Monstro VS V-Raptor…….

[Noah Yuan-Vogel]

We've hit a ceiling in which people just shout out for big improvements but there's little to actually improve nowadays so the small increments of improvements we achieve feel like baby steps instead of revolutions. This is just natural and Raptor is a natural progression of cameras.

I get where you're coming from but I disagree. When you have a fixed-gain raw camera competing against analog gain and conversion gain cameras, and you have red, who has been behind arri for dynamic range, the one primary image attribute most cinematographers single out as the most important, for over a decade, I think there are breakthroughs to be made that aren't just incremental baby steps. Maybe the new sensor and the new peltier cooling will give red the edge they so need, but if it did, I'd think they should be shouting it from the rooftops and proudly marketing directly on their xyla charts.

 

[Blair S. Paulsen]

Pick your poison

Pick your poison

One practical difference between Monstro and V-Raptor is data rates. The DSMC2 line uses wavelet compression, DSMC3 - unofficially Komodo and Raptor - uses a cosine based codec. Wavelets allow you to stuff a lot of image information into a 300MB/s bitstream. To package a similar amount of image metrics into a DCT codec requires roughly 800MB/s using current topologies. Wavelet based codecs like JPEG2000 or RedCode are efficient users of bandwidth, but require serious computational muscle to decode. DCT is less efficient, but much easier to decode.

So, which is better - less data coupled with higher client side hardware requirements, OR, more data but easier to decode on commodity hardware? RedCode was a key technology in developing digital cinema cameras able to deliver high resolution RAW imagery at data rates commensurate with available media/processing tech in 2007. I would argue it was still the best choice ten years later. In 2021, its a more nuanced issue.

With storage costs always declining (outside of COVID supply chain issues) - and much faster bus standards like USB 3.2 and TB3 (not to mention USB4 and TB4) - and CF Express variants capable of >800MB/s sustained writes hitting the market, RED went DCT. For some RedUsers, this change is not welcome - for others, the ability to work fluidly with RED files on modest computer systems is worth the extra storage requirements.

Bottom line - there's no free lunch, just trade offs - like most engineering ;-)

Cheers - #19

 

[Christoffer Glans]

I get where you're coming from but I disagree. When you have a fixed-gain raw camera competing against analog gain and conversion gain cameras, and you have red, who has been behind arri for dynamic range, the one primary image attribute most cinematographers single out as the most important, for over a decade, I think there are breakthroughs to be made that aren't just incremental baby steps. Maybe the new sensor and the new peltier cooling will give red the edge they so need, but if it did, I'd think they should be shouting it from the rooftops and proudly marketing directly on their xyla charts.

You are talking about minor differences between systems, which is entirely my point. We've reached a diminishing return where cameras from everyone start to blend together. Arri has been the king for a long time because they poured a lot of knowledge into their sensor, but it wasn't really perfect over the course of Alexas lifetime. Arris sensor is pretty shit when it comes to noise compared to cameras like Monstro. But the key point I'm talking about is that we've reached a point where the actual practical application of the extreme ends of digital cinema technology isn't really visible on the end result. In the hands of skilled professionals, almost any of the top-tier cameras will result in images that are almost indistinguishable from each other on the IMAX screen.

So this is what I mean by diminishing returns. We keep pushing specs on cameras, but it's not going to be revolutions in image fidelity. We will likely reach 20 stops DR, noiseless ISO6400 (without in-camera bullshittery) at 8K 150 fps by 2025. So what then? Is my question. The question is even relevant today. As I said, there are still things to improve upon, but the diminishing returns make it less relevant for each year going forward.

I'm holding a larger perspective in all of this, I'm not talking about just Red here. But the Monstro vs Raptor discussion leads to very little because the differences are so minute that anyone aiming to find a great and revolutionizing difference that makes it relevant to change from Monstro to Raptor has a hard time concluding anything substantial for it.

This is why the current cameras aren't going away, they don't drastically get price dumps, because they're still relevant. And as I've mentioned in other threads, today, knowing a camera is vastly superior than to always upgrade.


