Speculate: How much noise improvement from 5D3 to 5D4?

[KeithBreazeal]

The sCMOS design seems to be a solution. If it can be economically produced in a high mega pixel version, the sCMOS could be a great improvement for very low noise levels. I could see using it in the top of the line first.

http://www.scmos.com/

 

[lintoni]

Sporgon has speculated, elsewhere on CR, that Canon may have some hitherto unused features in their dual pixel sensors that they're saving for the higher end cameras. I find the idea appealing, but there's nothing (that I know of) to back up his speculation.

 

[Marsu42]

I find the idea appealing, but there's nothing (that I know of) to back up his speculation.

This is certainly the right place to discuss this possibility

I know for a fact that Canon doesn't use all features built into their DIGIC processors because Magic Lantern managed to unlock some of them. And I know from ML coding that Canon hides fw features like "af patterns": The 9-pt "diamond" systems have the ability to select multiple af points (not just one or all), but Canon simply chose not to make this feature accessible through their GUI - most likely to protect their premium cameras.

While the above it the Truth(tm), it's difficult to separate it from conspiracy theories. Btw do you know you only use 10% of your brains' power and can unlock the rest by paying a lot of €€€ for pseudo-religious mumbo-jumbo?

 

[lintoni]

I find the idea appealing, but there's nothing (that I know of) to back up his speculation.

This is certainly the right place to discuss this possibility

I know for a fact that Canon doesn't use all features built into their DIGIC processors because Magic Lantern managed to unlock some of them. And I know from ML coding that Canon hides fw features like "af patterns": The 9-pt "diamond" systems have the ability to select multiple af points (not just one or all), but Canon simply chose not to make this feature accessible through their GUI - most likely to protect their premium cameras.

While the above it the Truth(tm), it's difficult to separate it from conspiracy theories. Btw do you know you only use 10% of your brains' power and can unlock the rest by paying a lot of €€€ for pseudo-religious mumbo-jumbo?

You don't need to pay for it anymore, it's all on facebook, punctuating the cute cat and dog videos.

 

[VelocideX]

I note that the 7D2 has about 1/3 stop noise improvement (via increased sensitivity) from the 7D, which is disappointingly low.

And that's disappointingly inaccurate.

I own both cameras and - depending on the ISO - I see the very best part of a stop in noise improvement.

Seriously - treat anything you read from/about DxO as a work of fiction compared to Real World experience...

I was basing my information on http://www.clarkvision.com/reviews/evaluation-canon-7dii/

 

[VelocideX]

I appreciate that my original statement conflates signal and noise, when they are two separate issues. Perhaps then I should have been more specific and mentioned signal/noise improvement, which is higher than 1/3 of a stop.

My takeaway from this is that the signal/noise ratio for the 5D4 may be 1/3 to 1/2 stop higher than the 5D3 (at high ISOs), and that's basically the information I was seeking.

I agree that the banding issue at low ISOs is problematic for the 5D3. I expect this will be much better in the 5D4.

 

[davidcl0nel]

Btw: You do know that iso >6400 on 5d3 is simply digitally amplified, and you only loose dynamic range when shooting raw and going higher? The high iso values are for jpeg, video, marketing - but not for hq stills shooting.

I think its ISO 25600 native, and of course 51200 and 100k (called H1 and H2) are digital versions, which you can disable in custom menu... ?

I don't know exactly - but is there a "big" sensor (APSC/FF) with back-illumination technique? The small sensors for phones and small cameras have it since some generations. Shouldn't there an effect on big sensors too? Or isn't this as much, because of the lens for every pixel?

 

[Marsu42]

Btw: You do know that iso >6400 on 5d3 is simply digitally amplified, and you only loose dynamic range when shooting raw and going higher? The high iso values are for jpeg, video, marketing - but not for hq stills shooting.

I think its ISO 25600 native, and of course 51200 and 100k (called H1 and H2) are digital versions, which you can disable in custom menu... ?

Definitely wrong, the h1/h2 are as "digital" as 25k - Canon just calls them "extpanded iso" from tradition and to mark the fact that you're to expect an even more severe degradation of image quality.

To be precise, "digital" amplification can happen at several stages of the in-camera image data path - I'm just using it here as word to differentiate from "real" sensor analog settings which are full iso stops 100-1600/3200/6400 (depending on the camera model, ff, crop, ...).

Their marketing div certainly won't burden innocent Joe Sixpack users with the internal workings of their sensors, esp. as so many people think the camera's highest iso setting equals the real low light sensitivity of the sensor.

 

[jrista]

I believe ISO 25.6k is not digital, it is still done by an analog amp. It's just an analog amp that is downstream from the pixels. Supposedly that results in less noise. Not sure how, but I gather there are limitations with how much you can amplify the signal with per-pixel amplifiers.


The only truly digital ISO settings (where the adjustment is made AFTER ADC) are the expanded settings, as far as I know.

