No more FF35 Scarlet?

[toke lahti]

My Toshiba laptop can encode H.264 off of its internal web cam or play back 1080p H.264 at real time. The best it can do with a 4k 12 megapixel raw file from a red is 1-2 frames per second play back and 8-10 seconds per frame encode to another format.

To make a real comparable encoding/decoding power test, one should use same amount of pixels for both codecs.

Would somebody like to make this?

1. uncompressed 4k -> r3d
2. r3d 4k -> 4k uncompressed
3. uncompressed 4k -> 4k h.264
4. 4k h.264 -> 4k uncompressed

And take time to see how many fps?

Edit: looking at your decoding speeds, you get 30+Mpx/s with h.264 and 12-24Mpx/s with r3d. So power needed to decode r3d might be something like 3x per pixel?

 

[Rick Burnett]

If Canon wants to compete with Scarlet they con easily have competitor on market next year.

From what evidence? Just because Canon as a whole is a bigger company doesn't mean they contain all the best engineers or even have the capacity to create something better than they have.

Likewise, to reverse engineer a CMOS design is not going to be easy, not to mention I am sure Red owns patents on their technologies that make these designs possible.

Don't you think Canon would have CMOS technology by now to compete? I personally doubt they can do it at the price point that comes even close to what Red is offering. Mass production capability is not limited to Canon, it's out there for anyone who is going to produce a quantity of items. Apple doesn't make all their own equipment. Red can tap into these same manufacturing chains for their products, which they probably have done to make the Scarlet Fixed so cheap in comparison.

Given the video we've been seeing for the M-X sensor, it seems to me that Red is very far ahead of the competition on light sensitivity AND efficient data transfer for ANYTHING within even a reasonable price range.

 

[toke lahti]

From what evidence?

Of course pure speculation, but if you look at history, Nikon announced D90 as first dslr with hd video in August 2008 and Canon answered this with 5Dmk2 in September 2008.

Likewise, to reverse engineer a CMOS design is not going to be easy, not to mention I am sure Red owns patents on their technologies that make these designs possible.

I think it's not very hard for a company which has been designing and manufacturing cmos sensors for a decade to take sensor under a microscope and see how it is built.
If Red really has some patens on their sensors Canon can't copy the technics right away, but anyway I don't believe there's something magical in Red's sensors that other's couldn't do.
Btw, does Red design the sensors by themselves or do they just have exclusive deal with designer/manufacturer company?
Are MysteriumXs backlit sensors?

Don't you think Canon would have CMOS technology by now to compete?

The whole idea of Red is to break the japanice cartel of camera manufacturers, which hold back new tech as long as they can to make money with old tech. For example the whole model families of sony and panasonic are crippled so they don't step to each others toes. They might have several new models already designed, but wait for the best time window to start producing them to get maximal profits. Like Canon with 5Dmk2. They had model ready just waiting for Nikon to announce a competitor.

 

[David Rasberry]

To make a real comparable encoding/decoding power test, one should use same amount of pixels for both codecs.

Would somebody like to make this?

1. uncompressed 4k -> r3d
2. r3d 4k -> 4k uncompressed
3. uncompressed 4k -> 4k h.264
4. 4k h.264 -> 4k uncompressed

And take time to see how many fps?

Edit: looking at your decoding speeds, you get 30+Mpx/s with h.264 and 12-24Mpx/s with r3d. So power needed to decode r3d might be something like 3x per pixel?

With any of the Canon cameras and with HD H.264 you are only encoding/decoding 48 Mpx per second 1920x1080p at 24fps.
With Red 4k, to play a .r3d file at 1920x1080 on my laptop means processing 288 Mpx per second at 24fps.
1 minute of video from a Canon at highest quality is 330 megabytes. A minute of REDCODE 28, the lowest quality setting at 24fps is about 1.6 gigabytes. The highest quality data and frame rates from the new cameras will be nearly 5 times that much per minute. Uncompressed raw will be 12 times that data rate per minute or more.
The point remains that the processing power required for RED's level of performance, just to move the data around, is in super computer territory. Faster than most computer workstations can manage currently. The ASIC processor that is the heart of RED's new cameras is more complex and has more circuit layers than any that has ever been made. Truly ground breaking technology. Not cheap to develop or to fabricate.

 

[Rick Burnett]

The ASIC processor that is the heart of RED's new cameras is more complex and has more circuit layers than any that has ever been made. Truly ground breaking technology. Not cheap to develop or to fabricate.

