So Is a7 III Still The Dynamic Range King? (Not tolling, just asking)

[Mark Romero 2]

So in terms of hybrid cameras - and including the BMPCC4K even though it isn't a hybrid camera - is the a7 III still tops when it comes to dynamic range?

Purely dynamic range. Not colors, not ergonomics, not soul, not flame-proof-ness. Just DR.

I've read Sony a7 III with slog 2 has close to 14 stops of DR.

Is the BMPCC4K able to compete / beat it when shooting in RAW? (New BMRAW firmware was just released, BTW).

Will Z6 beat it once it gets ability to shoot in RAW via hdmi out? (I am reluctant to spend $1,000 getting a recorder, battery and SSD card for it).

What's the DR of the Panasonic S1 cameras like? Another site said that it had 12.2 stops when shooting HLG at ISO 400. I think V Log is a paid firmware update coming down the road, right? (Hopefully it won't be dependent upon the US Postal Service for delivery).

ALSO: I thought someone on this forum posted a few months back that the 8-bit codec of the a7 III actually helped increase dynamic range when compared to 10-bit codecs. Was I dreaming? Was I on drugs? Was my mind being controlled by @mercer cat???

(P.S. I know that Sony cameras - much like my wardrobe choices - have many serious flaws. No need to comment on either of those things here.)

[Zeng]

Have you tested it? In my testing experience none of these small cameras come closed to 14 stops. None. Even more so with Sony A7III 8bit output. It's just impossible to have 14 stops in that. 10bit log is said to carry up to 12 stops (11.5 according to cinematography.net), so... But whatever helps people feel more confidence in their tools. It's the final outcome that matters, after all.

[Mark Romero 2]

6 minutes ago, Zeng said:

Have you tested it? In my testing experience none of these small cameras come closed to 14 stops. None. Even more so with Sony A7III 8bit output. It's just impossible to have 14 stops in that. 10bit log is said to carry up to 12 stops (11.5 according to cinematography.net), so... But whatever helps people feel more confidence in their tools. It's the final outcome that matters, after all.

Thanks for the reply.

Haven't tested it myself. Still trying to decide on which camera to buy

Have seen lots of tests online though using different methods. I will definitely check out cinematography.net which I haven't seen before.

The most optimistic online tests I have seen was just about 14 stops. I don't recall seeing one under 12 stops (there might be some, just didn't see any. Again, I will check out the site you mentioned.) But different testers use different methods, and they don't seem standardized.

Part of this assumption is based on extrapolation on testing of the D750 by @Mattias Burling where he tested against BMCC (both RAW and PRORES codecs) and he felt confident that the D750 had 13 stops of DR when using flat picture profile. As my aps-c a6500 in Slog 2 has more DR than my D750 had in Flat profile, and the a7 III reportedly has more DR than my a6500, so...

Link To D750 Test by Mattias:

 

I've seen some tests of slog 2 versus 8-bit F Log, where slog 2 came out ahead, but I haven't seen any head-to-head against 10-bit F Log. Would love to see some.

[KnightsFan]

44 minutes ago, Mark Romero 2 said:

I've read Sony a7 III with slog 2 has close to 14 stops of DR.

The problem is there are many ways to measure DR. If you read "the Sony a7III has 14 stops of DR" and "the Arri Alexa has 14 stops of DR" both may be correct, but are utterly meaningless statements unless you also know how they were measured.

Many years ago, Cinema5D pegged the a7sII at like 14 stops. However, they later standardized their measurement to use SNR = 2, which gave the result of a7sII at 12. But whichever way you measure, it's ~2 stops less than the Alexa. Many members here will tell you that Cinema5D is untrustworthy, so take that as you will. I have yet to find another site that even pretends to do scientific, standardized tests of video DR.

14 minutes ago, Mark Romero 2 said:

I've seen some tests of slog 2 versus 8-bit F Log, where slog 2 came out ahead, but I haven't seen any head-to-head against 10-bit F Log.

Cinema5D puts the XT2 and XT3 at just over 11, so that confirms your finding. And again if you change your methods, maybe it will come out at 13, or 8, or 17--but in every case it should be a stop less than the a7sII when measured the same way.