 

[jdramirez]

The other day the mentioned that the 5d mkiv might be the BIG sensor people have been asking for and it should have the same pixel density as the 7d mkii.....

SO I WOULD GUESS THAT THE USABLE ISO WILL BE LOWER... NOT DRAMATICALLY LOWER, BUT LOWER. And that makes me sad.

 

[LetTheRightLensIn]

I note that the 7D2 has about 1/3 stop noise improvement (via increased sensitivity) from the 7D, which is disappointingly low.

And that's disappointingly inaccurate.

I own both cameras and - depending on the ISO - I see the very best part of a stop in noise improvement.

Seriously - treat anything you read from/about DxO as a work of fiction compared to Real World experience...

says one of the Canon forums biggest writer's of fiction....

(I will give you that sometimes the character of noise can make it work out better in some cases, like the 7D has less high ISO banding than the 5D2 so in scenes with much of the scene near black but a few parts clipping white, the 7D could actually outdo the 5D2 even when not distance limited. In other shots the larger frame of the 5D2 made it do much better though.)

 

[Marsu42]

The only truly digital ISO settings (where the adjustment is made AFTER ADC) are the expanded settings, as far as I know.

It's really hard to tell, the ML devs really dug into it while trying to optimize the digic registers to get more dynamic
range "for free". There are multiple amps involved, and they work in different ways on each iso setting, currently my impression is that they won't be able to figure it out completely.

What it comes down to "digital" vs "analog" is if it's "worth it" dumping dynamic range for probably a tiny bit of less noise because the amplification is done in the image data chain. For most purposes, at least for what I do, the answer is "no" - I'd rather shoot raw 6400 underexposed than go higher on my 6d.

Imho there are lots more people out there cranking iso up w/o knowing what the tradeoff is, that's why I'm (and the ML devs, if you read the thread over there) are rather verbal about the minimal benefits of post-sensor amplification on high iso settings.

 

[Keith_Reeder]

says one of the Canon forums biggest writer's of fiction...

Just you're incapable of getting these results out of your cameras, don't run away with the assumption that we're all as inept...

And as for "writers of fiction" - as well you bloody know, back when I used to put up with your BS on the likes of DPR, I backed my arguments up time and time and time again with images that proved the point I was making.

Like I say - don't transfer your lack of talent onto others. I don't have to lie - I leave that to others.

 

[Tugela]

Don't expect high ISO noise improvement, Canon is going for low ISO improvements, better color, things like that. Any High ISO noise improvement will be slight. Of course, the new Digic 6 processor will produce jpegs at higher ISO settings and less detail, but the actual raw sensor output will be a incremental improvement.

The Dual Pixel sensor does have promise if Canon can do more things with it than they have shown on the crop cameras. I suspect that there are some more tricks they can use and are reserving them for the more expensive models. That is the standard sales tactic, whether its a iphone, a new car, a TV set, or a Canon DSLR.

The Digic 6 is not new. It has been around on point and shoots since 2013.

 

[FTBPhotography]

Silly topic IMHO. Why people are so obsessed with noise is beyond me. I notice its mainly from amateurs who have never shot film. Everyone wants to shoot at ISO 25600 and want it to look like
ISO 100. I dont get it. Get over it people.

 

[Khufu]

Silly topic IMHO. Why people are so obsessed with noise is beyond me. I notice its mainly from amateurs who have never shot film. Everyone wants to shoot at ISO 25600 and want it to look like
ISO 100. I dont get it. Get over it people.

Oooor you get over it, person!

It's painfully obvious why people would want improved performance and capabilities. Do yourself a favour and stop painting yourself as such a bitter old-timer

 

[jrista]

Silly topic IMHO. Why people are so obsessed with noise is beyond me. I notice its mainly from amateurs who have never shot film. Everyone wants to shoot at ISO 25600 and want it to look like
ISO 100. I dont get it. Get over it people.

Actually, you have that backwards. We want to shoot at ISO 100 and not have it look like ISO 800. ;P That is the primary issue with Canon sensors...the noise ramps up as you get to LOWER ISO settings. ISO 100 has aroun 1.5-2x more read noise than ISO 200, which in turn has 1.5-2x more read noise than ISO 400, etc. That flattens the DR curve out, WASTING information that the sensor actually gathered and could otherwise be usable.


It's never been about shooting at ISO 25600 and having it look like ISO 100. It's about getting out of ISO 100 what is there to start with. Canon's high frequency, low parallelism, off-die ADC units are throwing away massive amounts of usable information, where as sensors from other brands are preserving 2-8x more information.

 

[telemaq76]

i just want clean iso-100 !

 

[TommyLee]

when will they realize the market wants...
12 mega pixels and 2 stops better noise than 5d3 ........... across the range

and 45 megapixels and all the noise reduction possible... try hard
.....

I guess the market ...means ....me

I would buy another lens or two to help this along...AFTER the lower noise camera is delivered..
...ok 3

....

or I could go back to sleep....


my vote