April 10, 2012

5D Mark III/D800 Hands-On Part 4: Mark II vs. Mark III vs. D800 - Candlelight

Well it's taking a bit longer to get these up than I'd hoped, but I think the evidence in this one is the most obvious of any test I've seen so far. This time the Canon 5D Mark II has been thrown into the mix, in addition to the 5D Mark III and the Nikon D800. We were in a room with large windows well into the night, and so there are a couple streetlights providing very basic illumination at the higher ISOs. Other than that the only light is the candle right in front of our model Sasha.

ISO 500 was chosen as the starting off point because it was the first ISO that really started to produce an image with the candle. Here are the settings for the video, this time the Nikon was recorded to ProRes 4:2:2 HQ from an Aja KiPro and the Mark III was recorded ALL-I internally:

  • ISO 500-25,600
  • Nikon 85mm f/1.4 lens
  • Default/Neutral Color Profiles
  • Canon ALL-I & Nikon KiPro ProRes 4:2:2 HQ
  • Noise Reduction Off
  • WB: 3200K
  • Shutter 50
  • F-stop stayed the same throughout - F/1.4

And the video (again be sure to download the 1080p file for better compression):

As you can see, the D800 is still significantly brighter at lower ISOs, and still a bit brighter at higher ISOs - this has not changed and the same exact lens was used on all three cameras. I'm actually going to try to figure out what ISOs these cameras should really be reporting based on some light meter readings in the next day or so. The results speak for themselves - the Nikon D800 is sharp, but very noisy. It's a camera that needs a lot of light, but when you give it that light, it has a beautiful picture coming out of the HDMI.

The 5D Mark II still holds up very well, but there is a noticeable shift to green at higher ISOs which has always been the case for that camera. The Mark III and the D800 do not have a shift in that way. I think some of the noise of both Canon cameras is being hidden by compression. With the Nikon going straight to ProRes, it's got the cleanest image, and thus the noise is going to be much more noticeable - but also easier to get rid of in post.

The Mark II is obviously a bit noisier than the Mark III, two stops or so (just like Canon said). It's still usable because its noise isn't too noticeable out of the box until around 2500-3200 - mostly helped by compression. The Nikon is just a big mess by 6400, and it would take heavy noise reduction to bring it back. I still don't care what people say about film-like grain, noise is still noise, and if it's not part of your aesthetic, it's going to stick out unless you get rid of it. The Mark III is very, very clean. I would have no problem shooting at 6400 ISO with the Mark III and using it in a professional piece. There's a little bit of noise, but it is mighty impressive. Not only that, but the Mark III holds skin tones quite a little bit better than the Mark II at 6400. The D800 is a bit yellowish in the skin tones, not quite as pleasing, but nothing a little color correction can't fix at 4:2:2 color sampling. None of the cameras were corrected for the candle, they were simply left at Tungsten to keep it consistent.

So we've got a couple more parts in this series left, but the D800 is definitely hurt by its high megapixel count in low-light. The D800 is a camera that should be left at 640-1000 (which is comparable to 1250 and 2000 on Canon) for the most part and lighting used to compensate. It's much less of a run and gun camera than the Mark III, which can shoot in almost any light, thanks to its improved light gathering ability. Nikon didn't know what Canon was going to release when the Mark III was announced, but you can bet what they will be working on in their next version: aliasing. Aliasing is pretty bad, but we'll get to that soon enough.

Another big benefit for using the HDMI on the Nikon, there's a lot more color separation in the light from the candle. The Mark II has some heavy banding and the Mark III is a bit better thanks to the ALL-I, but the D800 really shines because of the massive amount of information contained in the 8-bit 4:2:2 HDMI.

I wouldn't count out either camera yet - they are both fantastic performers, and it's getting to a point where you really need the right camera for the job at this price point. Need the uncompressed HDMI and the resolution: Nikon (unless we all want to modify the Mark III). Need the low-light: Canon.

Anyone think the Nikon noise is pleasant? (I have a tough time deciding.)