52 minutes ago, Mark Romero 2 said:

I thought someone on this forum posted a few months back that the 8-bit codec of the a7 III actually helped increase dynamic range when compared to 10-bit codecs.

Bit depth doesn't necessarily correlate exactly to dynamic range. You can make a 2 bit camera that has 20 stops of DR: anything below a certain value is a 0, anything 20 stops brighter is a 3, and then stick values for 1 and 2 somewhere in between. It would look terrible, obviously, because higher bit depth  reduces banding and other artifacts. There is pretty much no scenario in which an 8 bit encoding has an advantage over 10 bit encoding of the same image.

[ade towell]

From the cinema 5d dynamic range test the Panasonic S1 had more DR - 12.2 than the Sony FS7 - 12, and that was without V-log for the S1 so I would expect it  trumps the A7iii

[Mattias Burling]

I don't know about the A7iii but my A7sii could not touch the DR of the 13stop BMCC, D750 or even the 12stop XC10 and 5Dmkiii raw. 

So I doubt the A7iii could match a BMPC4K. At least I haven't seen any footage that points to it being able to. Would love it if it could. Maybe then I could justify it's imo to high price.

[Zeng]

It will always be a bit subjective in the shadows range. Just look at Geoff Boyle's tests. He has a comment section where he rates each of those high end cameras, with number of stops over and under the charts. With highlight clipping it's easier, although I've seen discussions where some even question this part, seeing 1/4 of a stop more, etc. In practical applications these hardly matter. In the shadows Geoff is ruthless - he discards anything where the color changes or the noise becomes apparent. For instance Alexa @ 800ISO is 5 stops over and 2 stops under in his tests. I'd probably find acceptable another stop in the underexposure. And so on. But in 8bits... you can try cramming those stops, but there's just not enough values per stop at that depth, etc. However, many find those log curves in 8bits acceptable and use them all the time. Whatever works for you, I guess.

[androidlad]

It you're solely talking about video then DR is between 11-12 stops, since the moving picture readout drive mode is limited to 12bit ADC.

Panasonic S1 lab test exceeded 12 stops, but it's really just heavy NR cheating the charts, which don't reveal detail loss in real world shooting scenarios.

1 hour ago, Zeng said:

Have you tested it? In my testing experience none of these small cameras come closed to 14 stops. None. Even more so with Sony A7III 8bit output. It's just impossible to have 14 stops in that. 10bit log is said to carry up to 12 stops (11.5 according to cinematography.net), so... But whatever helps people feel more confidence in their tools. It's the final outcome that matters, after all.

10bit log encoding can efficiently store up to 16 stops of dynamic range, that's why most film scans have been using 10bit cineon format.

[Zeng]

Probably you're right, 11-12 stops limited by ADC.

[Mako Sports]

I would guess 12 if your properly exposing Slog 2

[Mark Romero 2]

Firstly, thanks to everyone that has responded.  You've all contributed knowledge without starting a war, which makes me very happy to be a member of this forum.

Will respond individually to everyone's comments when I have a bit more time later today.

 

[tupp]

2 hours ago, Mark Romero 2 said:

ALSO: I thought someone on this forum posted a few months back that the 8-bit codec of the a7 III actually helped increase dynamic range when compared to 10-bit codecs.

2 hours ago, Zeng said:

Even more so with Sony A7III 8bit output. It's just impossible to have 14 stops in that. 10bit log is said to carry up to 12 stops (11.5 according to cinematography.net), so...

1 hour ago, androidlad said:

It you're solely talking about video then DR is between 11-12 stops, since the moving picture readout drive mode is limited to 12bit ADC.   [snip]    10bit log encoding can efficiently store up to 16 stops of dynamic range, that's why most film scans have been using 10bit cineon format.

Folks, just a friendly reminder:  Dynamic range and bit depth are two different and  completely independent properties.  A change in dynamic range does not affect bit depth, and vice versa.

 

 

1 hour ago, KnightsFan said:

Bit depth doesn't necessarily correlate exactly to dynamic range.

Bit depth and dynamic range do not correlate -- period.  They are completely different properties.

 

 

1 hour ago, KnightsFan said:

You can make a 2 bit camera that has 20 stops of DR: anything below a certain value is a 0, anything 20 stops brighter is a 3, and then stick values for 1 and 2 somewhere in between. It would look terrible, obviously, because higher bit depth  reduces banding and other artifacts.