I've pointed out on many occasions that true digital cinema revolutions don't have to do with traditional camera specs anymore: Z-depth data, in-camera camera tracking data for VFX, smaller camera bodies, fanless/weather-sealed camera bodies, concave sensor shapes, AI AF systems, wireless audio at high fidelity... and so on. Even if a camera comes out tomorrow with 20 stops DR, no noise up to ISO 12500 and 300 fps 8K... what will people actually do with it? Shoot the same kind of movies as they've always done and they will look almost identical to what we already see in the cinemas right now.

The next camera system to "win" on the top-tier digital cinema lineup has little to do with the image quality and everything to do with everything else.

 

[Tom Gleeson]

Christopher,

True words. Improvements are likely to stay incremental in the short term although in the long run these improvements will add up. The achilles heel of early video and Digital cinema cameras was dynamic range but this has been mostly addressed. While any improvement is welcome it is not the issue it used to be. You could argue that HDR finished projects could easily leverage more dynamic range but as you say even there it is a diminishing rate of return.

I am more than comfortable staying with my Monstro even with the improvements the Raptor offers. The Raptor XL is the camera that will directly"compete" with Monstro but even the XL is unlikely to make Monstro irrelevant.

If the Raptor and Raptor XL are moving away from Wavelet to Cosine compression I would not complain but I would love to hear Red's rational for this move. I do have to thank Red for not moving past 8K resolution. Even in 2021 I have some clients who baulk at 8K recording. I personally would love the option to shoot full sensor but record 4K or even 6K R3D. We can do this in Pro Res why not R3D?

 

[Curren Sheldon]

I do have to thank Red for not moving past 8K resolution. Even in 2021 I have some clients who baulk at 8K recording. I personally would love the option to shoot full sensor but record 4K or even 6K R3D. We can do this in Pro Res why not R3D?

Red has always pushed resolution, so I'm not saying they should do this, but a pure full-frame, 4k camera (8-9 megapixels), that has a compressed raw codec internally that has high frame rates (96-120fps) would be a huge hit right now. If the Komodo was pure 4k and had 96fps in that 4k, I'd buy it in a second. But, it is hard to deal with a 2x+ crop in 4k raw or astronomical file sizes in 6k.

Every manufacturer outside of Arri seems to be pushing resolution, when really many folks want high dynamic range, full-frame sensors, internal raw, and high frame rates. Strangely Sony comes closest with the FX6 (4k, full-frame, 120fps, raw to external recorder), but then their IQ is still behind every other manufacturer IMO, so not quite worth the tradeoff.

I would think Gemini was DSMC2's most popular camera. Great IQ, great low light, R3Ds at reasonable file sizes, and no crop or very little crop in what I think is the sweet spot for resolution: 4k-5k.

Hoping the next DSMC3 camera is a full-frame 4k-5k sensor, great low-light, power draw closer to 50wh, same high frame rates as Raptor in 4k-5k. Done. I would buy it in a heartbeat.

 

[Ethan_Cole]

Probably a day 1 or day 2 test for me. I have my Xyla and the 2 other things I use to look at that. Sensor, OLPF, and IQ (color, draw, etc) related tests are pretty much the first things I dig into on any camera. Likely testing all the OLPFs too.

I did a bunch of workflow tests already, which has been interesting. I'll have to post some stuff next week regarding that.

Phil, one question I have that is HUGE for me is how much quieter is the fan on the Raptor vs Monstro. I actually run a 40mm super quiet auxiliary fan on top of my camera to keep the internal camera at a low rpm. The result is a much quieter camera but it could be quieter. If the Raptor is a lot quieter THAT would be a HUGE reason for me to switch. You can sell a camera here Phil

 

[Noah Yuan-Vogel]

Hoping the next DSMC3 camera is a full-frame 4k-5k sensor, great low-light, power draw closer to 50wh, same high frame rates as Raptor in 4k-5k. Done. I would buy it in a heartbeat.

Yeah it's kind of amazing red hasn't tried that yet. If they could make something to compete with an LF mini (focus on DR at 4-5k VV with ND) at a very competitive price and with more sensitivity, red raw workflow and features, they could make a killing and dominate the market. Too bad they are so resolution obsessed at the cost of everything else, relegating their cameras more to specialty/VFX/low budget...

 

[Phil Holland]

Phil, one question I have that is HUGE for me is how much quieter is the fan on the Raptor vs Monstro.

