Welcome to our community

Be a part of something great, join today!

  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Weapon compression - WOW!

Thanks Phil, I think Gene meant different spatial resolution but at the same datarate...so for example:

6K WS @ 10:1 vs 5K WS @ 7:1

or

6K WS @ 6:1 vs 5K WS @ 4:1

The wavelet compression make things a little less predictable...what did you find out about this?

Thanks!

I have a different way of working when it comes to "tuning things" that much. I have my "way" to work at 6K-5K to equate to a 4K finish. For that my general mindset is 5:1 for "at speed" recording. On Weapon I will be shooting lower REDCODE RAW Compression Ratios likely for everything.

If I was shooting below 5K I would be always shooting at the lowest possible REDCODE RAW Compression Ratio, but I can count on one hand how many times I've shot below 5K at this point. Once for testing. Once on another shoot that needed 120fps.
 
Aren't the files 16bit to hold enough stops in the linear raw file?

Bit depth does not equal dynamic range (stops).

12bit has 4096 shades, 14bit has 16384 shades, they can both store infinite stops of dynamic range, 14bit just has finer gradiation of levels.

The limiting factor on the number of stops recorded in the file is the analogue to digital converter.

The R3d is always only 12bit linear, not sure what the A/D stage on the camera is as Red does not publish that spec.

Red specify weapon/dragon has a SNR of 80db, that would indicate a 13bit A/D converter - but they also specify 16.5 stops of dynamic range.
16.5 stops would require a 17bit A/D converter, marketing conflicting with math.
 
"they can both store infinite stops of dynamic range" - DR is always noise limited, so as soon as the quantization noise is greater than image noise the quantization noise is now the limiting factor. Bit depth determines quantization noise, so no, DR is not infinite regardless of bit depth.

"The R3d is always only 12bit linear" - um no.

No, we don't talk AtoD - it's proprietary.

Graeme
 
I was thinking the RAW file was the direct output from the A/D conversion compressed linearly to R3D without "remapping" the bit depth from the linear light (1 stop increment of light = 1 bit increment = double light = double bit).
But it looks like there is more "proprietary" things going on the signal chain. So no linear corrrelation between bits -> light stops in the R3D file.

(Did I understand right Graeme?)

Pat
 
The R3D file is designed so as not to limit the DR recorded from the sensor. But even in the linear system you describe, the relationship between bits and DR is not as simple as that (as I mention above about image noise and quantization noise). The main comment I was making though is that R3D is not 12 bit!

Graeme
 
The R3D file is designed so as not to limit the DR recorded from the sensor. But even in the linear system you describe, the relationship between bits and DR is not as simple as that (as I mention above about image noise and quantization noise). The main comment I was making though is that R3D is not 12 bit!

Graeme

Isn't it 14bit (in a 16bit wrapper) like pretty much all CMOS cameras out there?
 
Is there any compression gain from a Epic Dragon Mini Mag 512 to a Forged CF Weapon?

Yes. That's one of the main features of Weapon. Check out the DSMC Media Ops Guide:
http://s3.amazonaws.com/red_3/downloads/manuals/DSMC_Media_Ops_Guide/DSMC_Media_Operation_Guide_Rev-L.pdf

Both Weapon MG (225MB/s) and Weapon CF/Forged (300MB/s) have increased data bandwidth to allow for better compression ratios using 512GB and 1TB MINI-MAGs. 240GB MINI-MAGs are still being worked on to increase their write-to speed from what I understand.
 
Yes. That's one of the main features of Weapon. Check out the DSMC Media Ops Guide:
http://s3.amazonaws.com/red_3/downloads/manuals/DSMC_Media_Ops_Guide/DSMC_Media_Operation_Guide_Rev-L.pdf

Both Weapon MG (225MB/s) and Weapon CF/Forged (300MB/s) have increased data bandwidth to allow for better compression ratios using 512GB and 1TB MINI-MAGs. 240GB MINI-MAGs are still being worked on to increase their write-to speed from what I understand.
that bandwidth is SUPER LOW when you compare to other cameras.I clearly dont understand why it cant be more FUTURE PROOF ?
Since WEAPON CF body will remain same on 8K these numbers are not quite enough for GOOD quality.I dont want to buy another new tech superspeed side module tc. with post Dragon sensor.
Did you know Alexa 65 have 850MB/s data bandwidth?
Btw Why not working together with codex? RED SSD system too much archaic next to Codex.
Codex SXR Drive 2TB 20Gb/s
https://codexdigital.com/products/media/capture-drive-2
 
"The R3d is always only 12bit linear" - um no.
No, we don't talk AtoD - it's proprietary.
Graeme

Sorry, I should have written - The R3d is, currently, never more than 12bit linear.

ps. Any resolution on the chroma aliasing?