I very highly doubt this. I'm sorry but as a microelectronics engineer, chip design is driven by high quantity, and that means the fab processes that are in place have already been there and available to other people. And for this, the fabrication cost is not the problem, it's the design cost. I've worked on many high speed designs, including video cameras. Going to the fab is the easiest part. I've worked on designs so large, they needed a machine with 128G (and yes, that is gig) to time them properly.

The 5Dmk2 didn't come out in a month because Nikon had the D90 and all of a sudden Canon needed to play catch up. Given the fact the video in the D90 is so horribly bad (because I know, I own one), I wouldn't be surprised if they both knew about each other, and even Nikon knowing later since it produced such an inferior product video wise. (Still wise it is awesome though).

The technology to move lots of data fast is nothing new. Serdes technology (like your SATA drives) can be run in parallel to get massive speeds. As soon as you take something and make it dedicated in hardware, the speed increase is massive. But don't forget, Red has to play the balancing game, speed versus power too. So that will limit the size of their designs as well.

When you look at stuff like CMOS design, a lot of times you design it, you build it, you test it, you redesign, etc till you get what you want. Every aspect of the fabrication process cannot be easily simulated, especially noise, and Red already has a long time ahead of them from what they learned from the first sensor. Clearly Canon has not shown they have this ins CMOS design at the speeds Red has.

With RED it's the whole package that you are getting. It's the software and algorithms, the extremely fast sensor, the feedback, the quality. This comes with lots of testing. But for me, where I find Red's beauty is in their software. They get it.

You know, ARRI has been making amazing cameras as well, why hasn't Canon put them out of business with their amazing resources? Canon has a lot invested into their video camera lines as well so they have more to lose on their own products outselling their own products!

Because we are talking bleeding edge here. Chip design does not happen overnight, because you can't just throw more people on it, it takes time because it his highly complex. Canon, Nikon, etc have known about the Scarlet and Epic for a long time, yet, I don't see anything to compete. Clearly they've had time. Personally, I don't think they will until the mass produced Scarlets start showing up everywhere.

 

[Gavin Greenwalt]

Also just because Canon (tm) is huge doesn't mean Canon 5D team is large.

You would be surprised to find how small teams are at big companies. Lots of the team is taken up by people who don't affect the spec list. You have to write manuals for 36 languages. You have to run QA on 100s of computer specs.

Canon's 5D team might not be much larger than RED's DSMC team. After all while Canon gets to benefit from billions in stock value, they also have hundreds of products and have to turn out a new product every 12-18 months to satisfy stock holders.

 

[David Rasberry]

I very highly doubt this. I'm sorry but as a microelectronics engineer, chip design is driven by high quantity, and that means the fab processes that are in place have already been there and available to other people. And for this, the fabrication cost is not the problem, it's the design cost. I've worked on many high speed designs, including video cameras. Going to the fab is the easiest part. I've worked on designs so large, they needed a machine with 128G (and yes, that is gig) to time them properly.

ASICs are done in layers... a metal spin means that the overall design is good but one or more of the layers (of 30) needs to be changed. It is a MUCH better situation than a full re-spin. One takes a month... the other can take several months.

I have to say that 95+% of our ASIC works great. This is quite a feat considering that industry experts say this is the single most complicated ASIC ever attempted.

Jim

My main point is that processing full resolution raw images at full motion rates, up to 120 frames per second continuous to the limits of available storage space for the new Red models, requires many times the bandwidth and processor power that Canon has in its current gen products. Bandwidth and processing power increases cost, power consumption, and heat loads. Subsampling a sensor and encoding to MPEG as Canon does is much less demanding. Best Canon does to date is 8fps raw burst on an 18mpx sensor and 3fps continuous.

 

[David Rasberry]

With RED it's the whole package that you are getting. It's the software and algorithms, the extremely fast sensor, the feedback, the quality. This comes with lots of testing. But for me, where I find Red's beauty is in their software. They get it.

You know, ARRI has been making amazing cameras as well, why hasn't Canon put them out of business with their amazing resources? Canon has a lot invested into their video camera lines as well so they have more to lose on their own products outselling their own products!

Because we are talking bleeding edge here. Chip design does not happen overnight, because you can't just throw more people on it, it takes time because it his highly complex. Canon, Nikon, etc have known about the Scarlet and Epic for a long time, yet, I don't see anything to compete. Clearly they've had time. Personally, I don't think they will until the mass produced Scarlets start showing up everywhere.