Links: All Parts

Part 1: 5D Mark III/D800 Hands-On: Initial Impressions (Mark III)
Part 2: 5D Mark III/D800 Hands-On: Initial Impressions (D800)
Part 3: 5D Mark III/D800 Hands-On: ISO Range Test

Your Comment

80 Comments

I think I fell in love with your model and didn't read anything that came after the video

April 10, 2012

0
Reply
Michael Soomon

I feel you, bro. She's beautiful.

April 10, 2012

-1
Reply
tungah

^
this

April 11, 2012

-1
Reply
Raphael

Yes, she is and the music is very good too!!!

April 23, 2012

0
Reply
Thanos

Have you found anything more about the model? Qhat a perfect smile...

April 23, 2012

0
Reply
Martins

What a beautiful smile - I'l call her our 'Little Butterfly' - I wish I was 40 odd years younger!!!

What camera? - I would have liked to see the D4 included!!

April 23, 2012

0
Reply
Frank

All right, I understand this industry is dominated by guys but let's be a little more professional here. She's a model, an actress, and a friend, and she's in a few more pieces I'll be posting here - so compliments are fine but we're getting a little off topic.

The D4 is $6,000 - at that point it makes much more sense to buy a camera that is made to shoot video. From what I've seen the D4 has the same performance as the D3S in low-light, which would obviously be better than the D800, but again, at that price level it just doesn't make any sense.

April 23, 2012

0
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

Thanks for posting this. The mark III had some really strong results. I don't like the Nikon noise effect it looks terrible.

Cheers!

April 10, 2012

-1
Reply
Amelia

The Nikon is noisier, but I would like to see a comparison that shows them at equivalent brightness, not just equivalent exposure.

April 11, 2012

0
Reply

Did you find that recording 422 on the NIkon/KI added any latitude that could be pulled in post?

April 10, 2012

-1
Reply

I would love to see how the footage looks graded with noise reduction...As you point out noise reduction could be huge for the nikon due to it's nearly uncompressed files. It's the only way we could tell what iso the nikon is really usable at.

April 10, 2012

0
Reply
Lev

Right. And to see if recording at ProRes adds exposure latitude. hmm.

April 10, 2012

0
Reply

In the preliminary tests that me and my DP Shawn have done, it wasn't a matter of adding more latitude, but you can push and pull areas that are nearly overexposed a little easier without them falling apart. So in that sense you gain latitude, but there isn't any more information than the camera is physically recording, it's just a matter of the compression being better so there should be less noise when you're dealing with extreme color correction.

April 10, 2012

-2
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

The Mk III looks better at 12800 than the D800 at 500. LOL

I don't know if it is the codec, but D800 seems pretty unusable from start even for web videos, and post noise filters can only do so much before transforming your image into a acrylic painting. It's a blotchy noise pattern very noticeable and unpleasant in my opinion.

I was wondering what the D800 would look like at lower ISO's, because it's at least 2 stops over the Canons. Perhaps less bright but less noisier too. It's like we have some high-gain control going on from the very start.

Well, it's a great comparison. Definitely would not go for D800 hundred based on it. But than again, as you said it, different cameras offering different solutions and performances at different conditions.

Is it time to start renting these cameras instead of buying for specific job requirements? I always thought the whole concept of DSLR is the possibility for the end user to own a digital-cinema-ish camera, but this might be changing, IDK.

April 10, 2012

1
Reply
mauborba

Part of what you're seeing at the low-end near 500 ISO on the Nikon is underexposure noise - but when a scene is properly lit it's much cleaner in the correctly exposed areas.

April 11, 2012

0
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

Yes the D800 is a total non-starter for video. Who cares about uncompressed 422 when the image is unrecoverably miserable and full of aliasing and moire. If the 5d3 only wasn't so soft (and I'm sure the Digic5 could do proper downscaling rather than binning without overheating). If they can hack the 5d3 that will be the DSLR option. But with the new cams coming out DSLRs are probably just a convenience thing for video by summer.

April 11, 2012

0
Reply
Peter Jensen

You're saying that the D800 is full of moire and aliasing. But the samples in normal light, and even in low-light like Joy Ride, look better than the soft 5D III samples. And it appears the new 5D III codec at 90 mbits is no better than the D800s internal 24 mbit codec, in fact it produces a softer, lower resolution image. That's the real non-starter for many.

