Canon R3

[Fast Glass]

I just looked at the noise comparison, didn't realize Bryan already did the leg work.

I think it is interesting to note unless you are shooting at 100k ISO where the R6 takes a lead. The R5 looked ever so slightly noisier but also better detail. With noise reduction on at say ISO 6400, the R5 actually looked just the tiniest bit cleaner and ever so slightly better detail. This was the PS comparison, DPP was not quite as good.

In other words there was more difference between the software used and what settings and technique used than the difference in the two bodies. Partly there could be differences to mitigate it when Canon made them. But that is purely a guess on my part.

The R5 definitely compares well unless you consistently shoot at 100,000 ISO. And even then I would say it is roughly a 1/3 to 1/2 of a stop. But I believe that is because the R6 has a higher native ISO. So you are pushing the exposure an extra stop. But even at 51,000 and with noise reduction you are splitting hairs.

All this to say there is not a one stop difference in noise performance if you are making a fair comparison.

[Kayaker72]

I do agree with your overall point. Modern cameras are amazing, perform very well. Noise is much better controlled than a decade ago. While the R5 is not as good as the R6, it has better (lower) read noise at low ISOs than the 5DIV/5DIII.

Couple of quick points:

• The second you downsample the R5 to the R6 size you have lost the benefit of the extra resolution.
• Downsampling compared to RAW (which is still from DPP) is a more heavily processed image compared to a less processed image.
• Read noise as calculated on photons to photos is different than what Bryan did with Read noise being measured under complete black conditions and then values calculated (shown here). Bryan's test is a highly illuminated subject and involve a level of processing, sometimes a heightened level of processing.
• Read noise is is important for blacks, such as the nightscapes Kari does, but also for shadows and DR. Until recently, this is one of the reasons Sony outperformed Canon in low ISO DR tests, they had lower read noise. Canon mostly caught up at the 5DIV generation and caught up with the R5/R6/1DXIII generation.
• Interpreting Bryan's tests are somewhat subjective. I do prefer the noise of the R6 over the R5. It isn't a stop, but it is better.

I actually really like what Bryan does with his noise tests, DPreview as well for that matter. But, these tests are very different in that they are well illuminated vs as black as possible. Thus, they eliminate all but the worst of read noise (1-6 electron/photon variance doesn't mean much when your pixel is hit with 20,000 but means a lot when the pixel is only hit with 10 photons/electrons). Which is fine for those that shoot well illuminated subjects. But read noise can become an issue in poorly lit conditions.

[DavidEccleston]

Noise gets complicated. It increases as resolution increases, yes, but you can reduce that to an extent by scaling the noisier picture down to the same resolution as the cleaner image. You can use the high-megapixels when at low ISO, and process to a cleaner lower MP image when necessary. I'm sure the higher MP count still comes out a bit worse, but not to the extent that just comparing raw output would suggest. DXOMark's "sports ISO" score for he R5 vs R6 is like 3000 vs 3300. Assuming that's a highest-usable ISO based on some noise threshold, it's a fraction of a stop difference.

Brant, about your largest print comment. We've been conditioned to want 150+ DPI for images, but that only makes sense when at normal photo album / book type distances. As an image is printed larger, you need to be further away to view it. The image on a giant billboard is likely far less than 150 DPI. Requiring a large size at high resolution is a very niche market...

With 5K monitors being standard on iMacs for years at 5120x2880, or 14.7MP, and the R3 being 6000x4000 (according to https://petapixel.com/2021/07/28/can...-data-reveals/ ), it doesn't leave room for much horizontal cropping, or straightening, before you can't use a photo as a wallpaper image. Which shouldn't be the case for a brand-new professional camera. So in that sense, I agree, 24MP seems too low for 2021.

[Fast Glass]

@Kayaker72

I agree there are caveats, I know there will be a slight advantage to the R6 at the same output size. And there are limits to Bryan's noise samples for sure. But my point was it is not close to a stop loss of noise just because of inefficiency associated with the extra pixels. But at the same time, the extra resolution also defines the noise better and can do slightly better job of reducing noise. Also since there are still close in performance there could be hardware differences between the two in order to optimize them for their perspective resolution. But this would be a pretty big assumption on my part. In other words there are more variables yet.

But that was the only thing I wanted to clear up, we are in agreement.

I think the other thing to keep in mind, and why comparing it at the same resolution is so important, it's the output size.

What magnifies issues is not solely resolution, it is also output size.

Would I rather have a image at 50mp vs 20mp at say 20×30. Let's just say it was shot at f/11 so we introduce some diffraction. And let's just say it was shot at 1600 so there is a modest amount of noise. The 50mp image will still look better. All things considered.

The higher resolution image is going to be resolving stuff better, apart from the slight noise differences between the cameras overall noise performance at the same output size you will not be magnifying the noise and lens defects the same way when you hit 100% on your monitor. This is critical to keep in mind.