Isn't it 14bit (in a 16bit wrapper) like pretty much all CMOS cameras out there?

No and its not that simple - Arri use dual 14-bit A/D converters, low/high gain, to 16 bits of data then stored in 12bit LOG arri raw. Sony F65 is use 16bit A/D to 16bit Linear raw file. Dragon/Weapon is "proprietary" A/D rated at 80db which infers it can capture 13bits of signal, and not sure how weapon is suppose to have an additional .5 stops of dynamic range but the specs are the same 80db SNR as dragon.
 
I dont want to buy another new tech superspeed side module tc. with post Dragon sensor.
Did you know Alexa 65 have 850MB/s data bandwidth?.


Good point on future proof but Alexa65 needs that speed because of their lack of Raw compression. Shooting lower than 3:1 RC is overkill (IMHO). Now I'm not a highspeed guy and don't know if this should be the bench mark for RC compression/FPS/High bandwith decision.

Pat
 
Good point on future proof but Alexa65 needs that speed because of their lack of Raw compression. Shooting lower than 3:1 RC is overkill (IMHO). Now I'm not a highspeed guy and don't know if this should be the bench mark for RC compression/FPS/High bandwith decision.

Pat

im not gonna shoot uncompressed too :) however we are gonna upgrade to next sensor in few years.I dont want to switch my all data system again.
300MB/s speed is a JOKE when you spend more than $ 2K just for SINGLE 512GB media.Check out codex media specs.Red is following them 5 years behind.
Completely serious on my last sentence.
SXS Tech is more than 5 years old.Same speed with current RED MINI SSDs :)
 
Good point on future proof but Alexa65 needs that speed because of their lack of Raw compression. Shooting lower than 3:1 RC is overkill (IMHO). Now I'm not a highspeed guy and don't know if this should be the bench mark for RC compression/FPS/High bandwith decision.
Pat

Except, Weapon is suppose to be the future platform for the next sensor and the ssd is no longer modular.
Its already limited to 5:1 at 8k/24 and 12:1 @ 8k/60.
Whatever the next generation sensor after Dragon will be, higher data rate is a given = even more compression.
 
Except, Weapon is suppose to be the future platform for the next sensor and the ssd is no longer modular.
Its already limited to 5:1 at 8k/24 and 12:1 @ 8k/60.
Whatever the next generation sensor after Dragon will be, higher data rate is a given = even more compression.

Your assuming well always have the same media.

Who's to say a new "Mega Mag" isnt being worked on? Stands to reason, if RED is already working on the next sensor, then they are presumably working on the next media format (we've already had four media systems to date - hard drive, CF, 1.8" Mags, Mini Mags). RED has traditionally evolved in logical sequential steps. I have to imagjne this would apply to media as well.

Right now, 5:1 @ 24fps 8K is whats possible TODAY.

Who knows what these bumbers will be in a year or two...
 
Last edited:
that bandwidth is SUPER LOW when you compare to other cameras.I clearly dont understand why it cant be more FUTURE PROOF ?
Since WEAPON CF body will remain same on 8K these numbers are not quite enough for GOOD quality.I dont want to buy another new tech superspeed side module tc. with post Dragon sensor.
Did you know Alexa 65 have 850MB/s data bandwidth?
Btw Why not working together with codex? RED SSD system too much archaic next to Codex.

I'm well aware of what the Alexa 65 does, 6.5K Open Gate results in about 732MB/s using the Codex XR Drives. I've read your posts regarding CFast and have replied to those as well.

Weapon 8K isn't out yet, but what we do know is for "at speed" recording is it's 8K FF 24fps with REDCODE RAW 5:1 Compression Ratio. That is still the "standard" Compression Ratio used by most feature film productions, so I'm missing out the the "GOOD quality" complaint there, yes it would be lovely if it was 2:1-4:1. Similarly to the Alexa 65 6.5K, Weapon 8K supports up to 60fps for 8K FF from what I understand as of today. That's what we know as of now.

RED is focused at the moment at internal compressed raw workflows. And we literally have no idea what's coming down the pipe in the future. It's a much more compact system with a different workflow strategy.

