High ISO Compared: Sony A7S v. A7R v. Canon EOS 5D III

now we got 2: a7 II and a7r II

Then it's not much of a comparison image is it? It happens to be the first part of the image that got magnified on my computer, I also note that other highlight areas (lighted windows) do not show "demonstrably" better DR in the highlight areas either. At least not on this sample. They are actually much of a muchness. Even the Seattle Tower light areas have a little more detail on the 5D. You guys shot the images. I didn't.

It is a comparison. It's just not a comparison of what you want it to be.

This test was not designed as a DR test. It was designed as a noise test.

The 5D III cannot magically have more highlight detail b/c of more DR. If it has any more highlight detail over the A7S, it'll likely be b/c of a small increase in sensor efficiency on the A7S that makes the A7S overexpose a bit more easily (since matched focal plane exposures were used).

Please don't misuse the contents of this article for a purpose it wasn't designed for, and then say it's not much of a comparison. It's not a DR comparison, it's a noise comparison between cameras receiving equivalent amounts of light at different ISOs.

Seriously, there's no extra highlight detail in the 5DIII shots. Look here. If you see any other windows that show more detail, it's likely because the lights were shut off.

So, again, this is not a DR test. It's not how you do a DR test in any way. So please don't use it to make any judgement on DR. Seriously.

@Rishi Sanyal You make good points. Pity they seem lost on the guy who wants to see what isn't there. :-)

No one 'bellyaches' about huge differences in dynamic range between these cameras at high ISOs, and there was no shadow push at low ISOs so, once again, I have no idea why you're talking about some attribute this particular test shows absolutely nothing about.

You trying to talk about DR from this test is akin to drawing conclusions about the maximum speed of a car from a steering test. They're completely orthogonal.

GaryJP, it hurts to see your favorite brand smoked by a Sony doesn't it? Fanboyism simply blinds you to the truth and all the facts in the world won't change your mind.

It would if I had a favourite brand. For the rest, grow up.

@How.... tell me what are the serious high iso camera, if A7R and 5D3 are not.... you're funny...

What ISO are you speaking of? At 25600 the A7r is similar for example (slightly more noise but better detail, noise reduction should set them on par until that ISO). Would I sell pictures apart from web sized shots above ISO 12800 or 25600? I think not.

Nikon D4s is considered the high ISO leader, why not include that in the test?

Dabbler:

Probably because the D4s is very expensive. The Canon 1Dx is also not included. And the 1Dx is a significantly better high ISO camera than the Canon 5DIII.

The real questions are why not the Nikon Df and the Canon 6D?

Better high ISO on A99 then on Canikon full frames? STL costs sensor something like 1/3 to 1/2 EV of light and that's exactly why STLs have slightly worse high ISO performance. It's compromise make focus panel constantly illuminated.

Or you were sarcastic. Anyway I am Sony fan but you outperformed me by a long mile.

... my LCD display cos I know the exact exposure/colour tone & temperature before I shoot. The LCD is permanently closed inwards & only opened when I need to shoot at the waist level or ground level.

Its ironic how these are also precisely the reasons that allow mirrorless camera users to take better pictures than many traditional pentaprism OVF relics. Try looking at the great work found on m4/3, NEX, Fuji, Alpha forums/interest groups (on a photographic technique level, not pixel peeping level).

@Pretern... Ridiculous statement to say mirror less users take better pictures imo. Handling of smaller mirror less has pros and cons. The mirror may introduce slap (easy to avoid with lock up), but the A7r and OMDs for example show that shutter vibration is an issue. I'm furthermore not so sure whether an EVF is that important. I use both mirror less and DSLR and with a little bit of experience you know which setting to take to not clip highlights, for example. For beginners it is nice, of course. It may be he future, but I don't have a problem with both.

@otoklikBG,

I'm no fan of anything except innovation and technical excellence. Problem with you fans is that you don't get it cos you are too busy being a fan.

Of course the technical quality of the A99 shots are higher when I have up to 4.5 stops in-body image stabilisation advantage over any Canikon at focal lengths up to 70mm (with any lens). It is a MASSIVE GAME CHANGER. Even if you are half blind you can see the difference of 3-4 stops can't you?

