See update at the end.
ISO/Noise tests. 1080/24p, lens cap on. In FCP, native format, maxed the mids, applied same filter to every clip. Thanks to jjgross.
See pictures below.
You’ll see that the only suitable ISO values for 7D video are 160, 320, 640, 1250 and maybe 3200. Interesting!!
One thing that still remains a horrbile artifact of SOME 7D cameras (including my own :-() is the pattern noise at low ISO levels. See next post.
Read the complete post to see the test images…
Update: i am at a loss now… i receive messages from other 7D users that, on the contrary, only whole ISO numbers should be used… I think i’ve lost my own thread… 😉
Marvelsfilm's Blog 
These noise trends confirm similar findings for the Canon 5D2…multiples of 160 iso are better than muliples of 100 iso.
So much for using a linear Film iso analogy…digital is a different world.
Hi,
I find it hard to believe, that ISO 160, 320, etc. would be better than the full-stop ISOs. Of course, belief is not proof. I don’t have a 7D, but I have samples from two dozen copies of it; unfortunately none with 1/3 stop ISO suitable for noise measurement.
The speciality of ISO 160, 320, 640, 1250 and 2500 is, that pixel clipping on raw level occurs at (or around) 12650, in contrast to 15300 with ISO 200, 250, 400, 500, etc. (ISO 100 and 125 are on their own). The useful pixel values start at about 2048; thus the effective pixel value range with ISO 160, 320, etc. is about 10600, while the range with ISO 200, 250, 400, etc. is about 13150.
I noticed it several times, that some photographers measured the “base noise” in absolute terms (as the standard variation of the pixel values) and compared that between different ISOs – but that is nonsensical. The same noise measured in absolute terms is much greater relatively to the ISO 160 value range than to the range with ISO 200.
Now, to the posted crops with different ISOs: I don’t know, which program created them. A program dealing with the raw data has to be prepared to account for the different pixel value ranges. For example the preliminary support of the 7D in ACR regards all ISOs with the came clipping point, making ISO 160, 320… images darker than they should be (and less noisier); at the same time clipping will not be noticed, which becomes obvious when one looks at a clipped sky, rendered as pink by ACR.
I suggest you to shoot overexposed sky and look at its rendering; that gives an idea of the software’s understanding of the raw data.
Cheers
Gabor
So…. what you’re actually saying is, is that the 160/320/640 range is actually the opposite of the Highlight Priority? Thus limiting the dynamic range?
Very very interresting… We should write a seperate article about this.
Martin Beek
Yes, it is the opposite of HTP: the effective ISO with HTP is lower than selected (one full stop lower), thus normal metering causes underexposure or prevents overexposure.
ISO 320 is actually ISO 400, *overexposed*. This does not limit the dynamic range (nor does HTP increase it), it only “shifts” it.
The effect of using ISO 320 is the very same as using ISO 400 with +1/3 EV bias. Of course the higher exposure leads to lower noise – but it leads to clipping sooner.
If you create a four-shot set, ISO 200, 250, 320 and 400, the same scenery, constant illumination, metered with +4/3 EV (or even more) bias to make sure, that there is clipping (i.e. the exposure wqith ISO 400 is half of that with ISO 200), I create a demo from the raw files showing the difference between the clippings.
Cheers
Gabor
Addendum: to a complete demonstration, one more ISO 320 shot would be useful, with 1/3 EV LOWER bias than the others (this IN ADDITION to the shot with the same bias as the other shots).
I received raw files targeted towards this issue and created a demonstration with detailed explanation; see http://www.cryptobola.com/PhotoBola/Canon7D_ISO.htm
The result is in short:
– the 7D’s hardware does not support the fractional ISOs; they are created by underexposing the previous lower ISO respectively overexposing the following higher full-stop ISO,
– the +1/3 EV ISO steps, like 125, 250 reduce the dynamic range by 1/3 EV,
– the -1/3 EV ISO steps, like 160, 320 cause an “overmetering” by 1/3 EV, which may cause clipping,
– the ISO steps 100 and 125 cause about 1/4 EV higher exposure than it should be: ISO 100 is rather 119, thus the metering for ISO 100 may cause clipping.
Wow. So it only has ISO 100 200 400 800 1600 3200 6400 native.
And what about HTP? How does that influence noise?
Oh, wait, you measure in RAW. HTP doesn’t have anything to do with RAW, right?
Still trying to figure the camera out…
Gabor, from what I gather, the lens cap was on the camera so light has nothing to do with it. These are tests done in complete darkness. You can clearly see the difference in sensor noise when light exposure isn’t in the equation.
