"Native" ISO... is it real and does it make a difference

[Pinchers of Peril]

So my friend told me that it is better to use ISOs that are multiples of 160 since that is the camera's native ISO levels and the ISOs between those are just "pushed or pulled" digitally. Is this true? Is there any real world difference in the native vs non native ISOs. I've tried doing some research on it and have found conflicting information. I'm just trying to figure out if there are certain ISOs that I should use or avoid. -Thanks

 

[Loswr]

'Native' ISO is the converse of 'expanded' - the former is achieved by analog gain, the latter by digital gain. So, a camera may have a native range of 100-12800 ISO (analog) with L, H1 and H2 expanded settings, for example.

The concept you're referring to is 'base ISO', and whether that matters depends on the camera. First off, not all sensors have 160 as a base - it's 100 for some, 200 for others. In many cameras, the analog gain is applied in full stops from the base, and the 'tweeners' are digitally pushed or pulled 1/3-stop. If that's the case, then the pushing adds a little noise and the pulling removes a little noise. In practical terms, it doesn't make much real difference. In some cameras, the 1/3-stops are analog gain, too.

Regardless, a blanket statement to shoot at 1-stop multiples of ISO 160 is not universally applicable advice...

 

[Knut Skywalker]

This applies only when you are in videomode, if my information are right. Please correct me if I'm wrong, I think i heard it in one of those DSLR-video-tutorials on Vimeo.

 

[spinworkxroy]

i do believe it does make a slight difference these days BUT unless you pixel peep alot, you're not going to notice it. Especially with the newer cameras, the lower ISOs all the way to 1600 are almost identical..you're better off not worrying about "base" ISO.

 

[AprilForever]

Maybe you are talking about the 7D's native ISO of 160?

 

[insanitybeard]

Forgive my ignorance on this.... I understand the concept, but am interested to know what the base ISO of my 7D is- is it not infact ISO 100 as I assumed? If it is infact ISO 160 does that mean I may as well not bother using ISO 100 (from the point of view of capturing the most detail with the minimum of noise)?

 

[Marsu42]

Regardless, a blanket statement to shoot at 1-stop multiples of ISO 160 is not universally applicable advice...

Luckily, the Magic Lantern devs have figured out what iso is "best" - and it's rather surprising and more complicated than one might think because an analog and digital component is involved. Btw, setting the iso via ml gives better iq than via Canon

http://magiclantern.wikia.com/wiki/ISO#Then.2C_what_is_the_best_ISO.3F

Are ISO 160 multiples the best to use?

NO. They have harsh highlight rolloff and intentionally clipped details in highlights. I have no idea why.

Are ISO 100 multiples the best to use?

NO. While they do have smooth highlight rolloff, they are digitally pushed by a small amount (the exact value depends on picture style and other settings). What does this mean: a small amount of raw data, which actually has the best SNR possible, is simply thrown away.

Then, what is the best ISO?

To the best of my knowledge, the best ISOs are the ones available in recent Magic Lantern versions (April 2012 or later), obtained from ISO 100 multiples adjusted with a small amount of negative digital gain:

ISO 85, 175, 350, 700, 1400, 2800 - best for Neutral -4 and other low-contrast styles.
ISO 80, 160, 320, 640, 1250, 2500 - good for a wide range of situations.
ISO 70, 140, 280, 560, 1100, 2200 - best for high-contrast styles.

To enable them, set DIGIC ISO gain (in ISO submenu) at -0.2/-0.3/-0.5 EV and dial your ISO from ML menu or shortcut keys.

 

[dolina]

It matters to stills to! It's the difference between working to remove noise or remove lesser noise.

 

[MarkII]

Luckily, the Magic Lantern devs have figured out what iso is "best" - and it's rather surprising and more complicated than one might think...