At the airport on my way back to LA.

Simply put, it's much quieter. V-Raptor has a very different cooking system and design. New heat sinks, new fan, new thermoelectric cooling.

Too bad they are so resolution obsessed at the cost of everything else, relegating their cameras more to specialty/VFX/low budget...

What you think is resolution obsession is actually much more tied to "film-like" obsession. A 4K or 5K FF35 sensor would satisfy some, but that not what I think RED has ever been going for. Toss a S35 lens on a FF35 4K sensor and it's um.... Weird. Not to mention S16.

It's likely tough considering the growth of the market and many of the new cameras over the years, but it's likely important to remember that RED at it's core is after creating a digital film alternative in my mind. I could see a specialty camera like that, but that won't be their main bodies and sensors.

Film is an entirely different medium and nothing will ever be purely apples to apples, but in REDCODE RAW having images this clean at these sensitivies and an image that can standup to more discerning filmmakers is critical. The pixel size and design as well as the entire image pipeline is more critical than I think many here fully understand. It's often the core reason you don't see various cameras scratching the paint on higher end or professional shoots when it comes down to it.

I'm more curious if we'll ever see a sub-6K camera at this point in the timeline, even from a variety of manufacturers. One caveat. I was actually pretty surprised by Gemini. And people do love it. Hard to really know.

 

[Bobby Farokhzad]

I have been shooting on Monstro for over two years and I absolutely love the camera and the images it produces. It's not perfect and many things can be made better but it is a real powerhouse of a camera. The way I see it is it really comes down to the data rates. 800 MB/S might not sound like a big improvement over the 300 MB/S Monstro currently taps out at but it is, it really is. I don't know what HQ MQ and LQ equate to in terms of DSMC2 compression ratios and maybe someone from RED could chime in and give us the equivalent values but if the V Raptor can do full 8K at 60fps at near 3:1 or 4:1 compression that right there is a reason to upgrade for many of us. I for one (and I might be in the minority here) would gladly pay for extra storage in order to get my images as "uncompressed" as possible.

 

[Christoffer Glans]

Too bad they are so resolution obsessed at the cost of everything else, relegating their cameras more to specialty/VFX/low budget...

I'm pretty tired of the for vs against resolution debates going on. Both sides are wrong and right at the same time. The ones arguing for large resolutions craze about higher and higher resolutions but ignore factors like file sizes, compression, quality pixels, scan rates, frame rates, and so on. While those arguing against large resolutions ignore things like supersampling, better noise handling, reframe/stabilization purposes etc.

Both sides need to realize there's a balance to this. That balance has to do with everything in the industry, not just pixel peeping the result of that or this camera. You need enough resolution as an acquisition format that doesn't fill up storage space, but don't lack the drawbacks of lower resolutions.

With the sensors we have today and the delivery formats for almost every project in the world, 4K is the standard that is going to exist for a very long time and acquisition needs to be larger than delivery. We will not jump into 8K just because there are TVs and special screens made for it because content won't be 8K. Films are still mastered in 2K, shot in 2K or supersampled to 2K. 4K requires a lot in terms of VFX horsepower and most studios rather try AI upscaling algorithms than rendering in 4K. All while gaming are struggling to lower resolution and using the same kind of AI upscaling for new consoles and PC gaming with Raytracing. Everyone is aiming for 4K as a standard, both from under that resolution and from supersampling down from above it. Anyone thinking we're about to move into 8K on a global scale is downright delusional. It will be a niche group of people that don't drive sale tickets or anything major in the industry and gaming industry.