When we look on our monitors at two different output sizes because of resolution we don't get a accurate idea of what resolution is doing to the image in the real world. What we are actually doing is looking at an print that is significantly larger. Hence not really a fair comparison.

To simulate this on our monitors we need to uprez the lower mp file to the higher one. This way we get an idea of what it would look like at a given print size. If we are more interested in the affects of resolution of a very large print.

[Fast Glass]

@DavidEccleston

That's a good point. I feel like it would have been easy enough to have at least 30mp, I know it's only 6 mp difference. But as you pointed out, it's not that high anymore with high resolution monitors coming about. And with 8k not that far off 24mp won't even fill up the monitor natively. But by that point I would suspect that the R3 will be updated with a mark II. But still an interesting thought process.

[Kayaker72]

Just to step out of the details for a second, let's look at this another way.

The R3 is clearly being positioned as a high fps camera and it is a burner at 30 fps.

It is meant for those that want to capture an exact instant. Some critical moment.

With the 30 fps in mind...what resolution would Canon's recent cameras suggest is possible with their current technology?

The R5 can do 45 MPs at 20 fps, but does drop you down to 12 bit to do so. For simplicity, let's just say 45 MP = 45 MB files (file size increases with ISO and other factors), but for comparison, this is a reasonable assumption.

45 MB x 20 fps = 900 MB/sec; 900 MB/sec / 30 fps = 30 MB. So, using the single digic X sensor in the R5 as a benchmark, the best current Canon tech can do with a 30 fps camera is 30 MPs. As the R5 does overheat, hopefully the R3 does not. And R5 does 20 fps at 12 bit and hopefully the R3 does 30 fps at 14 bits, is 24 MP really that unexpected?

Next, do we really think there isn't a market for the R3 given its specs? Did the 1Dx cameras not sell well?

The R3 is a high FPS speed demon. It is not the resolution monster you are looking for.


Getting back into the details.

• Total noise vs Read Noise. These are not the same. Test conditions are important here. Read noise becomes more important the less light you have. DPreview, Bryan, DXO mark...their noise tests are well lit. This is important. Taking the illustration I started a bit further, if you take a pixel that receives 10 photons/electrons, but has read noise variability of +/- 3 electrons and add gain to get ISO 3200 (32 x) that is 7 to 13 electrons x 32 or a 224 to 416 e- expected signal range from different pixels exposed to the same low light source. That is noisy. But, saturate those pixels with a signal based on 20,000 photons/electrons per pixel but has the same read noise of +/- 3 electrons per pixel, that is 19,997 to 20,003 x 32 or a signal range based on 639,904 to 640,096 electrons/photons...not much variation in signal intensity...not much noise. In fact, other types of noise will take over from read noise. But, ignoring those other types of noise, the only difference in this example is how well the pixel was saturated. So, the noise right at the sensor is critical for scenes where you want true blacks. And, the general trend across multiple brands, higher MP = higher read noise. Does this equate to issues for all types of photography, not at all. But, this is why the Photons to Photos test is under "black" conditions. All the other tests are well illuminated. I am not surprised by well illuminated tests showing a different noise than tests conducted black. Other noise sources will become more prevalent.
• Great example on the retina display. Sure, if you want 1:1 resolution, 24 MP is a bit tight. It works, but it is a bit tight. But, how much is enough? If you want to follow this evaluation that calculates the typical human eye can resolve 530 ppi at 20 inches. I've seen prints at 150 ppi. I am happy with them. I do not feel conditioned as I have evaluated it. If you want to crop and display at 1:1 on your Mac, the R3 is not your camera. But, if we want to chase what the human eye can truly resolve, the R5 is 8192 pixels / 530 = 15.5 inches. No cropping. The 24 MP R3 would give you 387 ppi on that 15.5 inch display. Is that enough, or do you want 530 ppi? The R5 can fill a 27 inch wide screen at 300 ppi, the R3 would be 222 ppi on the same screen. Is that worth the trade offs? Double the file size/memory space? Or do you want 30 fps to catch the moment a javelin is launched?
• Output. Absolutely agree. Have to pick your output. Personally, I've gravitated toward the 24 x 16 print at 300 ppi, or 34.5 MPs, as my standard. I own an R5 for a reason. Truth be told, I did not pick it on resolution alone, better EVF, top display, whispers of a better AF, etc factored in. But, resolution was a factor and had the R6 been ~30 MP, I likely would have saved some money.

So much of this is personal preference. But the R3 is a tool. I suspect it will be, and already has been, used to take many great photos by many great photographers. It will not be for everyone.

[Fast Glass]

I don't think you really understand how that is not relevant in this comparison.

That only works at full resolution, which doesn't translate into final output size. All that is telling you because of the higher res you are getting more noise. No disagreement there. But it's not super helpful in this comparison. My point there is not a lot of loss in efficiency with adding extra pixels, at least not in the 45 to 50mp realm. Modern tech has minimized this significantly.