The charts here are instructive: http://home.comcast.net/~nikond70/Charts/PDR.htm

If you look at the 5DIII, you can see the waving up and down of the DR with intermediate ISO steps - exactly what you would expect for a mix of analogue and digital gain setting. In contrast, the 1DX plot is smooth, perhaps because it is using analogue gain for the intermediate steps.

The plots suggest a peak DR at ISO 160 on the 5DIII, which is consistent with the ML folks conclusion that the native ISO is somewhere around 80ish and everything else is a push/pull of that.

The effect is small, however, and probably not worth the hassle of fretting about when shooting... however I wish Canon had a RAW capture mode and metering that operated only at native ISO values.

 

[Marsu42]

structive: http://home.comcast.net/~nikond70/Charts/PDR.htm

Thanks for the link, great comparisons there - I'm surprised the current 18mp aps-c sensor has about the same dr as 5d2/3 up to iso 400.

And I'm really looking forward to my first ff camera - where my 60d is getting non-usable (iso1600+) the 5d2/3/6d(?) shows nearly no noise with very little loss of dr. But of course that's not because high iso on Canon ff is so good, but because low iso is so bad (select Nikon d600/d800 if you are ready for the shock...).

 

[MarkII]

Going from a 40D to 5DII I did not see much difference in low-ISO IQ, largely because it was already pretty hard to fault the 40D (which I still use). Where FF is clearly better is at high ISO - as the graphs show.

However, I think that most of the 'pop' you get with FF is down to the glass, partly because the lower pixel density sensor is less demanding on resolution and partly because of a tendency to end up shooting with shallower depth-of-field (and vignetting).

Also the FF RAW files seem to have more latitude for post-processing than I would have expected from the numbers. Extreme edits seem to give nicer results, even if you do hit the noise floor at a similar place - and frankly if you are below ISO800 on the 5DII/III, small errors in the exposure are much more significant than these 'native iso' effects etc.

 

[Edwin Herdman]

Thanks Marsu42 and Mark, too. I guess I'd better have a look at Magic Lantern then!

 

[Pinchers of Peril]

Wow thanks for all the info everybody. I am interested in magic lantern but still kind of nervous to put it on my camera. I know lot of people who use it with no problems but I have heard the occasional account of cameras getting messed up from it.

 

[Marsu42]

Wow thanks for all the info everybody. I am interested in magic lantern but still kind of nervous to put it on my camera. I know lot of people who use it with no problems but I have heard the occasional account of cameras getting messed up from it.

Don't get affected by the FUD - *stable* versions of ML are what they say - stable. I know because I ran it as pre-release on my 60d and it has issues than just like now on the 5d3 *alpha* builds - but that's what you'd expect from pre-release software. If you're on 5d3, just wait for the stable version (and donate so you support the effort).

No user has managed to brick his/her camera by ml stable unified (or it would have been reported), the worst thing that could happen (but doesn't) is that it locks up - in that case pull the battery out and put it back again, done. Believe me, you're missing out big w/o running ml which is the only reason I'm sticking with Canon (well, and because I've been using it since 1990).

 

[LetTheRightLensIn]

For one that guy cares only about the interal 8bit jpg/video engine, important to keep that in mind.

Regardless, a blanket statement to shoot at 1-stop multiples of ISO 160 is not universally applicable advice...

Luckily, the Magic Lantern devs have figured out what iso is "best" - and it's rather surprising and more complicated than one might think because an analog and digital component is involved. Btw, setting the iso via ml gives better iq than via Canon

http://magiclantern.wikia.com/wiki/ISO#Then.2C_what_is_the_best_ISO.3F

Are ISO 160 multiples the best to use?

NO. They have harsh highlight rolloff and intentionally clipped details in highlights. I have no idea why.

Are ISO 100 multiples the best to use?

NO. While they do have smooth highlight rolloff, they are digitally pushed by a small amount (the exact value depends on picture style and other settings). What does this mean: a small amount of raw data, which actually has the best SNR possible, is simply thrown away.

Then, what is the best ISO?