But when we shoot for 4K delivery, disregarding the fact that plenty of VFX is done sub-4K, we want the best possible 4K image in the end. And while 4K cameras seem to be the go-to solution for 4K delivery, that's not matching up with the reality of how this imaging tech actually works. Even if people don't do any reframing or stabilizing in post (which is still a very common thing to do on any camera with the intent of delivery to cinema or streaming, regardless of resolution and even before the age of digital cinema), supersampling from higher resolutions increase the fidelity of the 4K image extremely well. It's one of the best ways of improving image quality overall and is vastly misunderstood by so many people inside and outside the industry. But it's also a hyped-up feature that's been emphasized too much by, for example, Red. 8K is the perfect resolution for acquisition, it resolves to 4K extremely well, removes noise, and packs the color data beautifully. There's no reason to go higher, and the ones arguing for higher resolutions usually just say that this was the same argument as Arri had when they argued against 4K, except that it isn't. 4K is a newborn baby as a standard, barely able to be adopted by the games industry and VFX studios, which are driving forces for delivery formats. But the ones arguing for, talk about the next 8K standard while we haven't even arrived fully into 4K. All while tests has been done over and over showing that people have a hard time resolving differences between 4K and 8K, because it's a diminishing return going up. SD vs HD is extremely easy to see differences in, HD vs 4K is less easy, 4K vs 8K is much harder to see.

People on both sides need a reality check around this subject. 4K will be a standard for a very long time and the best way to get a good 4K is to shoot 8K with as good acquisition quality as possible. Going higher doesn't really produce better results compared to improving on the 8K resolution and making such cameras better. I think everything should be 8K, I don't mind 6K, still works great, but I would wish that we just end all of this and just go 8K on everything. Lock it in and start improve within that resolution, more fps, better pixels, better DR, better everything until we reached the diminishing return of image quality, then make cameras smaller, less cooling dependent, weather-sealed, fast workflows, better compression, new features (as mentioned before) never seen in any camera and so on.

I'm tired of hearing about resolution from both sides. Red proved the point with 8K, it's done, we arrived, focus on it, let's talk about other things that need attention in digital cinema technology.

 

[Björn Benckert]

One practical difference between Monstro and V-Raptor is data rates. The DSMC2 line uses wavelet compression, DSMC3 - unofficially Komodo and Raptor - uses a cosine based codec. Wavelets allow you to stuff a lot of image information into a 300MB/s bitstream. To package a similar amount of image metrics into a DCT codec requires roughly 800MB/s using current topologies. Wavelet based codecs like JPEG2000 or RedCode are efficient users of bandwidth, but require serious computational muscle to decode. DCT is less efficient, but much easier to decode.

So, which is better - less data coupled with higher client side hardware requirements, OR, more data but easier to decode on commodity hardware? RedCode was a key technology in developing digital cinema cameras able to deliver high resolution RAW imagery at data rates commensurate with available media/processing tech in 2007. I would argue it was still the best choice ten years later. In 2021, its a more nuanced issue.

With storage costs always declining (outside of COVID supply chain issues) - and much faster bus standards like USB 3.2 and TB3 (not to mention USB4 and TB4) - and CF Express variants capable of >800MB/s sustained writes hitting the market, RED went DCT. For some RedUsers, this change is not welcome - for others, the ability to work fluidly with RED files on modest computer systems is worth the extra storage requirements.

Bottom line - there's no free lunch, just trade offs - like most engineering ;-)

Cheers - #19

The new regim comes with huge problems for a lot of post houses that use 10gb e netowork. Simply put the Monstro files plays nicely over network the Raptor files... not so much.

Having the material on fast local storage is easy peasy when sitting on one machine editing, grading or doing VFX work. But when projects gets bigger and you start to add multiple computer heads then monstro files are simply way easier to handle. We play them full res no problem of our server towards at least 6 users simultainusly, if we where dealing with raptor files we would need to totally upgrade the infrastructure.

And as you write, Monstry / DSMC2 type compression takes a beast to compress and decompress the komodo type is easer to create and read but needs far more Mb/s to deliver the same quality. So the 800MB/s raptor file I doubt is twice as good then monstro compression. I would not be strange if they are on par.

 

[Michael Lindsay]

+ 1

we use 10gig and monstro decodes on our fastest machine (at full full quality 8k) 2-3 x RT... and as Björn says like for like Wavleets will delver the same quality in a much smaller package! I hope the Raptor XP offer both compression types..

The new regim comes with huge problems for a lot of post houses that use 10gb e netowork. Simply put the Monstro files plays nicely over network the Raptor files... not so much.

Having the material on fast local storage is easy peasy when sitting on one machine editing, grading or doing VFX work. But when projects gets bigger and you start to add multiple computer heads then monstro files are simply way easier to handle. We play them full res no problem of our server towards at least 6 users simultainusly, if we where dealing with raptor files we would need to totally upgrade the infrastructure.

