DRAGON debayer options - Red Rocket-X hardware versus RCX-P software only

[Bob Gundu]

Having spent part of the last couple of weeks at Neil's "LAB" I can say that there is so much going on in there with "NOT OUT YET" CAN'T TALK ABOUT" stuff, that is amazing what we were able to accomplish there, and while He nor I can say anything about it, I can say that tomorrow looks really good for DRAGON...

Perhaps the new macPro is announced tomorrow? Hmmmmmmm....

 

[Robert Ruffo New]

Something is seriously wrong with your tests.

Its making it look like post on dragon projects would be a nightmare and that the rocket-x is insanely slow.

Phil's clips were shot 2:35, not even full 2:1.

That is only 15% more resolution than 5K full on the Epic that we normally shoot, so I hoped your number were wrong.


Can't help on the very slow rocket-x front, but post workflow for Dragon does not seem nearly as bad as your test shows.


I used the same 5 clips provided by PHIL


Output 5 x 3sec clips to EXR Full Resolution / simultaneous transcodes - (software redline decode)

REDCINE-X PROFESSIONAL Build 21.1.30346 Beta


[10-08-13-58-07-917] Batch Job Complete: A002_C020_10030T Duration: 00:02:36.625
[10-08-13-58-09-387] Batch Job Complete: A002_C008_1003V8 Duration: 00:02:39.096
[10-08-13-58-09-856] Batch Job Complete: A001_C027_1003FG Duration: 00:02:39.565
[10-08-13-58-11-393] Batch Job Complete: A001_C069_10031T Duration: 00:02:41.102
[10-08-13-58-12-383] Batch Job Complete: A002_C021_10036V Duration: 00:02:42.092


15secs = 162secs render = 10x realtime


Output 5 x 3sec clips to EXR Full Resolution/ 1 at a time transcodes - (GPU)

[10-08-14-20-53-731] Batch Job Complete: A001_C027_1003FG Duration: 00:00:15.440
[10-08-14-21-07-289] Batch Job Complete: A001_C069_10031T Duration: 00:00:12.553
[10-08-14-21-21-411] Batch Job Complete: A002_C008_1003V8 Duration: 00:00:13.115
[10-08-14-21-35-465] Batch Job Complete: A002_C020_10030T Duration: 00:00:13.050
[10-08-14-21-49-495] Batch Job Complete: A002_C021_10036V Duration: 00:00:13.014


15secs = 67 secs render = 4.5x realtime



Software only decode is only a little slower than your Rocket-X decode



Single GPU decode (GTX680) is almost twice as fast as your Rocket-X decode.



If someone would explain the --deviceName gpu option for redline it would be even faster.


I used EXR as that is our normal workflow, other formats may be better or worse.
I did a quick AB of GPU/Software decode in AE, did not see anything horribly different, this is using the latest build which seems to have fixed the gpu artifacts.


Now I'd really like to see some full resolution 6144x3570 R3Ds posted.


Looking like GPU is a good solution and Rocket-X may be only useful for laptops, and even then?


PS. You can just look at the log file, you dont need some silly iphone stop watch to check render times.

6K si not 15% more pixels - pixel count is roughly 30% more. You have to multiply both horizontal and vertical and count the total.

Everybody told me to shut up and "the future is 4K" and blah blah blah but really 6K is not practical for most existing pipelines, most current destinations of footage for most people who, like me, do not always control how/where post is done - most computers out there being used for post cannot handle it and I don't have $7500 for another Red Rocket card now that the camera rental market is down to almost nonexistent.

I can hear the horror stories now - the missed deadlines the client complaints, the rentals cancelled to instead go with Alexa because of said horror stories.

I wish Dragon were 4K with bigger pixels.

 

[Ketch Rossi]

Perhaps the new macPro is announced tomorrow? Hmmmmmmm....

Ha ha, not literally....

 

[Shiblon Wixom]

We're all mostly going to shoot 4k HD and 5k HD at 8:1 compression, just like before, but with 2000 ISO and the motion mount, but you can go to 6k for VFX plates if needed. 17:1 compression seems excessive.

 

[Christopher*Seguine]

In my comparison it is, Phil's clips are 2:35 - 6144x2592, I almost always shoot 5k FF 5120x2700, so the posted clips are only about 15% more pixels.

6K si not 15% more pixels - pixel count is roughly 30% more.

 

[Shiblon Wixom]

6k 6144x2592 = 15,925,248 pixels, 2.37:1
5k 5120x2700 = 13,824,000 pixels, 1.89:1
which is about 15% difference, but the two values are are not the same aspect ratio.
5k 5120x2178 = 11,151,360 pixel, 2.35:1
which is 42% difference if comparing aspect ratios more approximately.

