that you shouldn't expect to see 4k being demosaiced and colour corrected and played back in real time.

RE: 2) GPU is not used for demosaicing. CPU is. It's hard to get the complexity of demosaic we need running on a GPU.

RE: 3) As posted earlier, current time for just the demosaic of 4.9k raw frame is 1.7 seconds (single core of quad G5, totally un-optimized development code). That's todays "touch slower" we hope to get that time down, but until we've got into the optimisation, I don't know how far down it will go. (People were worried when they'd heard that Dalsa was taking 8 or 4 seconds per frame.)

I don't want people to think either of two things:

a) it will be dog slow
b) it will be realtime everything

As we all know, in the real world, 4 processors doesn't make code 4 times as fast, as there overheads and other things at work.

The above is of today. Things can and will change.
 
Last edited:
As indicated that un-optimized development code running on a single core G5. No latest Intel chip with MMX/SSE & multi-threading. Forget that number, we will come out with much more accurate numbers later.
 
Graeme_Nattress said:
As wel all know, in the real world, 4 processors doesn't make code 4 times as fast, as there overheads and other things at work.
Click to expand...

Sorry Graeme to ask that again, but will REDCINE have a render Queue fonction to easly spread the rendering to the nodes' computers? As you say multiprocessors machines doesn't render as fast as multi monoprocessor machines, and I have "old pc" that I still use when I need to render heavy 3D.

BTW, my "6 years old pc" (dual PIII 1000, 1.5 GB RAM) has rendered twice slower than my "today pc" (dulacore P4 3G, 4 GB RAM) using vray in 3ds max 7. So in my opinion, even if it could seems a lot of difference, 2 "6 years old pc" = 1 "today pc", so it worth to have a effective render queue.
 
Um - I didn't say that - I said that 4 cores OR 4 processors never gives you 4 times the speed of one processor. There are memory and bus constraints. Some algorithms are not as parallelizable etc. etc. etc.

As for REDCINE - we're still working on details and your comments are noted.

In the camera is s super-computer - didn't we tell you that? :-)

Graeme
 
Thanks Graeme for the precision...

And thanks for the news, with RED I, i will get my fisrt super class SuperComputer, ehehehe....

BTW, i have nothing agains Super computer, just don't want to throw away my old ones.... recycling is the word :-)
 
No problem. Consider it forgotten. I understand it is in development. I was just surprised it could take so long per frame to run that process.

Rob Lohman said:
As indicated that un-optimized development code running on a single core G5. No latest Intel chip with MMX/SSE & multi-threading. Forget that number, we will come out with much more accurate numbers later.
Click to expand...
 
It's a pretty complex process. If you do some googling you'll find a lot about it, especially that the adaptive algorithms that make the best pictures are slow.

Graeme
 
There must be a reason why this wouldn't work, but I'm too curious not to ask: why not route the RAW files (once they are captured) through the camera's "supercomputer" to get realtime RGB instead of a few seconds a frame?
 
Two reasons - the camera algorithm is 4k RAW to RGB to 2k or 1080p or 720p, whereas in software, we go from 4k RAW to 4k RGB. Because we stay at 4k we use a different algorithm that's adaptive which gets us the best results for 4k use. In camera we use a different algorithm that's geared towards 2k and HD use. Hope that helps!

Graeme
 
Just curious. if we are shooting something and we need quick turn around, I guess we shoot it to RGB 2k or High def in cam and then we don't have to go through this process in post. Correct?

red 258
red 847
zoom 67
 
That makes sense. It also gives some sort of an idea as to how fast the eventual conversion process will be. If the chip inside Red One can convert 4K RAW to 2K RGB in realtime then a computer converting 4K RAW to 4K RGB is doing roughly (very rough) 4 times the work and should take at least 4 times realtime.

Thanks so much for your time, Graeme.
 
You must log in or register to reply here.
Back
Top