To the best of my knowledge, the best ISOs are the ones available in recent Magic Lantern versions (April 2012 or later), obtained from ISO 100 multiples adjusted with a small amount of negative digital gain:

ISO 85, 175, 350, 700, 1400, 2800 - best for Neutral -4 and other low-contrast styles.
ISO 80, 160, 320, 640, 1250, 2500 - good for a wide range of situations.
ISO 70, 140, 280, 560, 1100, 2200 - best for high-contrast styles.

To enable them, set DIGIC ISO gain (in ISO submenu) at -0.2/-0.3/-0.5 EV and dial your ISO from ML menu or shortcut keys.

 

[bycostello]

i keep my iso in 1 stop increments mainly so i don't have to twiddle the dial too much...

 

[jrista]

So my friend told me that it is better to use ISOs that are multiples of 160 since that is the camera's native ISO levels and the ISOs between those are just "pushed or pulled" digitally. Is this true? Is there any real world difference in the native vs non native ISOs. I've tried doing some research on it and have found conflicting information. I'm just trying to figure out if there are certain ISOs that I should use or avoid. -Thanks

This is a mistaken notion based on the behavior of Canon ISO settings and their noise characteristics. Canon uses a 1/3rd stop push/pull approach to achieving non-full stop ISO settings, which can result in some third-stops being less noisy and others being more noisy than their full-stop neighbors. To explain:

Canon sensors use ISO 100 as base ISO, and all of the standard full-stop ISO settings are indeed native. Third-stop settings are actually achieved by futzing with the exposure a bit, rather than directly amplifying the signal to those levels. For example, ISO 125 is actually ISO 100 with a 1/3rd stop underexposure, which is then digitally corrected, or "pushed" up to the correct exposure. ISO 160 is similar, only that it is ISO 200 with a 1/3rd stop overexposure which is then digitally "pulled" down to the correct exposure. ISO 125, since it is a digital push of 1/3rd of a stop, tends to be noisier than either ISO 160 or ISO 200. Conversely, ISO 160, since it is a digital pull of 1/3rd of a stop, tends to be less noisy than ISO 100. The consequence of this approach is that you lose 1/3rd of a stop DR for those intermediate stops of ISO. A third of a stop change in DR is rarely ever an issue in the very vast majority of circumstances though, especially at the lower ISO settings where you have more DR to work with anyway (which is the only time it exhibits...higher ISO's above 800 use an alternative approach.)

People who have noticed this quirk in Canon's noise behavior have made the rather naive assumption that it means Canon sensors actually have a base ISO of 160. On the contrary, when the actual ISO mechanics are investigated, Canon sensors most definitely have a true or native ISO 100, as well as a native ISO 200, 400, 800, and 1600 at the very least, and potentially more depending on the model. All third-stop ISO settings are achieved via push/pull, and at lower ISO settings that results in oscillating noise characteristics. This quirky approach to third-stop ISO settings is actually something I hope Canon moves away from when they move to a new 180nm fabrication process. A true sensor-level analog amplification to all ISO settings would be a better approach.

 

[fonts]

Luckily, the Magic Lantern devs have figured out what iso is "best" - and it's rather surprising and more complicated than one might think...

The charts here are instructive: http://home.comcast.net/~nikond70/Charts/PDR.htm

If you look at the 5DIII, you can see the waving up and down of the DR with intermediate ISO steps - exactly what you would expect for a mix of analogue and digital gain setting. In contrast, the 1DX plot is smooth, perhaps because it is using analogue gain for the intermediate steps.

The plots suggest a peak DR at ISO 160 on the 5DIII, which is consistent with the ML folks conclusion that the native ISO is somewhere around 80ish and everything else is a push/pull of that.

The effect is small, however, and probably not worth the hassle of fretting about when shooting... however I wish Canon had a RAW capture mode and metering that operated only at native ISO values.

What's insane (if its accurate) is the 5D Mark III/1Dx vs D600 .... God damn it :/