Still Photography Performance

You can’t talk about a digital camera without at least spending some time talking about the camera’s performance as, well, a camera. Personally, this is one part of camera reviews that I despise working on. It’s very difficult to make any kind of meaningful point or discussion here.

Exposure, Bracketing, and HDR

The metering system in the 5D mark III provides photographers with a range of ±5-stops in either 1/3 or 1/2 stop increments from the baseline middle gray value. Though both the meter in the viewfinder and the meter on the top LCD, only indicate a range of ±3-stops and you’ll have to count clicks for shutter/aperture increments beyond that.

The covered range is on par with Canon’s EOS-1D X and Nikon D800 and D4. There’s no longer a functional disadvantage to using the 5D compared to the 1D or Nikon’s bodies, at least as far as the range the meter allows you to operate over is concerned. In my experience, the limited display of ±3 stops isn’t an impediment, as the ±4 and ±5 stop settings aren’t frequently necessary for proper exposure compensation.

To go with the expanded metering range, the 5D mark III also allows for expanded bracketing in 2, 3, 5 and 7 frame groups, with as little as 1/3 and as much as 3 stops between frames. In the largest configuration, of 7 stops and a 3-stop gap, the camera will make exposures from -9Ev to +9Ev from the base exposure.

The bracketing range is certainly capable of being exceptionally large, in no small part because Canon doesn’t limit the bracketing step to only 1 stop max like Nikon does. However, I’m not sure that the full range is necessary. I’ve certainly never used it, and in my experience, such range can be excessive even when shooting for HDRs.

Canon has also included a built in HDR solution that combines 3 RAW images to produce an HDR JPEG. There is some flexibility in configuring the effect applied to the HDR, and of course, the camera doesn’t automatically throw away the 3 component RAWs, so you can rebuild the HDR later using 3^rd party software and the original RAWs. When shooting in HDR mode, the camera also shoots in continuous release mode without you having to change the camera over to it manually.

Ultimately, though, the HDR feature feels more like the kind of thing that belongs on an entry-level camera than a pro level one. That said, I don’t begrudge its inclusion, even if I don’t use it. Canon has balanced it’s accessibility between being buried enough that it’s not in the way for other use, and not so much that it’s hard or impossible to find when it’s wanted. It’s a top-level option from pressing the creative photo button.

Drive Modes, Frame Rate, and Buffer

The 5D mark 3 provides users with 7 drive modes; single shoot, continuous high, continuous low, silent single shot, silent continuous, 10-second self-timer/IR remote, and 2 second self-timer/IR remote.

The silent modes are new to the 5D line and offer users the ability to significantly curtail the sound of the camera when shooting. Reducing shutter noise is done in two ways. First, the drive and re-cocking speed of the mirror and shutter is reduced. Secondly, in single shot silent mode, the mirror and shutter aren’t re-cocked until the shutter release is released. The combination of the two techniques, allows the 5D mark III to be significantly quieter, though it’s not completely silent in the silent modes.

The self-timer/IR modes should be familiar to any recent Canon user; they provide a delayed release or the ability to use an IR remote, like the RC-6, to trigger the camera without touching it. Both the self-timer and IR releases can be used in lieu of a traditional wired cable release (such as the Canon TC-80N3 or RS-80N3) for reducing vibration in tripod shooting. Moreover, when mirror lockup is enabled, the self-timer will lock the mirror up before starting the timer to further reduce vibration.

The limited resolution increase for the 5D mark III over the 5D mark II, coupled with the significantly faster processor has allowed Canon to increase the maximum frame rate to 6 FPS (up from 3.9 in the 5D mark II). Given Canon’s claims that the Digic5+ processor is 17 times faster than the Digic4, it’s likely at this point that the mirror and shutter are now the limiting factors for frame rate on the 5D mark III.

Though the 2 FPS boost doesn’t sound all that impressive, it’s a 50% increase in performance form the 5D mark II. More importantly, at least in my experience, 6 FPS is right around the point where a camera can be used effectively in situations that demand high continuous frame rates. While I certainly enjoy the 10 FPS my EOS-1D mark III provides, I have never felt overly constrained with the 6FPS of my 5D mark III when shooting birds in flight or similar material.