Exactly.

 

You can also make a 16-bit camera that has 2 "stops" of DR.  In addition, you can make a camera with a given dynamic range that allows one to choose 8-bit, 10-bit or 12-bit depths (many such cameras actually exist).

 

 

1 hour ago, KnightsFan said:

There is pretty much no scenario in which an 8 bit encoding has an advantage over 10 bit encoding of the same image.

I can think of a few scenarios in which an 8-bit encoding would be more desirable than a 10-bit encoding of the same image -- if the 10-bit encoding has a lower resolution than the 8-bit encoding, the 8-bit image could exhibit more detail.

 

Furthermore, if the resolution of the 8-bit version has over four times the resolution of the 10-bit version, the 8-bit version will have more color depth than the 10-bit version.  Keep in mind that color depth and bit depth are not the same:  color depth = bit depth x resolution

 

Also, an uncompressed 8-bit version could very well have an advantage over a highly compressed 10-bit version.

[androidlad]

1 minute ago, tupp said:

Folks, just a friendly reminder:  Dynamic range and bit depth are two different and  completely independent properties.  A change in dynamic range does not affect bit depth, and vice versa.

Bit depth and dynamic range do not correlate -- period.  They are completely different properties.

Exactly.

You can also make a 16-bit camera that has 2 "stops" of DR.  In addition, you can make a camera with a given dynamic range that allows one to choose 8-bit, 10-bit or 12-bit depths (many such cameras actually exist).

I can think of a few scenarios in which an 8-bit encoding would be more desirable than a 10-bit encoding of the same image -- if the 8-bit encoding has a higher resolution than the 10-bit encoding, one gets more detail.

Furthermore, if the resolution of the 8-bit version has over four times the resolution of the 10-bit version, the 8-bit version will have more color depth than the 10-bit version.  Keep in mind that color depth and bit depth are not the same:  color depth = bit depth x resolution

Also, an uncompressed 8-bit version could very well have an advantage over a highly compressed 10-bit version.

Do you know what ADC is? ADC bitdepth directly determines how much DR can be extracted from the sensor. Sony sensors use single 12bit ADC for video. The new BMD 4.6K G2 uses dual 11bit ADC, ARRI uses dual 14bit ADC.

[tupp]

7 minutes ago, androidlad said:

ADC bitdepth directly determines how much DR can be extracted from the sensor

No, it doesn't.  Dynamic range and bit depth are completely independent properties.

 

Now, an ADC might be part of the signal pipeline that constricts DR, but that has nothing to do with bit depth.

 

 

7 minutes ago, androidlad said:

Sony sensors use single 12bit ADC for video. The new BMD 4.6K G2 uses dual 11bit ADC, ARRI uses dual 14bit ADC.

Again, the bit depth of those ADCs don't have anything to do with the DR.  Those ADCs could just as well be 4-bit with the same DR.

 

By the way, 16-bit ADCs exist.

[KnightsFan]

12 minutes ago, tupp said:

Bit depth and dynamic range do not correlate -- period.  They are completely different properties.

They are not tied to each other, exactly. I said "correlate" and "necessarily" because in real world, manufacturers usually add more bits to higher DR images to avoid artifacts. So they usually do correlate in the real world, but only because of convention and not because of some intrinsic property.

12 minutes ago, tupp said:

if the 8-bit encoding has a higher resolution than the 10-bit encoding, one gets more detail.

 

12 minutes ago, tupp said:

Also, an uncompressed 8-bit version could very well have an advantage over a highly compressed 10-bit version.

True, I implicitly encompassed both of those factors into saying an encoding of the same image. I should have been more specific.

8 minutes ago, androidlad said:

Do you know what ADC is? ADC bitdepth directly determines how much DR can be extracted from the sensor. Sony sensors use single 12bit ADC for video. The new BMD 4.6K G2 uses dual 11bit ADC, ARRI uses dual 14bit ADC.

Yes, the ADC bit depth does limit the DR assuming its linear, but the encoded image might not retain all that range in the final image.

[androidlad]

2 minutes ago, tupp said:

No, it doesn't.  Dynamic range and bit depth are completely independent properties.

