Sony A7R IV / A7S III / A9 II to feature 8K video, as new 60MP and 36MP full frame sensor specs leak

[HurtinMinorKey]

This isn't really "raw".  If it's not lossless, it ain't raw. 

[ND64]

On 12/20/2018 at 3:41 PM, androidlad said:

605000 is too good to be true for this pixel size. Probably its a base-only HDR trick. 

[wolf33d]

4 hours ago, Eric Calabros said:

605000 is too good to be true for this pixel size. Probably its a base-only HDR trick. 

Yeah this would mean 17.5 stops of DR  at a base iso of 12-15. Too good to be true unfortunately. 

I am dreaming of it though, a 36mpx A7S III with all of this new codec and DR and video modes. Problem is it would kill all Sony cameras including the A7R III. 36mpx is close to 42 and if it has better DR and better everything well... And the A7R III is a year old so hard to believe it will be renewed at the same time. 

Maybe an A7S III mid year and the A7RIV in October 2y after the RIII then. We'll see, hard to tell now 

[androidlad]

5 hours ago, Eric Calabros said:

605000 is too good to be true for this pixel size. Probably its a base-only HDR trick. 

 

20 minutes ago, wolf33d said:

Yeah this would mean 17.5 stops of DR  at a base iso of 12-15. Too good to be true unfortunately. 

I am dreaming of it though, a 36mpx A7S III with all of this new codec and DR and video modes. Problem is it would kill all Sony cameras including the A7R III. 36mpx is close to 42 and if it has better DR and better everything well... And the A7R III is a year old so hard to believe it will be renewed at the same time. 

Maybe an A7S III mid year and the A7RIV in October 2y after the RIII then. We'll see, hard to tell now 

The base FWC is true, it's achieved through multi-gain readout using 18-parallel ADC, similar to how ALEXA expands DR by using dual-gain architecture.

[Trek of Joy]

5 hours ago, wolf33d said:

Yeah this would mean 17.5 stops of DR  at a base iso of 12-15. Too good to be true unfortunately. 

I am dreaming of it though, a 36mpx A7S III with all of this new codec and DR and video modes. Problem is it would kill all Sony cameras including the A7R III. 36mpx is close to 42 and if it has better DR and better everything well... And the A7R III is a year old so hard to believe it will be renewed at the same time. 

Maybe an A7S III mid year and the A7RIV in October 2y after the RIII then. We'll see, hard to tell now 

You know being Sony it will be gimped I’m many ways because they certwill not stop heavily segmenting their models. 

[Rinad Amir]

On 12/17/2018 at 11:53 PM, ALFAOCS said:

Any ETA on the next Sony A7S III or the next Sony FS7?

Yah am thinking to go fullframe still cant decide wait for A7Siii or go back to Canon ? 

[webrunner5]

2 minutes ago, Rinad Amir said:

Yah am thinking to go fullframe still cant decide wait for A7Siii or go back to Canon ? 

Back to Canon, yeah you will be going back alright, back to about 2012 Tech.

[Vision]

1 minute ago, webrunner5 said:

Back to Canon, yeah you will be going back alright, back to about 2012 Tech.

1dx mark iii is on horizon if am correct , that would be some competition  

[ND64]

19 hours ago, androidlad said:

 

The base FWC is true, it's achieved through multi-gain readout using 18-parallel ADC, similar to how ALEXA expands DR by using dual-gain architecture.

Arri solution decreases the read noise.. nothing to do with FWC. The only thing than can create such a huge FWC is 2x2 pixel binning, which everybody can do, but usually don't. 

[webrunner5]

Yeah but you can probably buy a Canon C200 for what that will cost. I will take the C200 please. But yeah you are right it is due and it will probably be better hopefully. The mk II is damn good now so if it is better, man would be nice if you can afford it..

[KnightsFan]

1 hour ago, Eric Calabros said:

Arri solution decreases the read noise.. nothing to do with FWC. The only thing than can create such a huge FWC is 2x2 pixel binning, which everybody can do, but usually don't. 

Someone on Sony Rumors speculated that the pixel actually fills several times to make one exposure. Sort of like combining several short exposures into one longer one, but before the ADC and perhaps on a pixel level. I'm not sure if that's how it's actually achieved.

[webrunner5]

1 hour ago, KnightsFan said:

Someone on Sony Rumors speculated that the pixel actually fills several times to make one exposure. Sort of like combining several short exposures into one longer one, but before the ADC and perhaps on a pixel level. I'm not sure if that's how it's actually achieved.

It is probably along the same lines as the Smart HDR in the new Apple iPhones. The cpu is actually doing HDR in real time on video up to 30p in 4K. This stuff has gotten scary fast, Trillions of operations a second, and AI can do all kinds of crazy stuff now. So Sony is probably getting into Computational Mathematics in a big way now.

This new Sony a7s mk III might end up being a whole new kind of camera. Nothing we have ever seen before. A camera that actually thinks, I mean Really thinks. That is what the iPhone XS, XS Max does in HDR video, it actually thinks in a sense. It decides when more DR is needed or not based on content. It is not a constant thing like in the past. It is making sort of conscious decisions. Scary shit when you think about it. And it is in your pocket, Yikes! ?