Exactly. I question whether Canon will attempt to compete in this market at all as it is not their core market. Most consumers could care less and their top line cameras sales volume would not justify the investment if they stay in the under $10k market as they have historically.

 

[Jacek Szwarc-Bronikowski]

And what if... Canon makes - 5D Mark III, this year, with :

Full Frame, RAW video codec, articulated screen...

... for 2499 $ ?!

 

[Cüneyt Kaya]

And what if... Canon makes this year - 5D Mark III :

Full Frame, RAW video codec, articulated screen...

... for 2499$ ?!

then will mariah carey win an oscar next year..hehe

 

[Marcus Versén]

then will mariah carey win an oscar next year..hehe

That was just unnecessary and mean and cruel and and and..... i like it

 

[Rick Burnett]

My main point is that processing full resolution raw images at full motion rates, up to 120 frames per second continuous to the limits of available storage space for the new Red models, requires many times the bandwidth and processor power that Canon has in its current gen products. Bandwidth and processing power increases cost, power consumption, and heat loads. Subsampling a sensor and encoding to MPEG as Canon does is much less demanding. Best Canon does to date is 8fps raw burst on an 18mpx sensor and 3fps continuous.

In terms of camera ASIC's, that might be the case, I've only worked on one camera ASIC so I can't really say. And I also agree that the specs of what the Canon processors do compared with the specs of what we are expecting from the Red processors are VERY different. I mean, the fact is, I am an expert in the microelectronics field, so I can appreciate the difficulty in ASIC design. (By the way, 30 layers isn't that much. You'd be amazed at how many layers you deal with in ASIC design).

I guess it just amuses me when people are just like "well company x can just release something in no time to compete" when they don't realize how difficult designing something that works this fast AND efficiently is. Or that they could easily reverse engineer an integrated circuit. Don't get me wrong, it could be done, but it is VERY difficult and VERY expensive. Integrated circuits have many layers and you cannot see them all even with a microscope because they are on top of each other. Not to mention, if you use a different fab, all the timing and process variations will be different and I can guarantee you, by the time a company reverse engineered a sensor, and built their own the same, we'll be on like the Red Twelve or something

Canon has smart people, but money and size doesn't mean you have the smartest I've contracted with many large electronics companies and you can trust me on that.

 

[David Rasberry]

And what if... Canon makes - 5D Mark III, this year, with :

Full Frame, RAW video codec, articulated screen...

... for 2499 $ ?!

Wishful thinking.

 

[toke lahti]

With any of the Canon cameras and with HD H.264 you are only encoding/decoding 48 Mpx per second 1920x1080p at 24fps.
With Red 4k, to play a .r3d file at 1920x1080 on my laptop means processing 288 Mpx per second at 24fps.
1 minute of video from a Canon at highest quality is 330 megabytes. A minute of REDCODE 28, the lowest quality setting at 24fps is about 1.6 gigabytes.

Maybe you didn't get this:

To make a real comparable encoding/decoding power test, one should use same amount of pixels for both codecs.

I know very well Red's data rates and that it has more processing power than any dslr. I was just curious how much processing power these needs in realtion to each other.

The highest quality data and frame rates from the new cameras will be nearly 5 times that much per minute. Uncompressed raw will be 12 times that data rate per minute or more.
The point remains that the processing power required for RED's level of performance, just to move the data around, is in super computer territory. Faster than most computer workstations can manage currently. The ASIC processor that is the heart of RED's new cameras is more complex and has more circuit layers than any that has ever been made. Truly ground breaking technology. Not cheap to develop or to fabricate.

It's nice that you are so exited about this wonderful "one camera to free all artists", but this is no rocket science. To make something unheard of in silicon business would need all Jim's fortune and another one on top of that.
We are talking about dataspeeds of 1Gbit/s to little over 2Gbit/s from the ASIC. Every business grade router has this power to move data.
This means 16.6Gbit/s from the sensor, but for example displayport1.2 has data speed of 17.2Gbit/s and they are going to install that chip to every $100 GPU card on the planet.

Best Canon does to date is 8fps raw burst on an 18mpx sensor and 3fps continuous.

Hmmm,
7D can shoot 8 fps and compress that to jpg and fill the whole card with that speed. And the price is $1400. If there would be fast enough CF cards, it could do much more easily the same with raw. Like I said before that is over 400Mbit/s, so considering the price it is well in the ball park.

By the way for this codec comparison, it would be nice to add also m-jpg to the game.