 

Now, an ADC might be part of the signal pipeline constricts DR, but that has nothing to do with bit depth.

 

 

Again, the bit depth of those ADCs don't have anything to do with the DR.  Those ADCs could just as well be 4-bit with the same DR.

 

By the way, 16-bit ADCs exist.

It does. sensors respond to light in linear way. 12bit quantisation results in 12 stops of theoretical maximum DR.

ADCs have everything to do with DR. That's way ZCAM E2 has a low-noise mode that takes advantage of a 14bit ADC.

16bit ADC do exist but the framerate is in single digit at the moment.

[tupp]

17 minutes ago, androidlad said:

It does. sensors respond to light in linear way. 12bit quantisation results in 12 stops of theoretical maximum DR.

There are a zillion different 12-bit capable ADCs/cameras, but they all don't have the same dynamic range.  You can map 12-bits (linear, log or whatever) to 200 "stops" of dynamic range or map 12-bits to 3 "stops" of dynamic range.  Dynamic range and bit depth are two different completely independent properties.

 

 

17 minutes ago, androidlad said:

ADCs have everything to do with DR. That's way ZCAM E2 has a low-noise mode that takes advantage of a 14bit ADC. 

The E2's low noise mode may take advantage of the 14-bit depth (I would say that the E2's 14-bit depth takes advantage of it's low noise mode), but, nevertheless,  ZCAM could have just as easily mapped 8-bit, 10-bit or 12-bit to that low noise mode.  Bit depth and DR are independent properties.

 

 

17 minutes ago, androidlad said:

16bit ADC do exist but the framerate is in single digit at the moment.

Okay.  I haven't kept up with the progress (doesn't sound like there has been much 16-bit progress in the last ten years).

[androidlad]

8 minutes ago, tupp said:

There are a zillion different 12-bit capable ADCs/cameras, but they all don't have the same dynamic range.  You can map 12-bits (linear or log) to 200 "stops" of dynamic range or map 12-bits to 3 "stops" of dynamic range.  Dynamic range and bit depth are two different completely independent properties.

 

 

The E2's low noise mode may take advantage of the 14-bit depth (I would say that the E2's 14-bit depth takes advantage of it's low noise mode), but, nevertheless,  ZCAM could have just as easily mapped 8-bit, 10-bit or 12-bit to that low noise mode.  Bit depth and DR are independent properties.

 

 

Okay.  I haven't kept up with the progress (doesn't sound like there has been much 16-bit progress in the last ten years).

12bit ADC places an upper limit on the DR, which is 12 stops. Current silicon based sensors work in linear, so 1bit=1 stop.

Again, E2's low noise mode engages the 14bit ADC which is a hardware on the sensor, the video files are encoded with 10bit ProRes, which I'd say holds between 12-13 stops DR.

ARRI uses dual 14bit ADC to merge into a single 16bit stream, this is mapped to 12bit log for its 14+ stops of DR.

It seems that you're confusing linear ADC bitdepth with encoding curve bitdepth which can be log or linear or anything in between (for example, BRAW is non-linear RAW)

[Zeng]

OK.. seems like it's somewhat sensitive thing.. To get back into perspective, what usually matters to cinematographers is the picture they get. And in this particular case I just wanted to say that none of 8 bit cameras I tested looked good enough with their log picture. While Nikon Z6 and Fuji X-T3 look really good. But this is subjective guys, I'm looking at the picture, testing where it breaks, evaluating this for myself, and only then looking at what numbers/bits it has. Let's not start a war over it. In normal conditions with careful exposing one can make any of these 8 bit camera look just great, no questions.

[androidlad]

16 minutes ago, Zeng said:

OK.. seems like it's somewhat sensitive thing.. To get back into perspective, what usually matters to cinematographers is the picture they get. And in this particular case I just wanted to say that none of 8 bit cameras I tested looked good enough with their log picture. While Nikon Z6 and Fuji X-T3 look really good. But this is subjective guys, I'm looking at the picture, testing where it breaks, evaluating this for myself, and only then looking at what numbers/bits it has. Let's not start a war over it. In normal conditions with careful exposing one can make any of these 8 bit camera look just great, no questions.

The Possession of Hannah Grace, shot on A7S II in 8bit, looks gorgeous.