It's a valid criticism you've voiced before. And I can think of 3 solutions within the current ecosystem that provides "more bandwidth" if necessary. But it's just to early to make any judgments between these two systems. Both are wildly different flavors of ice cream.

Also, just to say it, you cannot purchase an Alexa 65. You can however purchase a RED Weapon 8K Dragon camera when that time comes.
 
Your assuming well always have the same media.
Who's to say a new "Mega Mag" isnt being worked on? Stands to reason, if RED is already working on the next sensor, then they are presumably working on the next media format (we've already had four media systems to date - hard drive, CF, 1.8" Mags, Mini Mags). RED has traditionally evolved in logical sequential steps. I have to imagjne this would apply to media as well.
Right now, 5:1 @ 24fps 8K is whats possible TODAY.
Who knows what these bumbers will be in a year or two...

New media would not increase the bandwidth of the signal path, its normal Sata3 limited to 520MB/sec, the computer world has already moved on to NVME/M.2/PCIE type interconnects.

I don't think uncompressed/codex is the right direction either - expensive, proprietary and negligible gain over minimal compression - but sticking with sata3 ssds on weapon doesnt seem very forward looking.
I suspect Red thought if people had to buy all new accessories and all new media (after just switching from 1.8 to minimag) - they would have a bigger mutiny.
Or possibly to benefit from more than sata3 data rates the rest of the camera would need to have a much more evolved re-design.

Nick you are correct the Alexa 65 currently maxs out at 28fps because its using the existing codex media, expect a public 60fps option soon, maybe at ibc.
 
Sorry, I should have written - The R3d is, currently, never more than 12bit linear.
that is not correct and I've already pointed that out. The listed spec for Epic Dragon is 16bit R3D.

Graeme
 
New media would not increase the bandwidth of the signal path, its normal Sata3 limited to 520MB/sec, the computer world has already moved on to NVME/M.2/PCIE type interconnects.

I don't think uncompressed/codex is the right direction either - expensive, proprietary and negligible gain over minimal compression - but sticking with sata3 ssds on weapon doesnt seem very forward looking.
I suspect Red thought if people had to buy all new accessories and all new media (after just switching from 1.8 to minimag) - they would have a bigger mutiny.
Or possibly to benefit from more than sata3 data rates the rest of the camera would need to have a much more evolved re-design.

Nick you are correct the Alexa 65 currently maxs out at 28fps because its using the existing codex media, expect a public 60fps option soon, maybe at ibc.

The Alexa 65 in London two weeks ago was only shooting 24fps, but yep, that 60fps is incoming "sometime". Curious to see how that's actually implemented and what it looks like.

We (here on REDuser) at one point discussed a "future RED camera" back when folks were asking "1000 fps". I personally was proposing merely an increase of 4-6X data rates/fps with the forward thinking of a SATA III.2 data rate, approximately 1969MB/s, which could imply a future similar to that desire. Imagine a world where 8K FF at 240-360fps is attainable. 5K at 384-576fps. And I imagine if things went that far we could see larger than 8K resolution and potentially even the format size shooting as well. I'm personally hoping RED stays within the 4-5 micron range on the pixel size based off of everything I've seen to date regarding pixel quality, design, and density. Also, optics at a certain point "are" apart of the issues we face when the pixel size gets too small, as well as various artifacts.

It would take a sensor redesign for sure and yes, likely new media if were are exploring data rates that high. I don't know what is under the hood within Weapon, but Weapon is "future minded" and I'm betting RED builds off this form factor/platform in the future. What is now and current is much better compression ratios with Weapon, which is fantastic for many of us.

All future, future, future talk really. I'm very interesting squeezing the most out of Weapons 5K-6K as well as 8K when that time comes.
 
that is not correct and I've already pointed that out. The listed spec for Epic Dragon is 16bit R3D.
Graeme

16bit is the listed spec but so is 80db SNR, which does not correlate.

The data rate of actual files from a Dragon recording various compression ratios and frame sizes indicates the bit depth never exceeds 12bit. Red patent document specs 12bit.

Taking a Dragon 6k R3d into Nuke, AE, or your favorite app - in a 32bit comp, exporting it as 12bit Lossless J2k, re-importing that J2k as another layer and doing a difference matte results in 0 difference in pixels.
I just mention J2k format because it is easy to create lossless files in various bit depths.

You have previously stated the process is all linear and you are not re-mapping to a log space like Arri does to maintain 14bit data in a 12bit file.

Not saying 12bits is necessarily bad, just that I see no way of getting more than 12bits from the currently recorded r3d files in the real world.
 
Back
Top