5D3 vs D800 - A Technical Analysis (UPDATED & EXTENDED) - With Samples

Started Mar 30, 2012 | Discussions thread
SubPrime Senior Member • Posts: 1,238
Question for Legion regarding methodology

You've clearly put a lot of effort into this, Legion.

First of all, I am brad agnostic. I own both a 5D2 and a D700 and have an equal investement in glass for each. Both the 5D3 and D800 look outstanding and I'll be upgrading to both at some stage in the neat future, but I’m in no hurry. As far as I can tell, there is practically no difference in ISO performance between them, especially when one looks at the images in their entirety.

As someone with a scientific background however, I have some questions about your methodology.

This comparison is also done in RAW converted to 36 MP and cropped.

Please excuse my ignorance as I don’t use ACR for raw conversion. May I ask how you converted the 22mpx up to 36mpx? Is there a way to do resize in ACR or was this done once you imported the files into Photoshop?

All crops are 100% crops with the 5D3 resized to the D800 and colour corrected to minimize differences in raw.

May I ask why you colour correct the images you subsequently based you judgements about colour on? That strikes me as counter intuitive. I could understand a white balance adjustment to level the playing field, but what colour correction did you apply?

This leads me to the perplexing observation that the D800 gave more accurate colour reproduction while the 5D3 had better looking colours. That seems like a contradiction. From an artistic director’s perspective, how can colours that are not accurate be better looking?

Above ISO 1600 the 5D3 just has both more detail and less noise than the D800. At 1600 ISO they are equal in both respects. At 3200 there is an edge to the 5D3 and at 6400 the 5D3 just pulls away and just keeps widening the gap.

The basis for sound methodology is producing results that can readily be duplicated by others. I was not able to duplicate yours, so I am curious about what parameters you applied in RAW. Could you please list the noise (colour and luminance) and sharpening values that you used? Perhaps a screen grab would be helpful?

The nature of the noise is part of what the raw image provides and part of what I am examining is the nature of the noise and how it responds to processing. My goal is to put out an unbiased view of real world results, because pictures aren't taken "in theory".

While I am inclined to agree with you, raw processing is a bit like cooking, and everyone has their own recipe. No two people will produce the same results when processing a raw (unless they simply spit out defaults).

A complete analysis would include before and after screen grabs to illustrate the effects of your adjustments. Looking at the raw files and comparing them to your results, I am a puzzled by how you got there by applying the same parameters to both files.

The way this test was done was to render both raw files @ 36MP and then adjust the noise sliders so there is 3% noise. Sharpening is equal, and is set to low, with more sharpening the difference becomes more exagerated. 3% noise @ 36MP

Is that 3% of colour? Luminance? Again the exact values would be very helpdful

PCMagazine and Consumer Reports among others use this standard and I find it widely agreeable.

I am not sure that PCMagazine and Consumer Reports are the go to resources for people wanting to buy $3,500 FF camera bodies, besides, I would be surprised if they even bothered to open raw files. CNET Asia’s review was so superficial that it read more like a PR release than any kind of analysis, so I wouldn't take their conclusion too seriously.

Now according to your method, you opened the files in RAW , resized, cropped, noise reduced and sharpened. Then you compared them side by side. Sounds fair and I look forward to trying it out as soon as I have your values.

I propose another method, and would like your opinion as to it’s validity.

1. Open both images in raw, resize and crop with all NR and Sharpening disabled.
2. Merge both images into the one file – as layers
3. Spit out a before NR image comparison.

4. Apply the same noise reduction and sharpening to both layers using something like Noiseware.
5. Spit out the final result

Às for AF, FPS and pop up flash, I guess it comes down to personal needs. Personally, the pop up flash is an annoyance. AF on the D700 was pretty solid but hunted in low light, so any improvement is a bonus. Low light AF is without a doubt, the real point of interest for me. FPS is great to have for those that rely on it, but accurate AF is what make or breaks low light/high ISO images.

Which brings us to Dynamic Range. The graph from Comcast shows the 5D3 lagging behind the D800 up to 1273, so I don’t know how you came to the conclusion that the DR differential only exist below 400 ISO. They are practically identical until 12,800.

Do you have your own method to measure these values?

At 25,600 ISO, the advantage of the 5D3 over the D800 is the same as the D800 has over the 5D3 at 800 ISO, yet you consider the differential insignificant at 800 ISO.


Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow