During pre-production for the narrative film that I am shooting as a companion piece to the 5D Mark III and D800 test, which is now on part 3, we decided to see the entire ISO range of both cameras and see how well they handled under and overexposure. I wanted to see how the internal codecs would stand up to this extreme test, so both cameras were set to the variable bitrate 28mbps codecs in the camera. The lenses were kept the same - the best of the best from both Canon and Nikon, the 70-200mm f/2.8, with the Canon being the newer version of that lens.
Here is the ISO Range Test Video - be sure to download in 1080p for better quality. On a side note, Vimeo's upload speed is obnoxiously slow, so that's why this was posted so much earlier than this article.
Below are the complete settings used for the test:
- Nikon and Canon 70-200mm f/2.8 lenses
- Factory Default for both Cameras
- High-ISO Noise Reduction Off
- Highlight Tone Protection and Dynamic Lighting Off
- Neutral Picture Profile
- Canon IPB and Standard Nikon Compression
- Noise Reduction Off
- WB: Cloudy
- Shutter 50 F/2.8
Since Nikon's liveview mode is a much bigger crop than Canon's, it's a little more difficult matching frames. The two lenses are exactly the same in terms of f-stop, so there shouldn't be much difference (if at all) in light transmission at similar zoom settings.
I wouldn't worry too much about the color differences, this is just concerned with ISO (though as usual Canon is a bit more saturated in their default settings than Nikon). One major thing that you notice immediately is that the D800 is significantly brighter at equivalent ISOs than the 5D Mark III. This actually seems to be less apparent the higher the ISO, but for everything under about 6400, the D800 is about 1/2 to 1/3 of a stop brighter at the same ISO. I'll know soon enough exactly which camera is underestimating or overestimating ISO. But in the meantime it makes it somewhat difficult to compare them equally, which is why I tried to give some equivalencies at the end of the video. Regardless of those differences, it's pretty clear that the 5D Mark III is cleaner. This isn't quite a definitive High-ISO test, it is more to test the range of under, and overexposure and what that does to the noise and the codec under natural lighting. The D800 also doesn't tell you ISO settings, instead it gives you Hi 0.3, 0.7, 1.0, and 2.0. When still images of these ISOs are brought into a program that can read metadata, the exact ISOs are 8,063 (0.3), 10,159 (0.7), 12,800 (1.0), and 25,600 (2.0).
By not changing the f-stop or shutter, you get a better sense of how the sensor and compression deals with noise. All sensors have a base ISO and they interpolate and boost gain to get to other ISO numbers. As you can see both cameras are noisier in the shadows at low ISOs when severely underexposed. A lot of this is the codec, but some of it is just noise from the sensor. Both of these cameras perform much better at standard compression when you attempt to expose as brightly as possible without clipping. The way they handle noise, they are both fairly clean at extreme ISOs when the main subject (in this case a wall), is 3-4 stops above middle grey (if your paper white is Zone X) - just before clipping.
People these days talk about filmic noise or grain, but obviously no such thing exists. It's not the same thing, and it performs much differently than actual silver halides. The one thing they do share in common is that the bigger the grain or noise particles, the more noticeable and obnoxious they are, but I'll leave final judgement on that until I upload the ProRes Nikon footage and the ALL-I Canon footage. Even though the Nikon has a bit more noise (1-2 stops at the same ISOs - which I stated previously and then corrected - silly me), the pattern isn't affected as greatly by the compression. I like the way that Nikon handles the H.264 compression in the shadows at lower ISOs, but when we get to around 8,000 or 10,000 ISO the compression is much less dramatic, and the noise becomes more apparent - and that's when the Mark III really shines.
I also did a quick resolution sample at 200% at two different ISOs for both cameras, to see how accurately they maintain detail as ISOs increase. They both do a good job keeping resolution, and it's much better to use post-production software to reduce noise rather than in-camera noise reduction. Like I said before, the Nikon is resolving more. Not significantly more, but enough to notice on fine detail like the ceiling. The review will get more exhaustive later on, but for now a few things are definitely clear, the 5D Mark III is cleaner (even when you compare the equivalent brightness ISOs), but the Nikon D800 is sharper.
Link: All Parts