If working on a feature, using a 2:1 shooting aspect ratio with 2.35:1 and 16x9 frames guides would probably work better than just going with a 16x9 HD or 2.4:1 shooting aspect ratio. So with a ratio of 2:1 in mind the % difference is probably closer to 30%.

 

[Christopher*Seguine]

Too bad Dragon will max out at 2:1, rather than 6k FF as initially indicated.

FF is very useful for the option of re-framing in post.

 

[Axel Mertes]

Hi All!

I still think that extracting one or a couple of single frame R3D files and exporting it/these with all options like mentioned in my former post would put the debate about deBayer quality from CPU vs. GPU vs. ROCKET vs. ROCKET-X into a better shape.

Arguing deBayer quality on downscaled stuff is IMHO pure nonsense.

Even the old ROCKET looked sharper first and many liked that look. But doing a simple sharpen 0.1 in e.g. eyeon Fusion on software decoded R3D files made them look identically sharp as the hardware decoded image. Looking at the details revealed that the sharpened software decoded image was far superior to the ROCKET. For VFX workflows this is key to understand and to get the best possible image quality for beginning the workflow. And in fact, in many cases we preferred the unsharpened look (like for female skin tones...).

So I think Neil should not invest time on encoding downscaled clips, but export a single frame and we can argue from that.
We still need to find out if e.g. GPU looks identical to software only.

The current GPU approach reminds me pretty much to what we've seen by Iridas SpeedGrade (now Adobe development leads, btw) with RED ONE firmware build 15. We were able to decode RED R3D files in realtime using CPU JPEG2000 decode of RAW, then deBayer in realtime on the GPU. OK, back then CPU and GPU were significantly slower than today, but it was pretty much the same architecture (if not really the same, who knows...). If RED would not have had the marriage to Scratch back then, we could have seen this GPU thing nearly from day one... (I did...)

And the GPU deBayer from Iridas / Adobe is pretty amazing too.

Cheers,
Axel

 

[Phil Holland]

Too bad Dragon will max out at 2:1, rather than 6k FF as initially indicated.

FF is very useful for the option of re-framing in post.

Huh? 6K FF is in there last I saw.

 

[Axel Mertes]

Huh? 6K FF is in there last I saw.

The homepage tells different things:

http://www.red.com/products/epic-dragon#tech-specs

MAX IMAGE AREA 6144 (h) x 3160 (v)

So 6144/3160 = 1,944... which is pretty close to 2:1... or DCI 1.9 : 1 (2048x1080y, 4096x2160y and scaled thats 6144x3240y - which is not fully covered).

So 6K FF must be an option to come. I think photographers would love it...

Axel

 

[Phil Holland]

The homepage tells different things:

http://www.red.com/products/epic-dragon#tech-specs

MAX IMAGE AREA 6144 (h) x 3160 (v)

So 6144/3160 = 1,944... which is pretty close to 2:1... or DCI 1.9 : 1 (2048x1080y, 4096x2160y and scaled thats 6144x3240y - which is not fully covered).

So 6K FF must be an option to come. I think photographers would love it...

Axel

Ah. That's where you have some confusion. Dragon 6K is a 1.94:1 sensor and that is indeed the "full" format, not 1.9:1 like Mysterium-X. From the manual:

6K Full: 6144x3160
6K 2:1: 6144x3072

That 6K Full resolution is the maximum physical recording size and the sensor isn't going to magically grow in vertical size there.

Interestingly enough the Dragon will ship default with 6K 2:1 as the set recording format. You can of course change it to whatever you'd like.

 

[Jaakko Rinne]

Everybody told me to shut up and "the future is 4K" and blah blah blah but really 6K is not practical for most existing pipelines, most current destinations of footage for most people who, like me, do not always control how/where post is done - most computers out there being used for post cannot handle it and I don't have $7500 for another Red Rocket card now that the camera rental market is down to almost nonexistent.

I can hear the horror stories now - the missed deadlines the client complaints, the rentals cancelled to instead go with Alexa because of said horror stories.

I think I remember this same sentiment about 4K not being practical being tossed around just a couple of years ago. 4K is practical now, and 6K will be soon enough.

I wish Dragon were 4K with bigger pixels.

So you could finish in 3.2K?

 

[Christopher*Seguine]

I was just going off the spec listed on the site:
Epic: 5K RAW (Full Frame, 2:1, 2.4:1 and Anamorphic 2:1)
Dragon: 6K RAW (2:1, 2.4:1)


6K Full was enabled on the camera you were using?


If so, thats great. They should update the specs on the site, if the majority of your work is 1.85, 2:1 can be seen as limiting.

For a 3996x2160 release, a source of 5194x2808 would be preferred to account for bayer pattern loss, if the dragon FF mode is enabled that allows for about 10% reframing in V and plenty in H , which is great.