 

[toke lahti]

The 5Dmk2 didn't come out in a month because Nikon had the D90 and all of a sudden Canon needed to play catch up. Given the fact the video in the D90 is so horribly bad (because I know, I own one), I wouldn't be surprised if they both knew about each other, and even Nikon knowing later since it produced such an inferior product video wise. (Still wise it is awesome though).

Can you suggest any reason why not?
Nikon's hd capability was huge news, although quality wasn't.
I too guess that at least Canon knew it was coming and made home run by timing their launch right. Otherwise there wouldn't be any media coverage headlining "Canon topped Nikon".

You know, ARRI has been making amazing cameras as well, why hasn't Canon put them out of business with their amazing resources? Canon has a lot invested into their video camera lines as well so they have more to lose on their own products outselling their own products!

Arri is in very different league moneywise (50x).
But that's the thing I was trying to say. Japanice Brands won't eat from their own cake until they have no other choise. And they are in the business for making money. Canon wouldn't get much profits out of selling 10 cameras, priced for 1 million per camera in a year.
Sony(F35) is in that league only because they are trying to sell whole hdcam-sr ecosystem to their customers.

Because we are talking bleeding edge here. Chip design does not happen overnight, because you can't just throw more people on it, it takes time because it his highly complex. Canon, Nikon, etc have known about the Scarlet and Epic for a long time, yet, I don't see anything to compete. Clearly they've had time. Personally, I don't think they will until the mass produced Scarlets start showing up everywhere.

Like you said in the end. They will come out _after_ the Scarlet has led the way. And they will try to steal market. Red has to be really quick with it's moves to beat the giants.
Nikon isn't very small company, but the economy of scale is so powerful that it took them 5 years to introduce FF after Canon.

 

[David Rasberry]

Before you can compress full resolution raw to some other format at full motion rates, you have to be able to do the A to D conversion off the sensor and capture it at full motion rates raw too. So far no consumer level DSLR camera manufacturer has demonstrated the capacity to do this. Their sensors are too slow.
Could they do it? Sure, at some cost and price point. Would they do it? Only if they perceive a market large enough to offset the development costs. Do their normal mass consumer markets care about this feature? No.
So far the only advances Canon has made on there consumer level HD video in the DSLR's is to add a few features like manual control and a selection of frame rates up to 30fps that should have been there from the beginning. The video quality hasn't improved much, and they apparently still have the same quality issues with the early next gen prototypes too from the rumors posted, so they still haven't made any really significant changes yet.
I see Canon doing the very minimum they can get by with to remain competitive in their normal consumer markets.

 

[Rick Burnett]

Can you suggest any reason why not?

That's easy, because from concept to the store shelves is not entirely controlled by Canon internally. These are VERY complicated devices and having worked on many, getting everything to function properly is not an easy task. For instance, how did the 7D get released with the bug it had that images were 'burned in' from previous images? Surely as large a company as they are they'd have tested for this?

The mythical man month, which is the title of a book, is a great explanation on how design today is not add more people to it and make it get done faster. Design doesn't work like this because in many cases the complexity with splitting up the design is far more prohibitive and cost penalized than taking longer and keeping it as one large project.

And seriously, the D90 doesn't even compare with the 5D, they aren't even in the same league of still camera.

I would imagine the problem is that the part of Canon that works on the still cameras is not the same part that works on their video cameras and until VDLSR became a desire, would have gladly kept them separated.

How do you know that Red announcing the concept of a VDSLR wasn't the catalyst for Nikon and Canon to both create such a product? Given the announcement time of Red and the existence of each product, that makes more feasible sense to me with my insight on manufacturing and timelines.

Their management, who makes decisions, might not have anticipated that the VDSLR market would be as huge as it is now. It was a gamble for anyone since the market didn't exist at all beforehand.

Not to mention that Canon or Nikon don't have a target right now. The Scarlet/Epic are not out yet so Canon or Nikon would have to guess from what Red has given us as to (a) What price point to target below and (b) what features and capability they need to hit. Once they make those targets, there are decisions in the ASIC world that you can't just go back and say 'Well, let's use a faster process, or different vendor of library cells', changes in those set a project back months.

I thank Canon and Nikon for one thing. Thanks for being a test model to see what it was we don't want so we could make sure Red knew that

 

[toke lahti]

That's easy, because from concept to the store shelves is not entirely controlled by Canon internally. These are VERY complicated devices and having worked on many, getting everything to function properly is not an easy task. For instance, how did the 7D get released with the bug it had that images were 'burned in' from previous images? Surely as large a company as they are they'd have tested for this?