Hi: I think I solved the mystery here. I’ve made a few prior posts, so I’ll just copy paste (with minor edits). Sorry in advance, this is long, but this explains everything.
his is actually a trick, but the results are nonetheless interesting, and here’s why we see what we see:
Sensor’s sensitivity (and I use term sensitivity here loosely) to light is controlled by gain (we all know this). More gain applied to the sensor (higher iso) — more signal loss (how many electrons/photons of light are recorded), hence more visible noise. This also explains why we lose contrast at low iso settings.
Now, let’s look at how fractional iso is achieved. Taking all the complexities out of the way, basically operation is reduced to sensor gain set to “native” gain mode with bias applied. So , at let’s say iso160, what is actually happening is that camera is recording at iso200 with -1/3 stop bias applied. That results in clipping of the highlights, so when color correction filter (e.g. curves) is applied to bring out visible noise pattern, the overall image appears to be cleaner. In other words, noise is there, but due to highlights clipping the final image in turn appears clearer.
In more detail… There are 19 photos above that show noise levels similar to tests in those videos. These are the values represented (low to high)
100-125-160-200-250-320-400-500-640-800-1000-1250-1600-2000-2500-3200-4000-5000-6400
these are all steps of one-third stop (1/3EV). Agreed? Good.
Now, let’s pick out “base reference” iso.
100 – 200 – 400 – 800 – 1600 – 3200 – 6400
these are all steps of one stop (0 EV) Agreed? Good.
next, let’s get a bit into more detail of how image is actually recorded (but not too complex). There are actually 2 stages that take place:
1. Analog Exposure (that’s physical light)
2. Digital Exposure (converting analog system to bits for image recording).
Agreed? Good.
Here’s why some images are clearer than others.
Fractional ISO is achieved in two ways:
1. Underexpose analog + Overexpose digital. Result: lose shadow detail (dynamic range)
2. Overexpose analog + Underexpose digital. Result: clipping of highlights
When we examine each image sample one by one, we can see that in case #1 we get more grain than base iso. In case #2 we get low noise picture. Of course, image degrades in higher iso ranges, but relatively we can see significant noise reduction in case#2. Here they are one by one (note, these are all 1/3EV – use website link for reference to images)
iso100 – base iso
iso125 – shot at 100, analog -1/3; digital +1/3 – increased noise/lost shadow
iso160 – shot at 200, analog +1/3; digital -1/3 – lower noise/highlight clip
iso200 – base iso
iso250 – shot at 200, analog -1/3; digital +1/3 – increased noise/lost shadow
iso320 – shot at 200, analog +1/3; digital -1/3 – lower noise/highlight clip
iso400 – base iso
iso500 – shot at 400, analog -1/3; digital +1/3 – increased noise/lost shadow
iso640 – shot at 800, analog +1/3; digital -1/3 – lower noise/highlight clip
iso800 – base iso
iso1000 – shot at800, analog -1/3; digital +1/3 – increased noise/lost shadow
iso1250 – shot at 1600, analog +1/3; digital -1/3 – lower noise/highlight clip
iso1600 – base iso
iso2000 – shot at1600, analog -1/3; digital +1/3 – increased noise/lost shadow
iso2500 – shot at 3200, analog +1/3; digital -1/3 – lower noise/highlight clip
iso3200 – base iso
iso4000 – shot at 3200, analog -1/3; digital +1/3 – increased noise/lost shadow
iso5000 – shot at 6400, analog +1/3; digital -1/3 – lower noise/highlight clip
mystery solved.
quick correction
“iso320 – shot at 200, analog +1/3; digital -1/3 – lower noise/highlight clip”
should read shot at 400.
I can confirm that for the 5D Mk II the best ISO to shoot at are 160, 320, 640, 1250 the lowest ISO settings, eg iso 160 is lower noise than ISO 100.
I had to do some extensive test as a part of a Diploma of Photography Course. I have found it quite handy to have in the back of my mind when shooting in low light and blending exposures for HDR.
Totally mis-analyzing test. Take a look in here and you guys (Marvels) may learn something:
http://shootintheshot.joshsilfen.com/2010/05/13/canon-hd-dslr-native-iso/
Got inspired by Tony Lorentzen’s video noise test and devised a still picture equivalent experiment, nothing fancy or scientific. If interested, please feel free to check the results out.
Interesting – so is it better to shoot on a multiple of 160 for reduced noise, or multiple of 100 to get the best dynamic range?
And, what is result of doing these tests with 1/2 stop increments? And why ? 😉
Gary
You should have read my otter articles on the subject first. There has been done a lot of experimenting and analyzing by very respected people that all contradict each other. I value Canon’s information and Fraunhofer institute’s tests over Zacuto’s shoot-outs.
Martin