So, back to the Rocket-X - Neil, is it really that slow? Can you post an EXR still so we can judge the debayer quality vs software?

Ah. That's where you have some confusion. Dragon 6K is a 1.94:1 sensor and that is indeed the "full" format, not 1.9:1 like Mysterium-X. From the manual:

 

[Blair S. Paulsen]

The RR-X deserves a fair shot

The RR-X deserves a fair shot

Please stop judging the potential of the RedRocket-X card from a snippet of early testing with lots of beta software/drivers in the mix.

RED is not perfect and some of their first gen products have issues, perhaps none more frustrating than the Rocket cards. That said, they also have a reputation for huge improvements in their second generation offerings. For starters, the RR-X has a 2 year warrantee (3rd year for an extra $395). It appears to be more robust in construction with better heat dissipation. It has more native horsepower.

Until we see what it can do with optimized drivers, firmware, etc its way too early to throw it under the bus. IF, after a few months, it turns out to be slower than expected, failure prone, etc then its open season. Until then, IMO the engineering team at RED has earned the benefit of the doubt - think of all the issues they have sorted out over the years. All I am saying, is give them a reasonable amount of time to dial it in before taking potshots.

Cheers - #19

 

[Christopher*Seguine]

Seriously? First its way faster and way better and now its never mind its beta but has a warranty?

We all know its beta and will improve.

Nobody else is posting Rocket-X info, so if someone is going to state its faster or the quality is better than software rendering, and just post vimeo clips that its impossible to tell, they shouldn't be surprised when they are questioned.
Its just coming off as silly.

There are many people anxiously awaiting dragon upgrades and with that the added post concerns.

THANKS TO Red, now there are other options - GPU.

So I'm sure I'm not the only one curious both about speed and quality of the Rocket-X.

Having bought several of the first rockets, and then bought them again as "repairs" - yes I am concerned about reliability and warranty, but mostly I am curious if/how it will help our workflow.

Should we buy Rocket-Xs for all our workstations, just for field use, or ?

Currently we do not use the original rocket for final renders or vfx handoffs , has that changed with the RocketX?

How much will additional size of dragon files slow down our workflow? and how best to compensate?

Just looking for accurate information,
beta or not.

Looking at the images closely on a Sony BVM grading monitor I've come to the following initial conclusions:

1) the image rendering quality of the RR-X card is way better than the old RR card - way better!

3) SPEED of renders! .... both for final renders and for checking out different options at FULL debayer ... no comparison ... if you have the need for speed, Red Rocket-X is the way to go (and I was running the software only renders on a beefy Mac Pro with a TITAN in the expansion chassis).

Neil

 

[Gabriele Turchi]

if someone have a clear example of how the rocket ( even the "" old one "") render inferior images compared the software ...could please post it ?

i assume that the difference is more noticeable in some particular circumstances ?
thanks

g

 

[Christopher*Seguine]

The original rocket applies a preset amount of sharpening or its debayer technique causes a sharpening of the image, it can not be adjusted. There maybe be other differences. I didnt say it was inferior, I just said we would not use it for final renders as usually you would like to control the amount sharpening and even area of sharpening in the image. It can be a problem for vfx shots, green screens, etc as sharpening will significantly enhance noise in shadow areas - specially in the blue channel. There was a thread awhile back in the rocket forum where someone posted examples of rocket vs. software decode.

If the Rocket-X does something similar? Neil? I would love to see an example posted.

 

[Robert Ruffo New]

I think I remember this same sentiment about 4K not being practical being tossed around just a couple of years ago. 4K is practical now, and 6K will be soon enough.


So you could finish in 3.2K?

Soon enough is not now. The problem is - to be fair not Red's fault - computers have been inching forward these past few years, not racing forward as they had before. Moore's Law is no longer true, not even for GPUs and even less for CPUs. The fastest PC you can buy today for $5000 is how much faster than a top-of-the-line $5000 (or so, at the time) Mac Pro 12 core from 2010? 20%? 30%? Its' certainly not many times faster, as would be the case for a same-price comparison 3 years later in the early 2000s.

Cameras are digital products and their value must be amortized immediately, as they loose value so fast. I don;t really care what my clients will be asking for and/or need 3 years from now. I care about today, and a year from now, two at the far end. I can buy what I'll need in two years in two years, for far less.