As far as the FPS and resolution are concerned. Sony was able to do 30fps, with some caveats, at 50mp. Remember this is being touted as a sports camera, to me this is the new standard for a pro sports camera in this price point. 24 is just not being very competitive in my opinion. I'm sure the R3 is an absolute monster in every way, but even with using the 1Dx III. That was my major gripe, resolution. Even though I loved that camera to death in every other way. I was grabbing it more often than the 5D IV even though it has more resolution. I had said before if this had 30mp, I'd be so much happier with this combo and it could actually make it a viable all rounder instead of being as niche as it is.

And if Canon is indeed developing an 85mp camera, that would make sense for a high resolution monster. Fuji already make 100mp offerings.

I realize not everyone has a need for high resolution, but as tech progresses higher quality comes with it and Canon is still living in the past and Nikon and Sony defiantly got this part right.

After seriously looking at the Sony and Nikon offerings. I just can't switch, it's a personal choice just because I can't see myself shooting anything else. Canon gets everything else right. So I'll just hang in there, but I'll be all to happy when they release a 50mp or higher offering for the pro market.

[Kayaker72]

. Modern tech has minimized this significantly.

My last take on this. I assume this is about the noise discussion. I'll phrase it like this, all of these statements are true (specifically about read noise):

• Modern tech has minimized "this" (specifically read noise) significantly
• As an example the R5 has lower read noise than the 5DIII
• The R6 still has lower read noise than the R5 which does lend itself to certain scenarios
• Higher MP cameras tend to have higher read noise than lower MP cameras.

Again, all of those statements are true. The reason why I even brought this up is the exchange you had with Kari about aurora photography. Low light/nightscape images are a scenario where read noise becomes more of an issue, thus there would be a negative impact for Kari's type of photography of more megapixels. For blacks, where signal to noise ratios are low, lower megapixel cameras still show a benefit over higher MP cameras. Now part of that discussion was motion blur, but even that could be effected as if you dial in your ISO to have acceptable noise in the blacks, a higher ISO would allow for faster shutter speed and less motion blur.

The impact for well lit images drops off dramatically, as David's DXO example showed, 3300 vs 3000, not much of a difference. So, if you want to say that "modern tech has minimized this significantly" for noise in well lit scenarios, I am going to agree most of the time. But, for blacks, the Photons to Photos measurements indicate the R6 is about 1 stop better in read noise control which will absolutely affect low light photography.

Can read noise still affect well lit conditions, yes. If you think about it, "expose to the right" (ETTR) was about two things: using the whole dynamic range of a sensor, and avoiding read noise. The left side of a histogram has a lower signal to noise ratio (the 10/3 scenario from above) while the right side of the histogram has a higher signal to noise ratio (the 20,000/3 scenario from above). If you think about it, "expose to the center" would have its benefits as, in ideal conditions, that would give you the most latitude in post processing. But, it is "ETT Right" as the right has a higher signal/noise ratio. Has modern tech minimized this, yes, the noise floor is lower, but it still exists even in some well lit conditions, for example, if you underexpose. I have seen this even with the R5. Quick examples, but a series of birds against snow, I exposed for the snow, tried to pull shadows up. I still really like the images, but there was noise in the birds. Modern tech has not minimized this to the extent I had no issues.

Final point on this, I do have a noise issue with the R5. I am still trying to understand it as it is inconsistent. It isn't awful and I will still praise the R5 as a camera (it is great). Others have seen it as well and think that Adobe is not processing modern Canon files efficiently. But, it is luma noise, which could easily be read noise, and my other theory is that as the camera warms up, you start to see more read noise. So, if it isn't the software, then the high MP of the R5 is contributing to this with both higher read noise (compared to R6), and then more heat caused by moving more data. Even in well lit situations, but testers might not see this as it would only show up during intensive use.

To sum up, I find it misleading to state that there are not trade offs going to higher MPs. I am sitting here with my R5 and have several instances of these trade offs. While I was going to upgrade my computer anyway, the R5 certainly has hastened that decision as my current computer freezes up half the time downloading R5 files. Never an issue with my previous cameras. Resolution increases linearly, file size increases exponentially. Absolutely there are trade offs, with diminishing returns. Different people want and value things differently. The R5 is an amazing camera. I have no regrets, it is worth it, modern tech is amazing, but there are absolutely tradeoffs to higher MPs. There is a balance to be struck.

[HDNitehawk]

A new definition,

noise

1, a discussion among photographers involving the mp count of a particular camera.

read noise

2, an internet thread discussing the subtle differences between high mp cameras and low mp cameras, often deteriorating in to a discussion of shadows and camera manufacture comparisons.

[DavidEccleston]

HD, don't throw shade on the discussion... it will increase read noise.