[KnightsFan]

@webrunner5 using the CPU to do HDR doesn't explain how they achieve a higher full well capacity with such a small pixel.

[webrunner5]

28 minutes ago, KnightsFan said:

@webrunner5 using the CPU to do HDR doesn't explain how they achieve a higher full well capacity with such a small pixel.

Sure it does, it is AI using Computational Mathematics. It is new Voodoo. It is substituting the same thing 3 times to achieve what a larger pixel would have data wise. ?

[KnightsFan]

7 minutes ago, webrunner5 said:

Sure it does, it is AI using Computational Mathematics. It is new Voodoo. It is substituting the same thing 3 times to achieve what a larger pixel would have data wise. ?

I don't think you can use a CPU to increase the saturation point of a photosite. We're talking about the full well capacity of the photosites. Not the dynamic range of the image, or how the image is processed, or whether there is a computer-controlled variable ND filter in front of the photosite. Because none of those effect the FWC.

[webrunner5]

54 minutes ago, KnightsFan said:

I don't think you can use a CPU to increase the saturation point of a photosite. We're talking about the full well capacity of the photosites. Not the dynamic range of the image, or how the image is processed, or whether there is a computer-controlled variable ND filter in front of the photosite. Because none of those effect the FWC.

It is just a matter of cramming more transistors,  and light emitting diodes in the same area and having enough sensitivity, enough horsepower, ergo, Computational Mathematics, to make it happen fast enough. This is rocket science on steroids now. It is coming about because of better nano science and AI to control all of it, with real logic, learned logic, not just being programmed to do the same stuff over and over. We are just now getting into enough computation power to make just about anything possible now. We are into trillions, not billions of items on a CPU, especially GPU's now, and trillions of intelligent actions happenings every second.

I guess on paper they are using the same photosite more than one time in the flash of a eye as they say. Hell I know know, it is beyond my learning.

[ND64]

13 hours ago, KnightsFan said:

Someone on Sony Rumors speculated that the pixel actually fills several times to make one exposure. Sort of like combining several short exposures into one longer one, but before the ADC and perhaps on a pixel level. I'm not sure if that's how it's actually achieved.

Nikon has some patents for that idea, but its difficult to make, and even if they managed to do it, its very unlikely they put it in a off the shelf product. I think its more simpler, just a pixel binning. 605000/4=151000 which is typical FWC in modern CMOS sensors. Note that the next step is 81500, almost half, as it should be. 

[joshhr]

Correct me if I’m wrong, but don’t these specs seem a little excessive for a mirror less body? It makes a lot more sense to see Sony trying to absolutely dominate the cinema cameras under $20k (C200,C300,FS Series, EVA 1, UMP, etc).

 

They’ve absolutely killed the mirrorless competition so maybe they’re looking to dominate the sub $15,000 cinema cameras. Considering cropped 4K 60 is still in the early stages right now doesn’t uncropped 8K 60 seem a little excessive for a mirrorless body, if it’s even possible for anything less than $20k.

 

Again I could be completely wrong and a lot of you guys seem to know more than what I do, just a thought.

[Cliff Totten]

On 12/22/2018 at 1:34 PM, KnightsFan said:

Someone on Sony Rumors speculated that the pixel actually fills several times to make one exposure. Sort of like combining several short exposures into one longer one, but before the ADC and perhaps on a pixel level. I'm not sure if that's how it's actually achieved.

That was me that posted that on Sony Alpha Rumors. I read that MIT is working on a sensor readout technique that virtually doesnt allow a photosite to saturate. It makes a photosite fill up, dump its value and fill up again and again until the readout cycle is complete.

If you were using a 1/30 second shutter setting, you could collect and "add up" the photons in your your brightest highlight areas 2 or 3 or more times in that 1/60 shutter opening.

Your dark shadow areas wont get much help because those photosites would never completely fill up but your highlights theoretically have no limit. (Execpt for the readout and processing speed of A/D conversion.) How fast can the processor handle a photosite fill, read, dump, add, fill, read, dump, add, fill, read, dump add in 1/30 or 1/60th of a second?

Thats a TON of adding work for each photosite. You would probably need a specialized processing chip to take advantage of this new way of handling this kind of A/D.....

Oh wait!!,...isnt that what Sony is saying? You must also buy their specially paired image processor to go with this sensor? ( otherwize it cant work with any other? )

Hmmmm.......

P.s. Im sure this is a gross over simplification of what MIT and maybe Sony are doing. We will need to see how this develops to truely know. I think is clear that Sony is doing something unusual to make these phototon collection numbers so high using photosites that ore only 4+microns large.

CT

[androidlad]

There's no temporal element to any of the HDR features in new Sony sensors, it just won't work for high speed imaging.

The 605K e- FWC at base is simply achieved by hexa-gain readout. Period.

 

IMX490, an 1/1.55" automotive sensor, achieves up to 760Ke- with only 3um photosite.