The mythical man month, which is the title of a book, is a great explanation on how design today is not add more people to it and make it get done faster. Design doesn't work like this because in many cases the complexity with splitting up the design is far more prohibitive and cost penalized than taking longer and keeping it as one large project.

Software bugs appear every now and then even with the biggest companies, so they don't proof anything.
You are talking about how impossible it is to design a camera from scratch in month. I was guessing whole different thing. What if Canon had 5Dmk2 already ready for 3 months before? Let's say 100 pre-production copies made? Or 10 months before? When they analyzed market reaction to D90, they started mass production right away. If needed, they could have made small changes eg. ASIC or sensor and postponed the launch 3 months.

How do you know that Red announcing the concept of a VDSLR wasn't the catalyst for Nikon and Canon to both create such a product? Given the announcement time of Red and the existence of each product, that makes more feasible sense to me with my insight on manufacturing and timelines.

I don't know and I'm not saying that I do. Red could very well been that. Seeing that somebody's going to release big sensor motion camera for less than $20,000 might have been the wake up call.

Not to mention that Canon or Nikon don't have a target right now. The Scarlet/Epic are not out yet so Canon or Nikon would have to guess from what Red has given us as to (a) What price point to target below and (b) what features and capability they need to hit. Once they make those targets, there are decisions in the ASIC world that you can't just go back and say 'Well, let's use a faster process, or different vendor of library cells', changes in those set a project back months.

They and we know quite much about the target and they don't have to release the competitor after 1 month Scarlet shpis. They can do some adjustments and release it 3 months after. If Scarlet sells 10,000 copies within first year, they might sell 100,000. So they get 90% of potential customers. Next year Scarlet sells same amount, but they sell 1000,000.
After that all 3rd party (like lenses!) will focus on their products and because of thatScarlet will fade.

All-in-all what my message is: even if Scarlet might be revolutionary product when it ships, it might not be that half an year after it ships.Or full year. Or 18 months. And this is why Red has to be very careful to listen our weak signals about not what our needs are now, but what they will be next year and the year after that. Canon can make new model in 1 year. Red needs 2 years. Red can't win that way.
I still think the key might be to focus on lenses and making a 4/3" version.

 

[Terry VerHaar]

All-in-all what my message is: even if Scarlet might be revolutionary product when it ships, it might not be that half an year after it ships.Or full year. Or 18 months. And this is why Red has to be very careful to listen our weak signals about not what our needs are now, but what they will be next year and the year after that. Canon can make new model in 1 year. Red needs 2 years. Red can't win that way.
I still think the key might be to focus on lenses and making a 4/3" version.

As for me, I think Jim has a pretty good track record on business strategy and seems to have a pretty good handle on product development, competition and getting it right. That's the horse I am planning to bet on!!

 

[Rick Burnett]

The difference here is that Canon didn't make a revolutionary camera in terms of video with the 5D, 7D or any of their other DSLRs in terms of technology. They could have added these features before with earlier cameras without significant cost. Why they didn't I can only speculate as to not wanting to cut into their other markets, especially if they can make products that require someone to buy more than one.

Red is creating something that goes beyond conventional electronics we have access to at a reasonable price for more of the independent film maker. They've got working samples of their technology now that they've shown. In electronics manufacturer, there is an easy principle most companies follow, and that is "First to market". You get out their first and become the name that is synonymous with the technology and your product will sell itself.

First, I've never seen a product that does as much as the Scarlet is proposed to do anywhere near that price point. Canon certainly doesn't offer a lot of interchangeable features among their products past the lens and intervalometer. Most use different batteries.

Two, you don't even know if Canon technology could even come close to competing with the Red technology in terms of final use. Their sensor could be at the max of it's limit with the circuitry they use. Clearly the GH1 sensor outperforms the 7D sensor in terms of color moire patterns. They've got a ton of problems to overcome.

I see no reason why Red can't win with their current model because they offer something no one else has, upgradability. Let's see, in one year I can upgrade my brain to a S35 and keep all my modules that I have on my Scarlet Fixed? Yes please!

I think the key is focusing on a sensor that blows away the competition because this is the one spot where decisions are going to be made. I want no moire, no alias, good low light performance and drastically reduced rolling shutter problems. Seems like my checklist is good to go with the Scarlet.