Double Dragon....

[Adam Watson]

Well this poster with over 13,000 posts still thinks you're a bit mixed up. ...

I admit that I was lumpung two problems together in an unclear fashion but I believe they are still valid complaints and "other cameras have problems" isn't a good excuse. We have somewhere north of 16 Epic cameras and around 4 Scarlets so every little problem is big deal for us, and when something doesn't mesh when it seems that it should, it's a poor reflection on us even if it's a design flaw by the manufacturer. I don't see any reason why I shouldn't complain even if I know very well that they can't please everybody and make a perfectly flawless camera. that's feedback. While I believe that you personally don't have a use for those lenses people still rent them from here and the Epic is the only camera we carry that doesn't work with all of our lenses, and the Dragon sensor's even bigger size makes me worried about how many more of our lenses won't work with it. Is that not a valid complaint?

 

[Phil Holland]

and the Dragon sensor's even bigger size makes me worried about how many more of our lenses won't work with it.

That's just the thing. Shooting at 5K (and likely 6K HD) on the Dragon means your lenses should work with what we know today. This is good news if you look at the current UHD spec and focus. 6K full, 2:1, and WS may be a bit outside of the coverage of certain lenses, especially older glass. If for some reason Dragon was capped at 5K in it's sensor dimensions 6K wouldn't exist as a concept.

I think many people are going to be utilizing 5K modes on Dragon for "standard production work" for a 4K finish and using 6K when desired or needed.

 

[Shane Betts]

Ad guys? Overkill? Nooo!;-)

 

[Zack McTee]

Hey Alex.. Its actually not the sensor, it's just a simple file processing math error. You can actually see it on the first one as well in random places. We are still using the engineering debayer tools until Graeme and Rob get all the code sorted and put into RCX.

I'm glad there are smarter people in this world than me. I salute Alex for noticing this, and I salute Jared for saying "yes actually you're right," and I respect Graeme and Rob for being smart enough to actually know how to address this.

 

[Stacey Spears]

BTW, why is wavelet decompression not a GPU task?

Same reason that H.264, VC-1, and MPEG-2 decode are not GPU tasks. NVIDIA, Intel, and AMD have special fixed function HW bolted onto the GPU to decode them. e.g. AMD calls it the UVD or Unified Video Decoder.

 

[Mike P.]

Much better dynamic range shot coming soon..

Did this happen already?

 

[Bob Gundu]

Not yet. But you did miss the 1 day free upgrade to Epic Dragon yesterday.

Did this happen already?

 

[Les Dittert]

Dragon software debayer for old cameras as well ?

Dragon software debayer for old cameras as well ?

Big question : Will the newly developed 'Dragon' debayer that is coming in RCX be the debayer for present 5K and 'old' sensor work as well ?
I would assume so, and would welcome it , even if it is slower than the current software 'debayer'.

 

[Mike P.]

Not yet. But you did miss the 1 day free upgrade to Epic Dragon yesterday.

HA! With my luck I wouldn't be surprised.

 

[Wey Wang]

When you look at the Dragon's sensor size, 6K is ideal for anamorphic. It effectively gives you the height for an anamorphic image (when cropped to 4:3) that you can then de-squeeze. A common complaint of the current Epic 5K anamorphic setting is that it is heavily cropped, losing one of the benefits of shooting anamorphic.

Effectively, what RED has done is produced a sensor that can handle a variety of formats and lenses that more accurately reflects the imaging size of film while maintaining one product. Sure, certain lenses vignette on the current MX sensor, but when you consider a 1.85 or 2.40 aspect ratio, the vignetting is often not an issue.

So, you shoot 6K Ana w/ anamorphic lenses or cropping to 2.40. You shoot 5K spherical when cropping to 1.85. Or you're like Fincher, shoot at 6K while framing for 4K and give yourself tons of freedom in post.

Edit:
Additionally, RED has positioned their cameras as Digital Stills and Motion. Using still lenses, there would be no concern of vignetting. You effectively get higher resolution (at a smaller imaging size) of a 5D Mk3. You can also shoot up to 89 fps vs the mk3's 6 fps.

 

[Phil Holland]

Effectively, what RED has done is produced a sensor that can handle a variety of formats and lenses that more accurately reflects the imaging size of film while maintaining one product.

You and I are on the same page Wey. Nicely worded.

 

[Jeff Kilgroe]

I admit that I was lumpung two problems together in an unclear fashion but I believe they are still valid complaints and "other cameras have problems" isn't a good excuse. We have somewhere north of 16 Epic cameras and around 4 Scarlets so every little problem is big deal for us, and when something doesn't mesh when it seems that it should, it's a poor reflection on us even if it's a design flaw by the manufacturer. I don't see any reason why I shouldn't complain even if I know very well that they can't please everybody and make a perfectly flawless camera. that's feedback. While I believe that you personally don't have a use for those lenses people still rent them from here and the Epic is the only camera we carry that doesn't work with all of our lenses, and the Dragon sensor's even bigger size makes me worried about how many more of our lenses won't work with it. Is that not a valid complaint?

