ahalpert
Major Contributor
Hmm i suppose you're right
Sony A7SIII = $3500, 10-bit 4K120p internal, 16-bit 4K60p RAW out
- Thread starter ahalpert
- Start date
Hmm i suppose you're right
Trek of Joy
Member
From the tests I've seen other than the slight crop at 120p, detail in all 4k modes is pretty much indistinguishable. Its just a hair less sharp than the a73 and the 60/120 is more detailed than the R5. Gordon Liang posted a SOOC Slog3 shot, gonna tinker with that. Hopefully more footage surfaces soon.
Chris
Do you have a link to Laing's SOOC Slog3 video sample. Can't find at any of his web sites, or Flickr.
Thanks.
ahalpert
Major Contributor
thanks
Carlos Morillo
Active member
Edited, to make a short version. Don't worry about 16 Bits on the A7SIII.
Now the long version.
I'm not 100% sure but I wouldn't worry to much about the 16 bits vs 14 bits. If you look at it from the stills perspective, even on medium format cameras, there have been discussions about how valuable the 16 bits are vs 14 bits. No CMOS sensor that's readily available that I'm aware of even if you look at a PhaseOne, GFX100 or on the video side of things like the Alexa 65, will give you 16 stops of clean noise free dynamic range. At a certain point the ADC converter's bit depth is exceeding what the CMOS can capture and you are basically just getting more steps or a more accurate readout of noise without that much benefit to signal. Some people argue the difference between 14 and 16 bits is huge and other's will say they can't see the difference just a bloated file. Either way I doubt it would make much difference on the A7SIII. I believe there was some drama not with the 150mp IQ4 or 102 mp GFX 100 but with the Hasselblad X1D and the GFX 50s/r. They are both using the same sesnsor and apparently both doing 14 bit readouts but for some reason Hasselblad chose to to still process/write in 16 bits and Fujifilm stuck with 14 bits. People argued the merits of the "phantom" bits between X1D and GFX 50. I think Hasselblad chose 16 bit files just to stay in line with there higher end cameras. It's not until you reach PhaseOne or Alexa levels of dynamic where 16 bit files even really matters. In terms of dynamic range the GFX 100 sits north of the X1D/pentax/gfx 50 type cameras but south of the PhaseOne so I think any benefit that you might get on the GFX 100 16 bit files over the 14 bit files certainly aren't huge. Probably the type of differences where you have to go hunting for them or someone would need to point out to you.
Now the long version.
I'm not 100% sure but I wouldn't worry to much about the 16 bits vs 14 bits. If you look at it from the stills perspective, even on medium format cameras, there have been discussions about how valuable the 16 bits are vs 14 bits. No CMOS sensor that's readily available that I'm aware of even if you look at a PhaseOne, GFX100 or on the video side of things like the Alexa 65, will give you 16 stops of clean noise free dynamic range. At a certain point the ADC converter's bit depth is exceeding what the CMOS can capture and you are basically just getting more steps or a more accurate readout of noise without that much benefit to signal. Some people argue the difference between 14 and 16 bits is huge and other's will say they can't see the difference just a bloated file. Either way I doubt it would make much difference on the A7SIII. I believe there was some drama not with the 150mp IQ4 or 102 mp GFX 100 but with the Hasselblad X1D and the GFX 50s/r. They are both using the same sesnsor and apparently both doing 14 bit readouts but for some reason Hasselblad chose to to still process/write in 16 bits and Fujifilm stuck with 14 bits. People argued the merits of the "phantom" bits between X1D and GFX 50. I think Hasselblad chose 16 bit files just to stay in line with there higher end cameras. It's not until you reach PhaseOne or Alexa levels of dynamic where 16 bit files even really matters. In terms of dynamic range the GFX 100 sits north of the X1D/pentax/gfx 50 type cameras but south of the PhaseOne so I think any benefit that you might get on the GFX 100 16 bit files over the 14 bit files certainly aren't huge. Probably the type of differences where you have to go hunting for them or someone would need to point out to you.
Last edited:
Trek of Joy
Member
tull_power
Active member
As far as I know, size of value for each colour (be it 8bit,12bit or 16bit) in independent of dynamic range. However, it follows that for any given dynamic range to be fully effective - that is - both in the shadows and in the highlights in the same frame - there must be enough precision to hold a meaningful value from black all the way to white. An example being SLOG3 recorded in 8bits on previous Sony cameras (not the A7SIII) - you did have the dynamic range but no way to record the sensor pixel's value accurately enough - resulting in banding and other artifacts.
On a seperate note, it looks like Canon are nervous about the R5 and R6 and might either recall them or make some last minute modification to combat the severe heat problems.
https://www.eoshd.com/news/will-can...ium=feed&utm_campaign=Feed:+EOSHD+(EOSHD.com)
joema
Active member
Thank you for pointing that out. I constantly hear supposedly-knowledgeable people equate bit depth with dynamic range. It's just not correct. There is a *loose* correlation because traditionally an expensive high-bit-depth cinema camera would generally be designed with a sensor having better dynamic range. But the bit depth doesn't *cause* the dynamic range, and (as you pointed out) doesn't even *enable* dynamic range. You could theoretically design an 8-bit sensor and processing path which had 16 stops of dynamic range, but there's no reason to do that.
Right now on the Arri Alexa web site, they have downloadable ProRes sample footage, some of which is only 10 bit. Those cameras will also shoot internally 12-bit ProRes 4444XQ or 12-bit log-encoded uncompressed ARRIRAW. That camera has great dynamic range and beautiful roll-off, but to my knowledge it's "only" 12 bits per color channel.
