Just thought I'd share some quick tests i did using my canon 7d and canon 5d mk2 cameras. I was inspired to do some tests after seeing some similar tests done on Marvelsfilm’s Blog (http://marvelsfilm.wordpress.com) .

Anyway, it seems pretty clear to me that highlight tone priority has a big effect on low end noise, but that is to be expected since it basically pushes the image a stop to retain an extra stop of highlight information. It is also clear that white balance has a big effect on noise and the best isos to use are 320,640,and 1250. I think I will probably just tend to shoot iso 320 with HTP on and light as close to 5000k as possible, especially avoiding high isos in tungsten light, and probably turning off HTP at isos above 640 or 1250.

I did the iso noise test first and noticed the 5d had a lot of noise but realized it was blue noise due to being set at tungsten wb whereas 7d was at awb which effectively pushes the blue channel almost 2 stops. that prompted me to do the whitebalance noise test with the 7d, which was done at iso1600.

Interesting results, though, i always felt that the 5dmk2 gave a green cast above iso3200 and now i see quite clearly that it does though i am not sure why.

BTW all of these stills were taken with cameras with lens caps on etc and an adjustment layer has been applied to the whole image. The adjustment layer was added in photoshop and is a simple levels filter with gamma cranked to 6.0 and black clipped to 1 to increase contrast and make the noise really pop and be visible and difference distinguishable.

Interesting, but I still always question the lens cap on noise test. Something doesn't sit well well with me when this is used as a control. I think maybe because I think lens, what is being focused on, etc changes noise and maybe throws cap on tests out the window.

I dont think this is the best noise test either, but it seems to me that it is still a reasonable accurate, simple way to comparitively see noise in black values depending on iso and whitebalance. It doesnt say much about whether noise is noticeable with a lens on and an properly exposed image, but its hard to avoid seeing that there is clearly a correlation between noise and iso and noise and whitebalance in scenes that comprise only of black values. It doesnt seem to me to be all that big of a stretch to say then that based on only this information its reasonable to prefer iso 320,640, and 1250 and to prefer shooting at 5000k if you want to minimize low end noise. It also helps that this is pretty much what one expects given that most silicon sensors natively balance better at 5000k and that there is some information out there saying that slrs use different methods of acheiving extra sensitivity at minor iso levels (maybe only one iso setting for each stop of sensitivity change is actually mostly hardware gain related whereas the other might be software sensitivity adjustments, giving them more noise). If another more in depth test is documented that tells a more detailed story about noise in this camera in practical environments id prefer that but i havent seen many tests like that as of yet.

Great post, Noah. Thanks for sharing.

the best isos to use are 320,640,and 1250.

You might be interested to know that the reason why 320, 640, 1250 have less noise is because they clip highlights. When you select "ISO 320", what the camera actually does behind the scenes is set the analog ISO to 400, then reduce brightness digitally with a -1/3 stop pull. It's like the opposite of HTP (highlight tone priority), but only 1/3 stop.

If you shoot with normal contrast settings (e.g. "Standard" profile), the camera will clip that 1/3 stop anyway, so then it is good to use 320. But if you use low contrast settings, the difference can be visible.


I think I will probably just tend to shoot iso 320 with HTP on and light as close to 5000k as possible, especially avoiding high isos in tungsten light, and probably turning off HTP at isos above 640 or 1250.


For what it's worth, HTP should always be enabled when the ISO is over 1600. The noise difference between 3200+HTP and 1600 alone is nada, but one of them always has 1 stop more highlight headroom.

Very interesting info Daniel. Is iso behavior common for most canon SLRs? or have you tested the 7d? if so further details on your tests would be appreciated.

Very interesting info Daniel. Is iso behavior common for most canon SLRs?


There are some similarities and differences. For example, The 5D, 1D3, and 1Ds3 used separate analog gain amplifiers for the intermediate ISO settings, rather than digital iso gain as in the 50D, 500D, etc. The 5D2 is digital gain.

or have you tested the 7d? if so further details on your tests would be appreciated.

I haven't tested it. And actually, I'm not very certain (yet) that the information I have is correct. I have only tested my 5D2 and found it correct. What I do for testing is to take a few shots of color checker chart with consistent illumination (e.g. daylight) at a variety of exposures and run them through IRIS, which is a free astrophotography program. Rawnalyze is another really great tool.