I can take 1/8th s shots handheld with my Zeiss f2.8 16-35mm, 24-70mm lenses at ISO 800. Other photogs would have to shoot at 1/30th s & ISO 3200 (while suffering the full effects of mirror vibration blurring + mirror slap noise) during a theatre performance.

Who do you think gets the job & the published picture?

Whow! You seem to be THE expert then! Good to know. I wonder how people made those great photos using 1Dx and D3/D4(S) then when shooting sports. Seems to be impossible when using DSLRs or non-SLTs.

@Preternatural : wow, and you call others 'ignorant with no understanding of photography' ...

Well, thank you very much for explaining my job to me. I must have missed so much not having your insight for the last 12 years I've been a professional wildlife photographer ... Gee, why am I using bulkier, heavier, more expensive conventional Canon DSLRs ?

Well, I'll tell you why : because the files ultimately outperform anything I've seen from Sony & co, hands down. And I'll tell you another secret : Sony is just not an option in pro nature & wildlife --> to my knowledge, there's not even a single colleague using Sony.

1/8th s shot on theatre performance, they are all on a seat?

Yawn.

No, it's not useful to my current photography, but then having the option would open up entirely new facets of photography to me. I can't take my Nex7 outside at a quarter moon and so much as hope to get photos of wildlife unless they're asleep. Sure I can use a flash and get 1 chance at a photo, not counting the subsequent photos of animal anuses as they head for the hills, or I can go ultra high ISO and completely silent shutter and take lots.

Dimly lit events? Concerts? No more need to push in PP to salvage a useable shutter speed. No need for an obtrusive flash.

Pretending that the option of shooting at extremely high ISO is useless to all only speaks to the ongoing trend of placing technical image quality over content and, if this site's comments are true, evidently over getting an image at all.

Give me a break. Anyone that had this capability on hand will absolutely find a way to use it.

Peter: Yeah, generally speaking, you got it.

Why did we spend time doing this? To visually show what the claim 'large pixels for High ISO' actually meant to the photographer, considering the significant cost in resolution that is paid.

Is that not valuable to those considering this camera, no less delivered before most pre-orders have even shipped? :)

HowaboutRAW: Do you have any evidence to back up the claims, such as these, you make:

"It's not so much the ISO 100,000 setting that's of importance in the year 2014, but how noise free the image is at ISO 25,600 that's important. Here the Nikon Df and D4s best the Sony easily 25600. And the Canon 6D is likely as good as the Sony A7s"

If so, please show us.

Rishi Sanyal:

I don't share my raws, but the testing I've done confirms this point, you can play with studio scenes here.

The point is that in shadows, at ISO 25600, the Sony has typical Sony cyan and magenta banding--that banding is the big problem. Grain is something else.

And the D4s has a bit less grain at ISO 25600, but most importantly zero magenta and cyan banding and/or blotching in deep shadows at that ISO. For the D4s the blotches in the shadows don't really show up until ISO 50,000. The Df is almost as good as the D4s.

The A7s is a good high ISO camera, like the Canon 6D, but it's not the king of the hill for readily useable high ISOs. (Above ISO 100,000 I really have no idea, since it's not realistic to shoot there with any of this gear.)

You obviously are a simple cam person & shoot in none too challenging situations. Plus, you do not know your photography much.

During the film days, PJs used to talk about the "need for speed" & use ISO 800 film, even push it to 1600 even though grain & colour suffers.

For professional shooting lighting levels we go above 6400 all the time. I've had shots that go to 10,000, 12800, 25,600 requiring extensive manual noise reduction work. Such as shots of the President of my country at night (no flash possible).

Need for speed involves understanding that shooting without flash often gives the best, natural results. Cue concert/theatre/dance performances. Even with flash, the most skilled photographers use fill flash at the lowest most imperceptible levels for a natural look. You want as much of the ambient light to fill the shot.

The A7S allows that. Even if native 12 megapixels ain't enough, the likes of Genuine Fractals /OnOne magic on super clean pixels with huge DR will rock.