Most fo our work is ads music videos, some corporates, fashion catalogs. We have never, ever ever been asked for a 4K finish. No one is interested. I mean not one client has ever even casually mentioned mild interest - and I hear the same reports from everyone I know in the business. Even the feature producers - distributors except at the very top of the top Hollywood end don't care about 4K Its' not because they;re stupid, it's because penetration of 4K TVs is less than a fraction of 1%, and they are hardly selling at all, so why bother? Most video production has no need for "future proofing" - and unlike SD, which was awful, 1080p and 2K are far from being crap - in many cases they show the majority a lot of the resolution the lenses and focus pulling slight imperfections were able to capture. Many visual experts say that 2K is about all most people's eyes can resolve at a comfortable seating distance. SD was far below that, so HD caught on fairly quickly after it was released - it was an obvious difference. Would people not see a movie at the theater because it was projected and/or acquired in SD? Yes, I think many would pass. Because it was 2K rather than 4K? Studies show that many people can't even tell the difference. Did anyone complain that Black Swan was "too low rez?" 16mm resolves about 2K and the answer, we know, is no.

Then there is the higher expense of 4K 3D effects and renders - no one wants to spend lost of extra $$$ for a basically non-existent audience, and 99% of video content makes most of its money or serves the entirety of its promotional purpose in a few months, not years. Computers can get faster, but we will still be pushing them to their limits within deadlines - lost of polygons, lots of very nice and complex 3D shaders - all that means more within always limited rendering resources than extra resolution that may or may not even be perceptible to most seats in the house.

I can see an Imax release wanting the most resolution they can get, but to be frank, that's not any project I am currently working on, is not any project the people who rent our cameras are working on, is almost certainly not what you, reading this, are working on. Most indy films are lucky if they can get some decent numbers on iTunes - any significant theatrical release, let alone 4K Ultra AVX and Imax, are highly, highly unlikely release venues so unlikely that spending extra money on the remote possibility is a very bad investment.

Here are Reduser, there is an imaginary belief that 4K finishing is in demand. It's just not, and buying cameras for future use rather than immediate needs is a sure way to waste a lot of money in depreciation.

I've said this before, but for the purposes of noise reduction, some occasional reframes, motion tracking and very precise rotoscoping that ends up being sub-pixel in precision, 4K was awesome, 5K was borderline overkill. There are many rarely discussed reasons to oversample, but there are limits. This for the 1080p and 2K finishes that are what 99.999% of clients ask for in the real world, clients who have said "You want to shoot 6K? Really? How are we going to edit that!?"

 

[Johnny Friday]

Robert: you have left out a big gap in your assumption: There are those of us that make a pretty decent living selling stock images. I for one have increased my sales by dramatic numbers and reached clients i would normally have NEVER sold to because my work is available at 4k & 5k. Now, to be fair i agree with much of what you say. Much of the work i do today for commercial or doco work is 1080p and it's for today only. Nobody is going to re-run a Corona commerical i shot ten years from now in 4k.

BUT, i do sell 4k & 5k stock to clients that use in feature films, or do 2d to 3d conversions. And in fact sold shots to producers of Dome Theater productions....so yes, there are a growing number of us that work in these K's and our clients in fact ask for it. I have three clients i am shooting for NOW doing 4k finish or 3180x2160....Funny that they are all overseas--China, Korea and UK.

4k tv market i think you have very wrong....i am only now seeing companies like OTF (off the fence productions) now gearing up for 4k productions....now 3d tv is certainly dead....but what i do know is i have a high demand and only growing for 4k clips and shorts.....most are these foreign broadcasters and some television manufacturers themselves that will host their 4k material on their site. Sony is a perfect example...they have employed a number of film makers to produce shorts in 4k and finish at 3180x2160 and that material will be hosted on their website for download to customers buying their 4k televisions......it's a complete other discussion on delivery methods over the airways i know....but you may not be working in this area, but there are those of us that do and more than you may know....

 

[Axel Mertes]

if someone have a clear example of how the rocket ( even the "" old one "") render inferior images compared the software ...could please post it ?

i assume that the difference is more noticeable in some particular circumstances ?
thanks

g

Hi Gabriele,

you asked for it, so here we go. This is a comparison of RED ROCKET vs. RED Software decoding.
Please read the text inside the images carefully to understand what you are looking at.

Actor: Gen Seto
Movie: Sushi in Suhl, 2012
With kind regards of StarCrest Media



Important: Neither the new GPU stuff nor the new RED ROCKET-X is being used here.

The source material is a RED R3D shot on RED ONE MX as far as I remember.


This image shows different kind of artefacts when pushing gain strongly:


https://dl.dropboxusercontent.com/u/22067399/RED%20Software%20SDK%20vs.%20RED%20ROCKET%20-%20GAIN.png

This image shows sharpness and aliasing issues:

https://dl.dropboxusercontent.com/u/22067399/RED%20Software%20SDK%20vs.%20RED%20ROCKET%20-%20SHARPNESS.png

We made this for our customers to understand why we use software only decoding for VFX and final RED output work.

As to our experience, the difference is always there.

Regards,
Axel