ISO 50

[acoll123]

Just curious - in general, lower ISO equals better Image Quality.

I shot a water fall last weekend with an ND filter trying to get lower shutter speeds to allow the water to blur. I also used the expanded ISO on my 5DII and set the ISO to L which I think equals 50. Technically, should the IQ be better or worse with ISO below 100 in a DSLR like my 5DII?

Thanks in advance.

 

[Loswr]

Technically, a little worse. ISO 100 is the lowest 'native' ISO, and therefore ISO 50 is 'pulled' (intentionally underexposed) in camera, and actually loses a bit of dynamic range compared to ISO 100 (you lose headroom in the highlights).

Native ISOs start at 100 and go up by full stops. Other ISO settings are pushed or pulled from the nearest native ISO, which can have undesired effects on DR and noise. The 'tweener' ISOs that are 1/3 above native, and are thus pushed, usually have more noise (often equivalent to the next-higher native ISO, e.g. noise at ISO 250 looks like ISO 400). The ISOs that are 1/3 below native are pulled, and like ISO 50, result in a loss of highlight headroom.

 

[henryblyth]

The 'tweener' ISOs that are 1/3 above native, and are thus pushed, usually have more noise (often equivalent to the next-higher native ISO, e.g. noise at ISO 250 looks like ISO 400). The ISOs that are 1/3 below native are pulled, and like ISO 50, result in a loss of highlight headroom.

This sounds like a pretty big problem when it comes to Auto ISO. Are there any advantages to these inbetween ISOs?

If a pushed ISO can roughly be the same as the next-higher native value, I would prefer to have the latter.
If a pulled ISO results in less dynamic range than its native value, I would also prefer to have the latter.

This is a shame, because Auto ISO is heavenly for working in Manual.

(Actually, pushed ISO doesn't sound that bad. Pulled ISO, however, I'm steering well clear of.)

H

 

[dr croubie]

You can set custom functions to use 1/3 or 1-stop values for ISO, does that also count for when in auto-mode? (i use manual Av/Tv with auto ISO a lot for animals going in and out of shadows quickly)

I heard (another poster on this forum) say the "native iso" of a 7d is 160. does anyone have any (ie, concrete) proof of this? (can't google, at work).

i read on a review of the IQ180 Medium Format back that its native ISO was something like 35 (ok, so that's an MF back).
Likewise, I would have thought that every camera sensor had its own "native ISO", different to every other sensor, dependant on manufacturing process, sensitivity of silicon and circuits used, etc.
"ISO" is just the native unmultiplied number that comes out of the pixel, timesed by 'something' so that every time we set any camera to the same settings, we get the same exposed picture. in that case, the "native iso" is just what comes out when the 'something' = 1.
Going down to 100 or down to 50 or whatever, doesn't mean that either is the 'native', it's just where the engineers/marketing department think they can get away with it...

 

[Sanjayfg]

I made a test on 5D2 for 50 ISO quality. I found it MUCH better than ISO 100.
However I do not know if it effects highlights.
When shooting landscapes if I have a tripod, I will always shoot on 50 ISO. I will keep in mind about the highlights and will either expose keeping that in mind or bracket it to merge the blown out areas later in post.
Regards,
Sanjay

 

[Loswr]

You can set custom functions to use 1/3 or 1-stop values for ISO, does that also count for when in auto-mode?

The C.Fn setting applies only to manually selected ISO. If you set the ISO to Auto, it will use 1/3 stop settings.

I made a test on 5D2 for 50 ISO quality. I found it MUCH better than ISO 100.

I'd be interested to see the images from that test...

 

[motorhead]

I heard (another poster on this forum) say the "native iso" of a 7d is 160. does anyone have any (ie, concrete) proof of this? (can't google, at work).

I remember a test on the 30D that showed that 160/320/640 etc were cleaner than the interspersed numbers. I though Canon had gone back to 100/200 etc, but I'm going on what I've read, not what I'm able to prove myself.

 

[epsiloneri]

I made a test on 5D2 for 50 ISO quality. I found it MUCH better than ISO 100.