And as you write, Monstry / DSMC2 type compression takes a beast to compress and decompress the komodo type is easer to create and read but needs far more Mb/s to deliver the same quality. So the 800MB/s raptor file I doubt is twice as good then monstro compression. I would not be strange if they are on par.

 

[Blair S. Paulsen]

(snip) And as you write, Monstry / DSMC2 type compression takes a beast to compress and decompress the komodo type is easer to create and read but needs far more Mb/s to deliver the same quality. So the 800MB/s raptor file I doubt is twice as good then monstro compression. I would not be strange if they are on par.

I would posit that 300MB/s of RedCode (JP2K variant) is roughly equal to 800MB/s DCT. Deep dive testing would be required to properly discriminate between them. IAC, I doubt the Raptor has the internals to do both. For networked post facilities I see two workarounds:

1) have crews keep under a 600MB/s data rate at capture so 10GigE can keep up. This might be as simple as sticking to MQ, but a few simple tests should sort out real numbers.

2) dump selects onto local storage as needed. 800MB/s over 10GigE may not be real time on real world infrastructure, but it shouldn't be worse than 1.3x real time unless your network is slammed.

Cheers - #19

 

[Brian Timmons]

One practical difference between Monstro and V-Raptor is data rates. The DSMC2 line uses wavelet compression, DSMC3 - unofficially Komodo and Raptor - uses a cosine based codec. Wavelets allow you to stuff a lot of image information into a 300MB/s bitstream. To package a similar amount of image metrics into a DCT codec requires roughly 800MB/s using current topologies. Wavelet based codecs like JPEG2000 or RedCode are efficient users of bandwidth, but require serious computational muscle to decode. DCT is less efficient, but much easier to decode.

So, which is better - less data coupled with higher client side hardware requirements, OR, more data but easier to decode on commodity hardware? RedCode was a key technology in developing digital cinema cameras able to deliver high resolution RAW imagery at data rates commensurate with available media/processing tech in 2007. I would argue it was still the best choice ten years later. In 2021, its a more nuanced issue.

With storage costs always declining (outside of COVID supply chain issues) - and much faster bus standards like USB 3.2 and TB3 (not to mention USB4 and TB4) - and CF Express variants capable of >800MB/s sustained writes hitting the market, RED went DCT. For some RedUsers, this change is not welcome - for others, the ability to work fluidly with RED files on modest computer systems is worth the extra storage requirements.

Bottom line - there's no free lunch, just trade offs - like most engineering ;-)

Cheers - #19

I sort of have a preference for wavelet myself and think the performance on most modern systems generally made playback a non issue.

Wonder if RED is taking longer view of things using DCT particularly where it relates to getting higher performance (resolution, frame rates) down the line.

One simple question is what would it take RED to get similar V-Raptor performance if they stayed using wavelet?
What would be the effect on the size, weight, power requirements, and heat dissipation/cooling requirements? Would it be anywhere near as quiet?

I think that would likely answer why the move to DCT.

I doubt Komodo would get 90 minutes on a single Canon BP battery or would even have such a responsive user interface and experience if it was based on wavelet. Despite the small size of Komodo, it's clearly the most quiet camera RED ever made.

It's likely that the overhead to perform wavelet compression and the hardware needed creates other performance issues and limitations that RED felt a need to move away from.

Very much agree Blair with your statement on trade offs.

Brian Timmons
BRITIM/MEDIA

 

[Blair S. Paulsen]

Good points Brian. I agree that the DCT encoder likely has a lower TDP than a wavelet encoder. Could be wrong, but pretty sure the complexity of wavelet encoding was a key impetus for spinning an ASIC for Epic and DSMC2. Furthermore, I suspect that decoding DCT has far better support in modern GPUs since its used by such a broad range of imaging applications.

Side note: the DCI pipeline uses wavelets (J2K variant IIRC) for the DCP sent to cinemas largely to support 12bpc color at data rates viable when the spec was established - 250mb/s (roughly 32MB/s). Its cake these days, but early installations required what was then a high spec server for playout.

