ISO Variance

March 27th

Personally, I don’t like the terms “ISO invariance” and “ISO variance,” but I can’t suggest better ones at the moment. 

Simplest explanation: An ISO invariant sensor is going to have a dynamic range chart that has a constant slope downwards as you increase ISO values. An ISO variant camera is going to have ups and downs, sometimes only a couple of clear jumps at particular ISO values where the digital gain is changed, sometimes changes at all values, as with the D5 low ISO values.

The D5 is acting somewhat different than we expected from the D4/D4s. Read noise—and thus dynamic range—is shifting in the low ISO values. That’s not typical of the Sony-Exmor sensor designs. It was typical of many Canon sensor designs. It’s not been typical of Nikon sensor designs, though there have been some previous ones—particularly the D3/D3s—that have similar levels of these shifts.

What happens is basically this: at a full stop ISO value—for example ISO 100—the read noise is some lowish level. But if you bump up 1/3 of a stop (ISO 125) the read noise goes up (and dynamic range down more than expected). Bump up another 1/3 of a stop and it’s up again. But bump up that final 1/3 of a stop (to ISO 200) and read noise drops below where it was at ISO 125 and dynamic range is better, too! This sawtooth pattern repeats up to about ISO 6400, where is changes slightly, then disappears completely. 

In practice, this means that if you’re pushing for every last little ounce of optimal data you can get out of the D5, you want to stay at ISO 100, 200, 400, 800, 1600, 3200 if you can. And you’d want to absolutely avoid ISO 160, 320, 640, 1250, and 2500, as you’d be better off bumping the ISO up a 1/3 of a stop in each case. Technically, you should avoid ISO 5000 and use ISO 6400 instead, but the change there isn’t near as dramatic as it is at the lower values.

I’m already seeing people go too far in their assumptions about this, and yet, there are significant considerations I’m trying to figure out that might be important.

The worst case happens at the lowest ISO values. But if you’re down at low ISO values, I’m not sure why you’d be picking ISO 125 or 160. You’re likely in good light, and you’re likely wanting maximal dynamic range, so stick with ISO 100, or 200 if you need a bump. 

But the sawtooth pattern does have some implications for auto ISO use. 

For the D3/D3s I wrote (in my book as well as elsewhere) that there really wasn’t a significant penalty for leaving the camera set to Automatic ISO with a maximum value of 800. That’s not perfectly accurate, as there’s a bit less than a stop differential in dynamic range that can occur between best case and worst case as the camera moves from ISO 200 to ISO 800. Still, the expected loss in an ISO invariant design would be about a stop per bump, so would be twice the loss you’d have if you used my D3/D3s advice for one of those bodies. With an ISO invariant camera, letting the camera crank up ISO means that you get a large shift in dynamic range as it does so. On an ISO variant camera, not so much, thus my Auto ISO advice.

I’m not sure what I’ll recommend on the D5. If I were to use exactly the same standard as for my D3 recommendation (same level of dynamic range change from maximum in Automatic ISO mode), I’d probably have to say ISO 640 should be the max for Auto ISO. Yet technically I’m measuring the D5 doing about as well as my D3s was at ISO 800 out to about ISO 1600, so why wouldn’t I say ISO 1600? 

This, of course, is one of the things I’ll have to put to test and evaluation over a fair number of circumstances before coming to any conclusion.

Nikon hasn’t made it easy to clearly evaluate how we should be shooting this camera. The D4 was far easier, as it was close to ISO invariant. The D5 doesn’t begin to distance itself from the D4 until about ISO 2500, at which point it’s better at every equal ISO by about the same amount (close to two thirds of a stop). 

But none of this takes into account something else: Nikon changed the noise reduction for JPEGs. It’s better. Clearly better at the same ISO and same settings. So JPEG shooters and NEF shooters may come to different conclusions. 

Progress isn’t always the simple, clean, logical one we hope for. The D5 is indeed progress. In particular, it’s pretty clear progress over the D3s, which is something that I would always hope to see when comparing every other generation of camera. The D4, too, had progress over the D3s, but mostly in the low ISO values. We’ve lost that progress and gained a different one: high ISO values. 

Indeed, the low ISO changes (D3, D4, D5) are enough that I have to conclude that Nikon knew that the wouldn’t be a D4x and wasn’t sure how the D800 would be received, but now there might be a D5x (or at least another clear iteration on the D8xx). 

Funny thing is, I don’t understand the value of Nikon keeping secrets in the low levels like this. They’d be making a lot more Nikon-user friends if they’d just come out and tell us about their design decisions and what motivated them. 

text and images © 2020 Thom Hogan
portions Copyright 1999-2019 Thom Hogan-- All Rights Reserved
Follow us on Twitter@bythom, hashtags #bythom, #dslrbodies