Full Frame Scarlet

[Subhadip Sen]

If you doubt the Scarlet's low light abilities, do download the lion R3D and push it to ISO 6400.

[David Rasberry]

Flawed assumption.

Mysterium sensor in RED ONE: 8 - 10 stops.
Mysterium-X sensor in RED ONE: 13.5 stops.
Mysterium-X sensor in EPIC with HDR-X: up to 18 stops.

All those sensors are the exact same size. Semiconductor design counts far more than pixel density.

There is a difference between sensitivity and dynamic range. You could have a 100 ISO sensor with 13 stops of dynamic range and a 2000 ISO sensor with 7 stops.
Seems like most current gen midrange priced digital video cameras 2/3" and larger have a native ISO rating around 200-400 range, but dynamic range varies considerably from around 9 stops to 13+.
More dynamic range pretty much comes from lower noise in the sensor and associated electronics, not increased sensitivity to light. Since you are generally working toward a display dynamic range of about 7 stops, the extra DR gives you some room to underexpose without incurring an unacceptable noise level when gain is boosted in camera or in post.

[Stephen Gentle]

There is a difference between sensitivity and dynamic range. You could have a 100 ISO sensor with 13 stops of dynamic range and a 2000 ISO sensor with 7 stops.
Seems like most current gen midrange priced digital video cameras 2/3" and larger have a native ISO rating around 200-400 range, but dynamic range varies considerably from around 9 stops to 13+.
More dynamic range pretty much comes from lower noise in the sensor and associated electronics, not increased sensitivity to light. Since you are generally working toward a display dynamic range of about 7 stops, the extra DR gives you some room to underexpose without incurring an unacceptable noise level when gain is boosted in camera or in post.

True. But the OP was talking about low light performance, not just sensitivity or DR. I just used the dynamic range measurements as an example.

My point was basically that better electronics and processing make far more difference* than simply lowering the pixel density (that is, having the same number of pixels on a larger sensor or less pixels on the same sized sensor) when it comes to low light performance, and I think that still stands.

[* at the differences in size we're talking here, at least. There would be a point where simply having the same design on a larger sensor makes a noticable improvement, but I'd expect it would be at a fair bit larger difference. ]

[David Rasberry]

True. But the OP was talking about low light performance, not just sensitivity or DR. I just used the dynamic range measurements as an example.

My point was basically that better electronics and processing make far more difference* than simply lowering the pixel density (that is, having the same number of pixels on a larger sensor or less pixels on the same sized sensor) when it comes to low light performance, and I think that still stands.

[* at the differences in size we're talking here, at least. There would be a point where simply having the same design on a larger sensor makes a noticable improvement, but I'd expect it would be at a fair bit larger difference. ]

Agree. Daniel Browning made the case some time back that for pixel sizes in the range we are talking about, 3.3-6 microns, there is little practical difference in real life performance. Total light gathering power scales more with total sensor area than with pixel size.

[Lee Saxon]

I have no problem with the claim that the 9mp/cm2 sensor is going to have good low light sensitivity. I'd even accept "as good as anyone really needs." But tell me "as good as the 3mp/cm2 sensor," and I'm gonna need to see it to believe it. Photosites three times as big are going to collect a lot more light. I don't see how you can get around that. I don't understand those disputing that.

If you don't believe me go point a D300 and a D700 at the same thing.

[Steve Wake]

If you don't believe me go point a D300 and a D700 at the same thing.

No argument there. With same gen sensors and processing the larger sensor and pixels will win.

A better comparison that matches the point made by others would be a D7000 at ISO 100 vs. a D700 at ISO 200 (both at their sweet spot).

D7000 with smaller pixels plus next gen processing is equal IMHO. D7000 at ISO 200 is inferior only under extreme conditions, and not by much.

And that's a borderline consumer/prosumer version of the sensor and ASICs vs. an earlier generation pro (equal to D3) version.

Technological progress marches on ....

[David Rasberry]

I have no problem with the claim that the 9mp/cm2 sensor is going to have good low light sensitivity. I'd even accept "as good as anyone really needs." But tell me "as good as the 3mp/cm2 sensor," and I'm gonna need to see it to believe it. Photosites three times as big are going to collect a lot more light. I don't see how you can get around that. I don't understand those disputing that.

If you don't believe me go point a D300 and a D700 at the same thing.