Cheers - #19

 

[Brian Timmons]

Good points Brian. I agree that the DCT encoder likely has a lower TDP than a wavelet encoder. Could be wrong, but pretty sure the complexity of wavelet encoding was a key impetus for spinning an ASIC for Epic and DSMC2. Furthermore, I suspect that decoding DCT has far better support in modern GPUs since its used by such a broad range of imaging applications.

Side note: the DCI pipeline uses wavelets (J2K variant IIRC) for the DCP sent to cinemas largely to support 12bpc color at data rates viable when the spec was established - 250mb/s (roughly 32MB/s). Its cake these days, but early installations required what was then a high spec server for playout.

Cheers - #19

The comparison with J2K is quite appropriate. Perhaps the early drive for wavelet compression had more to do with the media limitations of the day in regards to media size, TRUE sustained data rates, and pricing.

With the advances in storage particularly with CFAST and CF Express, RED is leveraging that speed to free up the need for more heavy duty processing and hardware (i.e expensive custom ASICS). Perhaps this latest line of DSMC cameras uses this approach as a basis for how they are designed.
You already said this in your previous post. I'm just agreeing with you here.

Yes the data is more than we are used to but drive capacity and speed and even network speeds have also increased substantially.
10TB is the new 1TB.

Brian Timmons
BRITIM/MEDIA

 

[Christoffer Glans]

Yes the data is more than we are used to but drive capacity and speed and even network speeds have also increased substantially.

While this is true, the problem is this...

if we where dealing with raptor files we would need to totally upgrade the infrastructure.

People like Björn, involved in a medium-sized post-house need to unnecessarily upgrade the infrastructure just to handle Raptor files. The question is how much gain you actually get with Raptor when you already own Monstros. Maybe for high-speed content, even cropping down for extra fps, but other than that, the surrounding upgrade costs for Raptor are the deciding factor people really need to think about. Red's concern isn't with how our workflows cost, it's with making the best cameras possible. And since most other companies already use a lot of heavy codecs I think their reasoning was that the benefits of lesser computational complexity and less power-hungry systems outweigh post-production workflow demands since the competition is already worse in that regard.

But for the smaller studios and post houses, it is definitely an important question. For myself, having a DSMC2 6K and a Komodo 6K, I can feel the reasoning while shooting. The DSMC2 is larger, more cumbersome and generally a pain to just grab and shoot with compared to the Komodo, but every time I shoot with Komodo I dread the file sizes it creates.

The change in compression, even though it enables a smaller camera with better cooling, less power draw for the performance, and better performance, is still the biggest drawback on the Raptor. It's very important that people look into the entirety of their workflow before deciding if they should get it. I would even argue that for some, the Monstro is a far better camera, especially if you don't need the extra high-speed fps.

 

[Dan Kanes]

Interesting waking thought:

Redcine-X / Redline-X could add a transcode tool from DCT R3D to Wavelet R3D.

R3D stills come to mind here - perhaps one of my favorite things are those R3D frame grabs…

 

[Phil Holland]

Interesting waking thought:

Redcine-X / Redline-X could add a transcode tool from DCT R3D to Wavelet R3D.

R3D stills come to mind here - perhaps one of my favorite things are those R3D frame grabs…

That would result in a generational loss IMO. You can still do R3D snapshots/stills now whether it's Wavelet or DCT based. Not really clear why this would be a thing.

 

[Dan Kanes]

That would result in a generational loss IMO. You can still do R3D snapshots/stills now whether it's Wavelet or DCT based. Not really clear why this would be a thing.

I’m merely pointing out the existing ability of generating an R3D still from an R3D (which I believe was always wavelet in the past) - means the structure exists to already output a wavelet R3D format in the software.

Why would anyone want to transcode from DCT R3D to wavelet R3D ?

Decreased file size, and the workflows touted by the people above me in the thread highlighting a core benefit of a Monstro workflow over a Raptor workflow.

Many of us do not see generational loss as a major issue if the quality level is “good enough” - there can be ugly generational loss - but it’s not always ugly, and in fact people are striving to replicate the generational loss of yore through film grain simulation.

Transcode from R3D to an R3D mezzanine codec to reduce bandwidth is better than transcode to prores proxy.

I could be wrong, but this was one of the core tenets of the RedRay player ? Decent looking wavelet compression for distribution.

Just one man’s opinion.