I do agree that these are valid or concerns, I wasn't trying to make excuses. I don't do much for rentals these days, but have in the past. I know how customer perception, expectations, whatever play a role. Although, I'm not totally sure I share your same level of concerns. There are always limitations to what any camera and lens combination can do. The older lenses you list predate the S35 format entirely. In most situations, 5K on the MX sensor can be covered by any lens that covers the full Super 35 aperture. The VariPrimes and Angie 25-250 were not designed to do so. They cover 4K on the M/MX sensor just fine, but it's smaller than S35, it's in line with 35mm Academy spec. The full aperture S35 window is 24.89mm x 18.66mm that needs a diagonal of 31.11mm for a full coverage circle. The MX sensor's maximum recordable are is 27.64mm x 14.58mm which has a diagonal measurement or coverage circle diameter of 31.25mm. People are often concerned with "will this lens cover 5K??". Well, any modern lens that covers full-aperture S35 is usually a safe bet.

The Optimo DP's are another example of how modern lens designs have produced restrictions going the other way. Several of these "digital oriented" lens designs are very restrictive. The Optimo DP's can't be used on any camera with a mechanical shutter mechanism because of how far they extend back toward the image plane. That means no use on any film camera, no use on an F65 if the rotary shutter option is installed... Compromises abound in this industry. I'm not saying it's right, nor am I making excuses. Just the way it is. Sometimes you have to hold your customers' hands a bit. I'm a firm believer that the customer is always right, except for when they're wrong. If they want to shoot full frame 6K on the Dragon Epic when it arrives, fine. You're the rental house, provide them with a solution. I'm sure 6K in this sized sensor won't be the last resolution increment by any means and sensors may continue to get larger, but I also expect them to continue increasing density. Where will 8K, 12K, etc.. lead us? Will we venture into 65/70mm IMAX and medium format territory? I truly hope so, I have an affinity for medium format photography.

Dragon does offer us the 5K resolution to fit within the smaller 35mm window. Looks like it's a 28.94mm diagonal. 6KFF 6144x3160 is indeed larger and yes, many cine-style lenses designed for PL mount on 35mm cinema cameras are probably not going to cover that, with the exception of some of the longer focal lengths. So, yes, there is going to be that occasional customer who will take issue with this. There are also other advantages pointed out above in this thread like the ability of the larger sensor area to provide better vertical reach for anamorphic or capturing the entire image circle of a fisheye lens.

I've found that to keep my rental customers happy, I had to be as much of a consultant and guide on their journey as an equipment supplier. Rentals were never my intended path, it's just something that happened along the way because there were times early on with the RED One, and EPIC to some extent, where rental rates were too good to pass up.

 

[Timur Civan]

When you look at the Dragon's sensor size, 6K is ideal for anamorphic. It effectively gives you the height for an anamorphic image (when cropped to 4:3) that you can then de-squeeze. A common complaint of the current Epic 5K anamorphic setting is that it is heavily cropped, losing one of the benefits of shooting anamorphic.

Effectively, what RED has done is produced a sensor that can handle a variety of formats and lenses that more accurately reflects the imaging size of film while maintaining one product. Sure, certain lenses vignette on the current MX sensor, but when you consider a 1.85 or 2.40 aspect ratio, the vignetting is often not an issue.

So, you shoot 6K Ana w/ anamorphic lenses or cropping to 2.40. You shoot 5K spherical when cropping to 1.85. Or you're like Fincher, shoot at 6K while framing for 4K and give yourself tons of freedom in post.

Edit:
Additionally, RED has positioned their cameras as Digital Stills and Motion. Using still lenses, there would be no concern of vignetting. You effectively get higher resolution (at a smaller imaging size) of a 5D Mk3. You can also shoot up to 89 fps vs the mk3's 6 fps.

THANK YOU.

 

[Jason Osterday]

Well said Wey. Could not agree more.

 

[Sanjin Jukic]

Canon and Carl Zeiss already have PL primes and zooms ready to cover 42mm+ image circle!!!

I already suggested RED to call Leica and ask for a kind of cooperation about cine PL redesign from discontinued but an extraordinary Leica R primes and zooms.

Those lenses will cover 6K without a problem!!!!

BTW. Also I asked more than two years ago for a sort of new sensor design here but unfortunately RED didn't take me serious.

And today RED is facing a tough competition that some of the competitors are beating pretty hard in many ways...just saying it!!!

 

[Björn Benckert]

Canon and Carl Zeiss already have PL primes and zooms ready to cover 42mm+ image circle!!!