Canon advertises the 5D mark III’s buffer as being good for at least 13 RAW frames. However, because Canon uses lossless compression for their RAW files, the size of the buffer in images will vary with the size of the files stored in it. At low ISOs where there is less noise in the images, and thus the images are less random and therefore more compressible, you’ll see the most available buffer space. While at high ISOs you’ll see less. The table below summarizes the reported buffer sizes based on camera ISO.

Additionally because the buffer is constantly being emptied to the flash card while the camera is shooting, the bust length can be increased further by using faster flash cards. Canon suggests that UDMA7 cards will provide an 18-frame burst length.

Based on the buffer size, reports form magic lantern, and images of the 5D mark III’s main board I expect there’s 256MB of SDRAM. With 256MB of buffer RAM, Canon is clearly factoring in a conservative 15-20MB/s card write speed in their calculations. If you’re using extremely slow/old cards, you may run into problems with the very slow write speeds. This may or may not have been a contributing factor in the problems I had with some old (circa 2009) SanDisk Extreme 30MB/s cards.

With 400x UDMA compact flash cards I see somewhere around 16 frames in a burst at continuous high. On the other hand, the SD interface is quite a bit more limited. With either a 400x or 600x SDHC or SDXC card, at best I only see 15 frames before the buffer is completely exhausted and the frame rate drops considerably.

At 13 frames, the buffers is good for just over 2 seconds of continuous shooting at continuous high (6FPS), and just over 4 seconds of shooting at continuous low or continuous silent (3 FPS). Personally I would have preferred a guaranteed 3 seconds of buffer at 6 FPS, with UDMA cards stretching that further. As it stands, Canon’s decision to limit the onboard RAM, likely to cut costs with a side effect of some more product differentiation, pushes up my costs to get a 3-second buffer considerably by forcing me towards very expensive 1033x Compact Flash cards.

I tend to leave my camera in either continuous high or continuous low regardless of what I’m shooting. Continuous low’s 3 FPS frame rate is slow enough that it’s very easy to squeeze off only one frame when that’s what is desired. However, it leaves you the ability to shoot multiple frames without the quality sapping effect of repeatedly mashing the shutter release.

Here again, Canon has left out software configuration options that are in the 1D series that certainly help photographers tune their cameras to meet their needs. Specifically in this case, I’m talking about the ability to configure the actual FPS for the continuous drive modes. With a 1D series body, I can tune the FPS rate to match what I need, and therefore balance how long my buffer will last in seconds.

For example, dropping 1 FPS from 6 to 5 FPS, while using a 400x CF card, would allow you almost another full second of shooting.

Image Quality

A number of factors make up what I’m grouping together under the title image quality; these include noise, dynamic range, and color accuracy to cover the most important ones. I’m not going to go into any great depth on any of these factors, ultimately because I don’t see any need to rehash what’s already been covered in great depth by DXOMark et al. The reality is that unlike say a lens, there’s just not that much variance in the electronics, including a camera sensor. Making a survey of the existing published sensor performance data, further substantiates that there isn’t much need for another ground-up set up measurements.

Canon’s 5D mark III, like most other Canon CMOS DSLRs, exhibits a dynamic range curve that shows a marked flattened shoulder at low ISOs. This is almost certainly due to Canon’s continued use of off chip amplifiers and ADCs, which introduce more noise into the readout, as seen in by significantly higher read noises at low ISOs compared to the exceptionally low read noises at high ISOs.

The results of this is that Canon’s dynamaic range performance is depressingly poor. At least at low ISOs. On the other hand, sometimes I do feel that the differences, which currently amount to about 2 stops of shadow detail, is somewhat overstated in the bigger picture. First, not everybody can shoot at ISO 100 or even ISO 400 all the time, and once you start shooting at higher ISOs the gap Nikon currently enjoys evaporates completely. By ISO 400, Nikon’s lead is only about a 1-stop advantage, if that, and from ISO 1600 onwards there’s no difference to talk about.