Well.... since I actually work in the film/digital video arena. And more than that, own an 8500 sq foot sound stage and shoot regularly on RED. Yes, I could say I'm intimately familiar with the process. I'm talking about still photography where you have the luxury of time dealing with a single frame.

If you've got the luxury of time and a still subject why does it matter either way? A camera with no viewfinder at all (read: mirrorless with manual focus assist) would be at least as suitable (I'd say more) to those circumstances as an optical viewfinder.

But turn that around and take away the luxury of a still subject and ample time and you're SOL if you completely miss exposure early.

I use my screen for dialing in my focus. Then shift to optical/mirror viewing so I can see all of the light as it is coming through the lens. I don't like the limitation of the screen's ability. Once screens become wide gamut, it will likely be less of an issue.

And when you look through the lens, you are seeing wide open aperture, which doesn't accurately render what the final image will look like. On the mirrorless, the screen is showing you *exactly* what the final image will look like, with the aperture set as the final image will be. I have not seen any issues with my NEX-6 and color rendition. And with focus peaking and 9x zoom, my manual focus old minolta lenses are fun to use and accurately focused. My eyes are not getting any younger, and looking through the glass is definitely not all people think it would be when you have focus zoom and peaking available.

I am having fun picking up old lenses at garage sales or ebay for pennies on the dollar and getting the matching lens mount adapter. I have a huge selection of nice lenses now, and I've spent less than $500 on, in total.

Superchalupa brings up a great point about aperture that people like Dester forget to understand.

Thanks for sharing.

Look we get it Dester, you shoot canon and decided to mention something you don't like about Sony and we all hope viewfinders improve. There will come a day when the resolution and readout speed of a viewfinder is equal to that of the sensor that feeds it.

2014 is a landmark year for photography. The Olympus omd1, panasonic gh4, fuji xt1, and Sony a6000 are all incredibly fast and cheap cameras for what they offer.....and canon is no where to be found (yet - we all want them to) blazing a trail alongside these companies.

Deal with it and cheer on competition that helps all of us.

You are the perfect example of the ignorant pentaprism-OVF dinosaur.

The whole reason why I moved to the A99/A7 paradigm is the same reason the mirrorless brigade moved to theirs.

You dinosaurs spew the nonsense about wanting to see what your eyes/lens is seeing. So LOOK! Why do you need to put a heavy cam to your eyes to see what your eyes are seeing?

When I put camera VF to eye, I want to see what the SENSOR is seeing. I want to see an immediate exact prediction of what the sensor will GET.

WYSIWYG. That is the paradigm you dinosaurs fail to get. That is why you need to chimp and check your LCD after every shot.

Whereas I finally get to enjoy what it was like to shoot film (not needing or being able to check the shot, just concentrating on what is coming next) during the digital age.

OVF in the digital era is half-baked half-screwed...

But let me repeat: this 2 EV advantage is really only for darker tones at the highest ISOs (204k & 409k, judging from my visual tests). The advantage starts to decrease as you go to brighter tones.

That's no easy task, so the A7S should be applauded for what it's doing here. Whether or not this advantage actually makes a significant difference to your type of shooting is another matter entirely, of course!

I'll see if I can get my hands on normalized full SNR curves to correlate with the visual analyses presented in this article.

I assume you are a sports photographer who must shoot football games at night in dark stadiums at very fast shutter speeds, mosc. I am still trying to figure out who else needs ISO 100,000++

Yeah it is better as a high ISO (above 6400) camera than either the A7 or A7r. The 5D and D610 and A99 too.

saying it twice does not make it so. If you feel that is the case, when the canon/a7R is normalized to 12mp, which very much gives an advantage to the a7S, good bless you. Go buy one.

HowaboutRAW:

"Downsampling does NOT remove noise, it's a near total myth."

Not sure how you're defining 'remove noise', but downsampling increases SNR. So, relative to signal, it does decrease noise.

HowaboutRAW:

Er, thank you - because that was the entire point of this article.

You claimed downsampling does not reduce noise. That's completely incorrect, which can be shown time and again simply by hitting the 'Print' button (which down-samples the image to 8MP) in our studio scene.