April 12, 2012

-1
Reply
Markus

Not really.
Fact is that the video from 5D3 can be sharpened in post much more than both Mk2 and certainly the D800 The soft output form the 5d3 is a non-problem. The simple reason for this is the lack of Moiré and false details/colors. There is a reason Canon choosed the new sensor size dividible by 1920.

A sharp image does not equal resoluted detail, a sharp image, especially when it comes to DSLR video is just aliasing and false details that makes you think it looks good.

Don't believe me, check this video out: http://vimeo.com/39292404

April 13, 2012

-1
Reply
Snake

I've been shooting with a 5DmkII for ages and moire is rarely an issue for me as I don't shoot charts or patters. So I see the D800 as being just as effective in for film as the 5DmkII. Far from a non starter. I considered the 5DIII but the biggest letdown was resolution. This camera is simply not true 1080p. Results based on my early test shots are similar to 720p fauxed to 1080. So while having little moire is nice, the soft output makes the 5DIII at total non starter for anything video related.

The situation is so bad, and if you haven't please head over here
https://vimeo.com/39171656
that people are mutilating their 5DIII's to remove the OLP filter in the hopes of capturing something even close to what the built in coded of the D800 does.

The 5DIII is just horrible at capturing fine detail. The other issue I have is the noisy all I codec. It suffers from mosquito noise even at base ISO. You can swap to the 28mbps codec but then that falls appart on high motion scenes.

Sharpening in post? you can't sharpen resolution in period. sharpening the 5DIII output introduces horrible artifacts like halos when trying to match the lowest quality D800 footage, say nothing of the 4:2:2 color out which the 5DIII is simply unable to match even with the OLP removed.

It pains me but I'm selling my 5DII (and 5DIII) and going nikon for my video needs. The 5DIII is ok for web output where resolution isn't important. But for critical work where detail is king, the D800 is simply unmached by the canon. Since we already know the problem is the OLP filter, canon will no be able to deliver true 1080p video via a firmware update.

April 23, 2012

0
Reply
Yosh

Interesting test, but i'd love to see the results of these videos with basic post work, simple CC and NR, also perhaps shooting with cinestyle video profiles would be a better idea for both(tassinflat for nikon), can you do that for us? thanks in advance. I mean as of now this test doesn't help me make a clear assessment . So yeah.

April 11, 2012

1
Reply
Madhur

Joe has the cameras for a limited time, and with NAB about to happen he may not have much of a chance to do more testing. We'll do our best though!

April 11, 2012

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

I'll get more in depth about this as soon as I can, but there is a major issue with trying to use other profiles with the D800, like tassinflat. There is some sort of pre-processing that happens and it completely destroys the shadows and makes the footage unusable (at least with the profiles I tried so far). This may not be for all profiles, and I believe Samuel Hurtado just posted a beta profile for the D800, which if I have the time, I will try to use.

April 11, 2012

0
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

Thanks a lot guys, looking forward to learn more. cheers!

April 12, 2012

0
Reply
Madhur

Joe -

Can you tell me what is unusable about the TassinFlat picture profile? I've been using it on a few projects, and it seems great with the D800. ( https://vimeo.com/40788982 )

I'm curious as it sounds like you've had problems, but it has produced some great images for me (I've also been using Hurtado's flaat profiles, but haven't done any side-by-sides yet).

April 22, 2012

0
Reply

It turned the shadows into a grey/green mush of compression. It made it completely unusable. It works fine on parts that are normally exposed but it absolutely destroyed the shadows. I didn't have time to use Samuel's other profile but it's possible that it might fix this.

It seems like you got good results, so I don't know what was going on, it was installed correctly.

April 23, 2012

-1
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

I think the problem you're talking about is due to extreme underexposure combined with the wide dynamic range. I've seen it when shooting TassinFlat, but it's nothing a final grade can't fix. From Samuel himself:

"that intermediate 422 image has about 13.5 stops of DR, but very little information in the last 1.5 stops (very few brightness values devoted to that part), so pushing this with a strong curve is better than doing it on the final 420 H.264 image, but will only take you so far.