Moreover, while it’s easy and valid to criticize the camera on the technical points, it’s important to remember that 2 stops of dynamic range isn’t going to make a mediocre photographer a great one. In some respects, it’s far to easy with the internet and the ability to readily quantify camera performance, to get mired in quantitative details and forget about actually making images.

Admittedly, this is the kind of point that someone who’s emotionally invested in their camera gear might bring up to try and sooth their angst at using the lesser performing product. I certainly found myself wondering if that was the case too. However, the reality for me, at least, is that the bulk of my time shooting isn’t at ISO 100, it’s at ISO 400 or 800, and that means I’m not especially disadvantaged by the 5D mark III’s more limited dynamic range.

That said, make no mistake Canon absolutely needs to improve their low ISO dynamic range going forward.

Noise performance has never been a significant practical issue for me with the 5D mark III. Though at the extremely high ISOs, especially H1 (51,200) and H2 (102,400) I’ve found images pick up a purplish tonality due to chroma noise. That said, careful noise reduction at those ISOs can produce usable images, at least in cases where having an image is more important than having a good looking image. Ultimately, the question of noise performance is going depend significantly on what you intend to do with the camera. Astrophotography and fine art uses are going to be much more demanding than say journalism uses.

One of the biggest frustrations I have with Canon’s cameras is their implementation of ISO or gain. They simply don’t make it easy on the photographer to get the optimal quality, in terms of lowest noise and best dynamic range, easily.

With the 5D mark III, Canon continues to use their Base/Push/Pull gain system that they use in non-1D cameras. As I understand it, the camera’s read amplifiers work in full stop steps providing analog ISOs of 100, 200, 400, 800, 1600, 3200, 6400, 12800, and 25600. ISO’s 1/3^rd stop faster, that is ISO 125, 250, 500, 1000, 2000, 4000, 8000, and 16,000, are created by underexposing the base ISO by 1/3^rd of a stop and pushing the resulting data by a 1/3^rd stop. The final +2/3rds ISOs (i.e. 160, 320, 640, 1250, 2500, 5000, and 10,000, 20,000) are derived by over exposing at the next higher full ISO and pull the results by a third stop.

Expanded ISOs, as always, are pure multiplications of the closest base ISO; 51,200 and 102,400 are ISO 25,600 under exposed by 1 and 2 stops respectively. ISO 50 is ISO 100 under exposed by 1 stop.

What this means is that instead of simply choosing the lowest ISO to get the best quality image, you’ll want to choose the ISOs in a much more complicated fashion. The typical advice is to use multiples of 160, then multiples of 100, and avoid multiples of 125. This advice is certainly oversimplified.

Below ISO 3200, you will want to use multiples of ISO 160 for the best noise performance (so ISO 160, 320, 640, 1250 and 2500). Above ISO 3200, you’ll want to use the lowest ISO you can.

ISO 100 and ISO 50 can be used if necessary to achieve longer exposure times, but they produce an image that has less dynamic range and more noise than ISO 160, and comparable dynamic range and noise performance to ISO 320.

What’s even more frustrating with Canon is that their EOS-1D bodies use a completely different gain system that biases the quality towards completely different ISOs. While you want to favor +2/3^rd stops below ISO 3200 on the 5D mark III, you’ll want to favor full stops below ISO 1600 on a 1D Mark III, and the EOS-1D X has a gain structure that suggests favoring the lowest ISO you can use without any other considerations.

Much like the situation with dynamic range, Canon really needs to address their system for generating ISOs in two ways.

First, the systems between 1D and non-1D cameras should be aligned so that different rules for ideal image quality don’t apply. Where ISO 160 and 320 are the best quality settings on the 5D mark III, they’re ones that should be avoided on a 1D or 1Ds Mark III or 1D mark IV, and a EOS-1D X user can just use the lowest ISO that gets them the shot.

Second, Canon should move towards a gain system where quality decreases monotonically with the increase in ISO. That is the rule for the best image quality should simply be to shoot at the lowest possible ISO.