I already suggested RED to call Leica and ask for a kind of cooperation about cine PL redesign from discontinued but an extraordinary Leica R primes and zooms.

Those lenses will cover 6K without a problem!!!!

BTW. Also I asked more than two years ago for a sort of new sensor design here but unfortunately RED didn't take me serious.

And today RED is facing a tough competition that some of the competitors are beating pretty hard in many ways...just saying it!!!

Question: What other current sensor do you think beats the MX?
To me the only sensor that I might consider as better preforming is the F65, but it does not really come in a coffe box, and for sure not at the same price.

Just saying, we needed a stil frame from different angle the other day for a test. So we set up the epic filming and then our (I know it's not the top dog but still) canon D650 shooting a raw snapshot next to it... then pulling the colors and looking at the two side by side I was quite shocked how far behind the D650 was. In my mind the still camera would at least deliver something that matched in quality but epic actually blows the canon out of the water even for snaps.

I think RED are working as fast as they can... the thing is now actually also some other camera manufactures are doing that as well... which is a big difference from before.

 

[Steven Caesare]

Because, as much as people refuse to believe it, wavelet decoding is not a task that lends itself to, or is truly benefited by, GPU processing. It also doesn't help when we have a number of "back-seat developers" who like to dredge up stuff via Google or whatever and they stumble across one very popular paper about wavelet decoding being done on a common GPU. It is an unfortunate paper that seems to give more false hopes and assumptions to people than anything else. Yes, it's being done, yes it can be done. It's being done at a level of compression that is not on par with what R3D is doing. The performance, even in that paper's example, is terrible when compared to what larger CPUs are doing.

It's a matter of using the right tool for the job. For now, the best tools for wavelet decoding are CPUs. Or dedicated wavelet-decoding processors in the form of the RED Rocket card, as an example. These days, GPUs are the rockstars of the computing world. People think they can do anything, but the real truth is they're quite shallow and and don't perform when stretched beyond the scope of their talent. If GPUs were the miracle computing device most people make them out to be, we would have ditched our desktop and workstation CPUs long ago in favor of nVidia Fermi and Kepler processors.

That said, I'm sure there's quite a bit of optimization and performance still to be had with R3D decoding on CPUs. And possibly with some level of GPU-assist. At the rate that R3D continues to improve and evolve, along with the new sensors like MX, now Dragon, I'm not so sure they have had the time and resources to fully dive into concentrating on ultra-optimization of their algorithms. I'm only speculating here, but typically premature optimization can be the death of many great software concepts.

Indeed Jeff, well stated.

Additionally, the issue is not just wavelet decode. The resulting RAW data must then be debayered to RGB as well. And then looks applied, etc... Depending on the applicability of that, you may have to incur roundtrip time to/from GPU land back to the CPU for further processing, and then ultimately back out to the video card for display.

In many cases, relatively "slow" trips over the I/O bus to-and-from the GPU can be far slower than simply doing the operations on-CPU and then only traversing the bus once for final look application and rendering to the monitor.

As you've stated, GPU's do some things extremely well... but that scope is relatively narrow. GPU horsepower is not a panacea for all things video. Well, at least not yet, anyway...

-sc

 

[Elsie N]

Is the Rocket a CPU or a GPU?

 

[Sanjin Jukic]

Question: What other current sensor do you think beats the MX?
To me the only sensor that I might consider as better preforming is the F65, but it does not really come in a coffe box, and for sure not at the same price.

Just saying, we needed a stil frame from different angle the other day for a test. So we set up the epic filming and then our (I know it's not the top dog but still) canon D650 shooting a raw snapshot next to it... then pulling the colors and looking at the two side by side I was quite shocked how far behind the D650 was. In my mind the still camera would at least deliver something that matched in quality but epic actually blows the canon out of the water even for snaps.

I think RED are working as fast as they can... the thing is now actually also some other camera manufactures are doing that as well... which is a big difference from before.

MX was just an extension of M sensor design actually its (X) means an extension that added the new A/D board and OLPF.

My understanding of being a leader in the field of cine digital camera design means that at every moment of time your product shouldn't be catch by the competitors.

But with MX is a far away to be that case as I said before.

Also JJ reported here several times that RED was just learning from the scratch and that of course it's the main reason that they had so many issues.

Hope that Dragon sensor design will get fixed almost all of that issues!!!!

What I wanted to say is that I expected those issues to be fixed in a time when Epic was launched.

 

[Maik Müller]

Is the Rocket a CPU or a GPU?

Per definition the chip is an ASIC, which you just can't easily replace with a CPU or GPU.
There is not just a re-labeled mainstream-gpu-core on that rocket as some people might think.

I guess its closer to reality to assume there is partially the chipset of a RED One on that PCI-card, to say it the easy way. I'm sure its a bit more complicated than that.