We think the Ninja V will also record 12-bit log-encoded ProRes RAW from the A7SIII, which might send the raw data as 16 bit, possibly with lower bits zero'd out due to issues with HDMI standardization, encoder ASIC, etc. Therefore I don't see why all the concern on some forums about whether it's 12, 14 or 16 bits. I wonder of some of that is coming from photographic people who are familiar with 14-bit raw stills and think somehow 12 bit raw video is a downgrade.
What would be really cool is if mirrorless cameras could sustainably record 14-bit raw stills at 24 fps. They are probably just one generation from doing that.
ahalpert
Major Contributor
The bit-depth of the ADC reading the sensor and the bit-depth in which you encode the files are 2 separate matters.
"In other words, 12-bits provides enough room to encode roughly 12 stops of dynamic range, while 14 bits gives the extra space to retain up to around 14EV"
https://m.dpreview.com/articles/465...e-not-the-number-of-colors-you-get-to-capture
Arri says "Images in the camera are processed in 16 bit linear and then encoded in 12 bit log on the recording medium. "
I think the same thing happens with sound. You can record really low signals (ie whispers on a recording with levels set for regular speaking). But if your bit-depth is too low, you won't be able to raise the level of those whispers in post without destroying the quality, and therefore your DR is limited
ahalpert
Major Contributor
LOL
I bought the zoom f6 just before the pandemic, it has 32-bit float and their website says "The Zoom F6 professional field recorder features both 32-bit float recording and dual A/D converters, providing an unprecedented amount of dynamic range."
Samuel H
Major Contributor
If the a7S III has the DR improvement that some people are talking about, it's probably using a 14-bit ADC even in video mode.
With standard (linear) ADCs, the DR is bounded by the bit depth of the ADCs, but most cameras don't reach that limit because of the read-out noise. Plus, simple theories usually talk about a weird single-pixel camera: averaging informationo from nearby pixels can provide the gradation that the last bit from the ADC is lacking.
How all of this turns out in practice depends on a lot of details. Sometimes ADC bit depth matters a lot for dynamic range, sometimes it doesn't. Dpreview often checks this:
On the GH5s, 14-bit mode gives you around half a stop more than 12-bit mode: link
On the a9, there's basically no DR difference between the 12-bit and the 14-bit modes: link
but those are cameras that don't have huge DR to begin with...
On the a7R II, 14-bit gives you between one and two stops more DR than 12-bit mode (I'd say it's very close to two stops): link
With standard (linear) ADCs, the DR is bounded by the bit depth of the ADCs, but most cameras don't reach that limit because of the read-out noise. Plus, simple theories usually talk about a weird single-pixel camera: averaging informationo from nearby pixels can provide the gradation that the last bit from the ADC is lacking.
How all of this turns out in practice depends on a lot of details. Sometimes ADC bit depth matters a lot for dynamic range, sometimes it doesn't. Dpreview often checks this:
On the GH5s, 14-bit mode gives you around half a stop more than 12-bit mode: link
On the a9, there's basically no DR difference between the 12-bit and the 14-bit modes: link
but those are cameras that don't have huge DR to begin with...
On the a7R II, 14-bit gives you between one and two stops more DR than 12-bit mode (I'd say it's very close to two stops): link
Peter C.
Veteran
From what I've seen this isn't a marketing gimmick, it really works and is pretty amazing. It might not have a lot of benefit for dedicated professional sound operators who are monitoring and riding their faders to maintain the proper levels, but for solo videographers who often have to leave recorders unattended it's a blessing.
Bit depth is not a loose correlation with sensor DR, it's a *direct* correlation to the A/D. Each bit is a doubling of code values, 1,2,4,8,16..so on. With linear light, each f-stop is also a doubling of light. For 12 bit linear, there is maximum 12 stops of sensor DR, 10 is 10, 14 is 14 and so on. That allocates just a single code value (gray shade) to the last stop, the one just above pure black, not even accounting for noise. To avoid noisy low key in practice, two extra bits are preferable. After the linear acquisition, sensor IRE gets remapped for output in accordance with the gamma transformation. This is where the bit depth can be less, 10 bit, 8 bit. The gamma uses a 1/2 float mathematical transform that points to a lookup table. The lookup table normalizes the distribution of code values more equitably so that each stop on the gamma curve contains a similar number of code values.
This was already covered by Ahalpert and Samuel H.
ReelFilmGeek
Well-known member
Does anyone know if there are any raw files from the camera available to play around with yet. The video side seems great but if the still images don't hold up then why make it a hybrid camera.
ahalpert
Major Contributor
Matti on youtube is raving about the a7siii: https://youtu.be/eTtwxF5GIKc
I take YT reviews with a grain of salt. But he said the camera was a lot of FUN to shoot with.
That was what I loved about my S1 - so straightforward and reliable, I could just focus on what I was shooting. If I could have that experience with great autofocus, even better low-light and DR, and uncropped slow-mo...wow
I'm thinking about getting a sigma 24-70 to go on gimbal with it
I take YT reviews with a grain of salt. But he said the camera was a lot of FUN to shoot with.
That was what I loved about my S1 - so straightforward and reliable, I could just focus on what I was shooting. If I could have that experience with great autofocus, even better low-light and DR, and uncropped slow-mo...wow
I'm thinking about getting a sigma 24-70 to go on gimbal with it
ReelFilmGeek
Well-known member
Was just looking at the sigma 24-70 myself if I decide to stay with sony. The lens has a useable manual focus which i like when compared to other to something like the tamron lens. And Yeah like you said seems like the camera will make for a great "just pick up and shoot" camera while being a workhorse b camera for one man band/small production companies. I can't wait to get my hands on one to test the slow motion options against the fs5 into an odyssey 7Q