The argument that Dan Browning made is that the larger pixel might have greater capacity as individual pixels, but for a pixel of half the width, 4 pixels in the same area would have lower sample noise at the margin because you get 4 samples for the same area vs 1. One offsets the other. Within that narrow a size range there isn't much difference in the net results if everything else is the same
Comparing any two cameras of different model and generation doesn't really tell you much, especially cheap cameras, because nothing is ever the same. You can't attribute it to the difference in pixel size alone.
However as the overall sensor area gets larger, there is a greater gain in light sensitivity over total noise. If the pixel size stays the same a sensor twice the diameter and 4 times the area should show about a 1 stop gain in sensitivity at the same relative aperture due to collecting more light(inverse square law). I expect we will see about 1-1/2 stop difference in sensitivity for Scarlet vs Epic, but about the same dynamic range overall.

[Robert Jackson]

If the pixel size stays the same a sensor twice the diameter and 4 times the area should show about a 1 stop gain in sensitivity at the same relative aperture due to collecting more light(inverse square law). I expect we will see about 1-1/2 stop difference in sensitivity for Scarlet vs Epic, but about the same dynamic range overall.

Yeah, exactly. And I'm sure someone will really need that extra stop, but that's why you go over the specials when you're putting together a budget and schedule. There are always some specials and the most economical way to deal with those is to rent a camera that can handle the problem for the day(s) you're shooting in a cave or whatever. You don't walk around with a high-speed camera for the whole shoot because you're going to smash a glass of water against a wall on day 11.

The full-frame Scarlet/Epic is going to be amazing, in all likelihood. So are the Epic 645 and 617. But there's this mindset that's really pervasive anymore about how you need to go out and buy something that will do everything you ever imagined needing to do and then you're somehow complete. In communications theory they called it the "magic box." If you get what's in the magic box all your problems are solved and you are completed as a human being. That's the core of advertising. It slices, it dices...it even makes mounds of julienne fries! You've got all the bases covered!

But really it's about having the bases covered that you need covered the other 98% of the time. When I bought the camera pictured below I was sure it would do everything. I'd seldom ever use 16mm again. I had super-speed optics and a much larger "sensor" and it was very quiet and all was well with the world. And uh...it's heavy (I eventually gave myself a herniated disc using it), much more difficult to focus than 16mm and the format drives up costs to the point that it's difficult to find gigs that make financial sense. And so I have several 35mm cameras that I can use for this or that (establishing shots on 35mm are nice for the added detail), but I still use 16mm more than anything. YMMV.

BTW, it's the same thing with lenses. I love exotic glass. There are lenses out there that are designed and built to cause people to lose sleep. Like Ultra Primes. The 8mm rectilinear is one of them. So's the 180mm T1.9. That lens is like optical crack cocaine. There are all kinds of fetish items out there. And the reputations they develop are usually warranted. But you can do a lot of amazing work on "mediocre" lenses. You don't have to go all Ozu or anything, but realizing that an Ultra Prime isn't the only thing keeping you from being a master cinematographer is good, too.

[David Rasberry]

I think the quest for ultra high ISO sensitivities beyond about 1600-2000 has little relevance to most day in day out shooting conditions. We have lived with film stocks that for the most part top out at 500 ISO rating, and typically are good for about 2 stops push processing with decent results.
Too much sensitivity can make shooting under daylight conditions difficult, a base ISO sensitivity in the 400-800 range with enough DR to get 3+/- stops of latitude for varying conditions would seem to me very satisfactory. These aren't parking lot security cameras.

[Robert Jackson]

I think the quest for ultra high ISO sensitivities beyond about 1600-2000 has little relevance to most day in day out shooting conditions. We have lived with film stocks that for the most part top out at 500 ISO rating, and typically are good for about 2 stops push processing with decent results.
Too much sensitivity can make shooting under daylight conditions difficult, a base ISO sensitivity in the 400-800 range with enough DR to get 3+/- stops of latitude for varying conditions would seem to me very satisfactory. These aren't parking lot security cameras.

Yeah. If I can get up to 320 or even 500 and still retain relatively wide DR, particularly in regard to highlights...well, I'm fine. The thing that kills digital for me so many times is that weird color skew as the highlights approach the point of clipping. A lot of times patches of skin get a strange yellow or salmon cast around a shiny cheek or forehead and it pulls me right out of the moment. If the highlights can approach clipping in a less video-y way I'm pretty much cool with most any kind of rational limitation in regards to maximum usable ISO. Most of what I've seen from the first generation of Red sensors has been pretty good in that regard. Hopefully the newest generation will be even better, particularly with the new HDR abilities.

BTW, I don't know what they're using to shoot The Colbert Report, but that show has the most surreal digital-y skintones I've ever seen. Sometimes it looks like he's wearing a plastic mask in wide shots. The Daily Show is almost as bad. We watch one or the other of them from the DVR before going to sleep at night and I frequently feign horror and hide my face when the skintones look impossibly inhuman.