"btw: it's even worse with the Canons, where that intermediate image has very similar characteristics, and only 11.5 stops of DR"

It's the last 1.5 stops on the toe of the curve that contain very little information, and therefore it is more prone to blocking up. As Samuel mentions, the Canons do it too with the Cinestyle profile, though the cast appears to be magenta instead of green, which isn't a surprise at all given the vast difference in color hues from each camera.

By the way, from my tests I have found that these problem areas actually grade out ok, as long as you take a little care. When shooting really dark scenes, you have two options: shoot a profile that introduces some muck in the deepest shadows (that can pretty easily be graded out in post), or shoot a narrower profile that registers the shadows as black with zero detail. I'd prefer the extra range that I can play with later, but to each his own.

Also, with all that extra DR the Nikon is spitting out, you can bring up the exposure by 2-3 stops safely to retain much more shadow detail, and keep those areas out of the danger zone.

April 24, 2012

0
Reply

I wish the model was turned more towards the candle in the Canon footy to match the angle of the Nikon. Thanks for the test though!

April 11, 2012

-1
Reply

Since the Nikon has a 1.2x crop, it's a little difficult trying to get the same framing, but I here what you're saying.

April 11, 2012

0
Reply
avatar
Joe Marine
Editor-at-Large
Shooter/Writer/Director

Nikon has a 1.2x crop??? Never seen that info before. It's a full frame 35mm sensor.

April 11, 2012

0
Reply
JS

yes it is a full frame sensor - the crop is on the viewfinder/LCD screen

April 11, 2012

0
Reply

It has a crop mode. You can select full 35mm, 1.2, or 1.5 modes.

April 11, 2012

0
Reply

To clarify, when the Nikon is in liveview, it does more than just crop top and bottom, it windowboxes. This is very clearly stated in the manual if anyone doesn't believe me. You can't ever get the full sensor with top/bottom 16:9 crop. This has to do with the way Nikon lineskips to get closer to 1080. It's actually a little more, and then it downsamples the the rest of the lines to get to 1080. This is why my framing can never be the same with the same lens. I also mentioned this in my first nikon post. It's either 1.1 or 1.2 crop - I did the math on part 2 of this review.

April 11, 2012

0
Reply
Joe Marine

Joe, it just looks like the model isn't looking in the same direction, as indicated by the shadow from her nose. Her head appears to be turned more to her left and the candle looks more in line, in the Canon.

April 11, 2012

1
Reply

"The Mark II has some heavy banding and the Mark III is a bit better thanks to the ALL-I"

it has already been established that ALL-I is easier to edit on slow machines, but IPB actually delivers better image quality than ALL-I (less codec artifacting):
http://www.dvxuser.com/V6/showthread.php?279229-Canon-5D-Mark-III-IPB-co...

also, my take about the "different brightness levels at same ISO" is that, with the D800 having a lot more DR, it has to look brighter in the midtones and shadows: it has to bring everything except the highlights a bit up to make room for that increased DR in the shadows to be recorded; but it's just a different curve, not higher ISO

April 11, 2012

1
Reply

Thanks for the test. At least for once there is a tester that acknowledgment the different ISO of the different camera. Fro photo test site it was at least 2/3 stop and I think it is enough to be sited because from two frame grab I would say that the Nikon 6400 is very close to the Canon 12 800. The second thing is that the overall image has much better DR in the Nikon than the Canon. The result is that you see much more info in the shadows but more noise because they are crushed in the Canon ones. The last thing is the resolution (the 5d2 looked even sharper than the 5d3 in the 200% test), the Nikon could accept a lot of NR to achieve the same softness than the 5d3.

I am not saying that the Nikon is better, but if you take the three things above, you are going to get much more closer result. The last thing I would say is that I would be a bit more generous about the usable ISO of the both of these camera. Your test is not a low light test but what I would call a no light test. Shooting only with a candle is a no light test. General shooting condition with even just street light is much higher than this and I would say both could be use at least 1/1.5 stop higher. As some have said it would be nice to get the same exposure and do some crushing and NR to see the difference.