This article was looking at any tangible gains w/ the A7S over the A7R - in normalized comparisons - due to the larger pixels of the A7S and/or any other optimizations to the sensor/signal processing.

The conclusion was that differences exist - at higher ISOs - where simple normalization does not reduce noise enough to bring the SNR of the higher resolution sensor up to that of the lower resolution sensor.

Clear now? Downsampling reduces noise, but in this case not enough - at the higher ISOs - to match the performance of the A7S.

Again, that was the entire point of this article, & presumably one of the reasons behind Sony making the A7S to begin with (video considerations aside).

"who doesn't know the larger pixel can handle higher isos in low light with less noise than smaller pixels?"

It would seem you missed the point about the normalized comparison?

Yes bigger pixels collect more light, but normalizing the comparison is a way of simulating equal resolution across the A7S & A7R. Then the comparison gets interesting, b/c the A7R's pixels have to theoretically have sqrt(3) less read noise than the A7S' pixels to match high ISO performance after normalization. That's b/c, generally speaking:

All else held equal, a n times higher res sensor has to have per-pixel read noise that is sqrt(n) times lower than the per-pixel read noise of the lower res sensor.

This is because extra aggregate read noise from reading more pixels. And this says nothing about other factors at play: readout speed, QE, etc.

So the point is actually far more subtle than your simple 'bigger pixels are better' argument. And so we wanted to empirically test the performance.

No, you don't upscale the lower resolution camera to the higher resolution to compare noise. The higher resolution camera starts at a disadvantage for pixel-level analysis b/c of its smaller pixels. For comparison to a lower resolution camera, you simulate the pixel size of the lower res camera by downscaling. This is an algorithm available to you that effectively simulates what would have happened in the hardware had the signal from n pixels been 'binned' to 1 pixel - by adding together the signal from the n pixels just as they would've be 'added' automatically in hardware had there been no subpixels within that region.

If you upscale the lower res camera image, what on earth are you simulating? Dividing up 1 pixel into n pixels? How's an algorithm going to do that - that is, create information where none is available?

Think of it this way: you start w/ a low noise, low res image, and a high noise, high res image.

If you upscale the low res image, its noise stays the same, it doesn't really gain any res, and so your comparison leaves you right back where you started: a low noise, low res image, and a high noise, high res image.

If you, OTOH, downscale the high res image, its noise decreases, and its res matches (our supersedes) the low-res image. So you now have your low noise, low res image, and a new image with potentially the same, or some degree worse, noise, with the same or better res image.

Which scenario above do you actually learn something from? Key in that statement is 'potentially the same, or some degree' worse. That's what you learn by opting to downscale, not upscale. If you upscale - you already know the answer to begin with when it comes to noise.

Also - upscaling typically makes the images incomparable. The noise is different and the scene detail resolved is different. Downscaling removes one of those factors (scene detail resolved) so you can focus on the noise differences.

In general - you don't change two variables at the same time. You control for one, while assessing the effects of the other.

As for your statement: " unacceptable noise in low light of 24mp aps/36mp ff with dubious gains in resolving power."

(1) Unacceptable noise? This entire comparison shows that the 36 MP A7R holds up quite well all the way up to its max ISO (25,600). It's not unacceptable at all. It's actually better than the 22MP Canon 5D Mark III, so I have no idea what point you're trying trying to make about Canon & its 'middle path'.

(2) 'Dubious gains in resolving power'? That's demonstrably an inaccurate statement, as a simple MTF analysis will show you. So will the images above (if you select 'Native Resolution'), by the way.

Rishi, thank you for so fully addressing a topic that I, too, don't fully understand. You've dealt very thoroughly and politely with some underlying misconceptions.

Thanks trustya - that's appreciated. If I can help even one person think through a concept and get it for themselves, that's a good ROI in my book. I also learn something in the process sometimes - the underlying source of the misconception. And questioning why you believe what you believe is great - you either reaffirm what you believe or you consider changing your point of view. One should always be open to the latter.

You're not comprehending the noise-per-sensor thing. The A7r is no worse than the A7s for stills.