Sure, if you overexpose by a factor of two and then divide the result by two, you will get lower noise on everything in the frame that is not saturated. But you will sacrifice half your dynamic range (the highlights).

 

[Loswr]

I heard (another poster on this forum) say the "native iso" of a 7d is 160. does anyone have any (ie, concrete) proof of this? (can't google, at work).

remember a test on the 30D that showed that 160/320/640 etc were cleaner than the interspersed numbers. I though Canon had gone back to 100/200 etc, but I'm going on what I've read, not what I'm able to prove myself.

Only Canon knows what is 'native'. But I believe that the 'full stop' ISOs of 100, 200, 400, etc., are native. The forum posts claiming that ISO 160 is native, and that 160/320/640 are cleaner, are not incompatible with that. ISO 160/320/640 etc. are 'pulled' (although only by 1/3 stop, where ISO 50 is pulled a full stop). When you intentionally overexpose then reduce the exposure in post (even if that 'post' is in-camera before the RAW file is written), you reduce noise, so the '-1/3' tweeners have less noise than full stop ISOs, but also less highlight headroom.

You can see the effects of ISO on read noise in the data in the lower plot on this linked page for the 40D (which is probably the same source for the 30D test that motorhead refers to, and is linked from that page). ISO 160/320/640/etc., have a higher S:N ratio than the full stops (but there are no corresponding data for dynamic range, which if there were, would show that it's reduced). If you roll your mouse down the list of ISO values in the top graph, you can see the 'combing' in the histogram at the 'fake' 1/3-stop ISOs, whereas the full stop curves are smooth.

Interestingly, the page linked above suggests that the 'tweener' ISOs are not 'fake' (i.e. pushed or pulled) on the 5D and 1DIII - those cameras apparently use an additional analog amplifier. As a result, all the ISO curves are smooth, and in that case, the S:N is highest at the full stop ISOs (which includes ISO 50). I have no idea if an additional analog amplifier is also used in the 5DII (or the 7D, for that matter). If so, it's consistent with Sanjayfg's statement that ISO 50 is better than ISO 100 (lowest noise without sacrificing the highlight headroom).

Note that all of the data in the above links are read noise only (e.g. shots with the lens cap on and at the fastest shutter speed). Shot noise (and other, less significant sources like thermal noise) are ignored.

 

[Loswr]

The 5D Mark 4 II uses a Digic 4 while all the cameras in the test use a Digic 3 or earlier. The DIGIC chips actually do the control themselves so the DIGIC 4 may or may not do this in the same way (remember it uses 14 bits for image, more than before so you get 16,384 shades of light as opposed to the Digic III;s 4,096 shades from 12 bits))

More misinformation. Digic III is 14-bits (check the specs of the 40D or 1DsIII). Actually, 12- vs. 14-bits is not terribly relevant - even Canon's 14-bit cameras deliver only about 9 stops of useable dynamic range. FWIW, that's based on testing I've done personally with my 5DII and 7D.

 

[dr croubie]

Actually, 12- vs. 14-bits is not terribly relevant - even Canon's 14-bit cameras deliver only about 9 stops of useable dynamic range. FWIW, that's based on testing I've done personally with my 5DII and 7D.

DxOMark has some nice nerdly-graphs and such.
ranks the 5d2 at 23.7 usable bits (ie, 7.9 bits per pixel). 7d comes in at 22 (7.33bpp). CF the best full format (according to them), d3x @ 24.7 (8.2bpp) and the best of anything Phase One p65+ @ 26 (8.66bpp).

dynamic range the 5d2 comes in at 11.9evs, 7d at 11.7evs, the pentax k5 tops out 14.1evs, and d3x 13.7evs for comparison.

So changing the ADC from 14 to 15 or 16 bpp is not going to do much, at least how i interpret those results, because the sensor can't even deliver enough range to use the full number of bits as it is (although, speaking from my experience with digital audio processing, more bits will help 'smooth out' the quantization noise, make colour transitions look smoother and such).