April 11, 2012

0
Reply
Danyyyel

Shouldn't you be using the multiples of ISO 160? Apparently that is when the sensor (at least the 5d2s) works best? I may be missing something, though :)

April 11, 2012

-1
Reply
monkeyflow

yes, but Joe hit some of those multiples in the test. To be fair, the test should have generic ISO steps because we don't know the sweetspot of multiples for the Nikon Camera yet. For my 5dmkII in high iso range I use 640 and 1250. Even at night shots with some basic lighting and a fast lens, I have not needed to go over 1250. And 1250 on the MKII is clean.

April 11, 2012

-1
Reply

Wow. Does the D800 rely on "in-camera" noise reduction? Or does the camera have it at all? Or is the 5D's that much cleaner?

April 11, 2012

0
Reply
Chris

canon's higher ISOs are actually different than Nikon. What canon calls X nikon calls Y and canon looks darker therefore canon seems to be fudging their numbers to appear higher (or nikon is understating theirs).

this renders the comparison mostly worthelss. you can't trust ISO numbers from camera makers since they have been playing these games to get better noise scores. You should compare similar brightness/exposure to get an effective comparison.

April 23, 2012

0
Reply
Yosh

Almost all videos go through post production so I would rather have the D800 and use my own denoising than to have the Mark III do it for me. And if thats what it looks like with practically no light. Just having an ipad in the room on flashlight will cure a lot of the Nikon noise.

April 11, 2012

-1
Reply
ghost

who shoots at 3200 without seeking some light or a very fast lens? If it's dark it's dark, shoot the scene as it is in the dark as cameramen having been doing on film for decades. This obsession for the pursuit of low light at low budgets is ridiculous. If you want low low candlelight light then swap out your Canon or D800 for a GH2 and a Voigtlander 0.95.

April 11, 2012

-1
Reply
Dan

GH2 is crushed by the mkIII for noise and light sensitivity. GH2 has rigid grid like pixels/banding in noise for high ISO's - and I'm talking about the hacked version. I'll take a MKIII anyday over it and take out the LP filter for true resolution.

April 11, 2012

0
Reply

I'd probably take the GH2, fast lense and additional lighting but you seem to be missing the original point.

Would you take a MkIII over a 500 ASA 35mm film stock? How about a MkIII over a Red with M-mount and f/0.95 lense? How often do you shoot a narrative feature using thousands in cameras and lenses but can't secure a lighting kit / generator?

April 11, 2012

-1
Reply
nobody

but the issue is that the GH2 crushes the 5DIII in detail.
http://www.eoshd.com/content/7631/panasonic-gh2-vs-5d-mark-iii

I just can't get over the smudgy look of the 5DIII. How can a 3500 dollar camera have such poor video quality? Even if it is less noisy, you'll take a hit which you can't recover in post resolution wise all the time. sharpening only adds faux detail at the expense of artifacts like halowing and ugly sharpened bokeh. you can't sharpen resolution in which the 5DIII fails to deliver.

much rather go for the D800 clean HDMI out which more closely matches the GH2 output with the benefits of full frame depth of field.

April 23, 2012

1
Reply
Yosh

You can also just rent the EF 50mm 1.0 ;) a few places have them even though they are discontinued

April 11, 2012

0
Reply
John Jeffreys

I would really like to see a comparison that takes into account the fact that the Nikon has a brighter image. ISO 500 on the Nikon is so much brighter than ISO 500 on either Canon. I knew the cameras rendered different from my experience in shooting stills, but I didn't realize they were that far apart.

April 11, 2012

0
Reply

agreed. canon seems to be overstating their ISO by quite a bit (or nikon understating it).

April 23, 2012

0
Reply
Yosh

Totally agree. Lets see the same strips side-by-side where the brightness is equal - I would guess the nikon at 500-1000 ISO and the Canons around 4000-6000 ISO (maybe even higher). THEN we would have something to compare.

April 24, 2012

0
Reply
Jason

"and the same exact lens was used on all three cameras"

hehe

April 11, 2012

-1
Reply
Raphael

am i wrong but if you light the scene correctly then noise becomes redundant ? higher iso settings compensate for low light so if the light is low do some creative lighting which will eliminate the need for high iso settings then grain is not an issue

April 12, 2012

-1
Reply

Pages