DXOMark concludes that the Canon EOS R3 is the ‘best low light performer’

Jan 22, 2012
4,473
1,329
I shot the R3 in low light yesterday evening and found that it's about a stop better than the R5. I'm not talking about per-pixel apples and oranges, but versus the R5 downsampled to 24 megapixels. It still shot about a stop better. I don't know if it's from the sensor or also partly from some extra noise cooking going on, but whatever they're doing, it's working pretty well.

The 24 megapixel aspect is, surprisingly, not much of an issue for anything that isn't reach-limited. It really does shoot above its megapixel weight. The comment from some days ago that it's about the sharpness/detail equivalent of the 5D4's 30 MP is probably exactly right. But at this point - despite loving the R3 - I don't plan on keeping it because the things I shoot typically *are* reach-limited, so the R5 is going to be better for me.

My average published shot is perhaps 35 percent of the frame, and typically winds up 300 dpi. To give perspective, with the R3, if you shot in landscape orientation something you wanted to be on the cover of a magazine, you'd be able to crop only 1/8th of the height of the image before you sank below 300 pixels per inch. And, yes, you can get good cover shots with 4 megapixels, yada yada, but if you brought a 4 megapixel sensor to this knife fight, you'd get gunned down most times by someone else. I had the same decision back when I had the A9 Mark II (24mp) and the A7r Mark IV (61MP). I found eventually that I kept using the A7r4 whenever I went to shoot wildlife, even though it was really the remit of the A9; and it was just for the resolution. Been there before. But I do recognize my use case isn't normal.

As an aside, I see a lot of people assuming the R1 will be a higher resolution sensor, but that actually doesn't conform to any Canon precedent. I expect the R1 will be a really cool camera with a sensor that will - as it has for the last decade, without fail - disappoint those wishing for resolution similar to that found in competing cameras. I'm hoping, instead, that we'll see an R5 Mark II in two years (yes, it'll be that long) that adopts much of the R3 tech and even more resolution.
"1 stop better". That is a wow claim.
 
Upvote 0
Jan 29, 2011
10,675
6,121
As an aside, I see a lot of people assuming the R1 will be a higher resolution sensor, but that actually doesn't conform to any Canon precedent. I expect the R1 will be a really cool camera with a sensor that will - as it has for the last decade, without fail - disappoint those wishing for resolution similar to that found in competing cameras. I'm hoping, instead, that we'll see an R5 Mark II in two years (yes, it'll be that long) that adopts much of the R3 tech and even more resolution.
Well the 1DS, 1DS II and 1DS III easily beat (I and II) or matched any other resolution available, so I don’t know quite what precedent you are thinking of.

I also have no doubt Canon have done a ton of research and will produce the camera they think will sell best, current prospective purchasers are using 45mp (Nikon) and 50mp (Sony) options.
 
  • Like
Reactions: 1 user
Upvote 0
Which is which?
The one on the right (file name ending in 22909) is from the R5. I'll do a proper series in the next day or so on a tripod with over and under exposures at two or three high ISOs (maybe 12k, 25k, 50k).

I am biased to my own sort of work, which is very light-limited, forcing me to underexpose in order to get a shutter speed that doesn't ruin the shot. It's this that I'm most interested in testing.
 
  • Like
Reactions: 1 user
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,355
22,528
To tide you over, here is an imperfect, non-tripod, slightly different angle quick test shot of R5 and R3 at 50k ISO underexposed a couple stops and raised one stop. I expect that the more controlled tests will show similar results....
50k iso underexposed by one stop is equivalent to iso 100k. The official iso range of the R5 is up to 51k wheres the R3 and the R6 both go up to 102k, so you really are taking this to beyond the extreme range of the R5. It would be interesting to see how the R3 does vs the R6. If you are working at these very high isos, then the resolution of a high density sensor is wasted in any case as the noise wipes out fine detail.
 
Last edited:
  • Like
Reactions: 3 users
Upvote 0
50k iso underexposed by one stop is equivalent to iso 100k. The official iso range of the R5 is up to 51k wheres the R3 and the R6 both go up to 102k, so you really are taking this to beyond the extreme range of the R5. It would be interesting to see how the R3 does vs the R6. If you are working at these very high isos, then the resolution of a high density sensor is wasted in any case as the noise wipes out fine detail.

I think this is mostly a fair point, Alan. It might be most fair to compare 25k ISO, which is within the specs of both cameras when underexposed. That said, that the R3 goes another stop higher within its spec is indeed a quite an advantage for high ISO shooters. I look forward to some bench tests on all this.

By the way, I've never allowed ISO to go to 50K before. With my 5D Mark IV, I capped it at 12K in the settings, and with my R5, I capped it at 25K. That 25K was sort of a desperation ISO setting, where I might be able to use a picture if I didn't crop at all. Maybe. That's my impression of the 50k shots I took a couple nights ago with the R3.

I expect there to be not nearly the same sort of difference in perceived quality at 1600 to 6400 ISO - as is shown by some of the comparisons you already see on the internet.
 
  • Like
Reactions: 1 user
Upvote 0

DBounce

Canon Eos R3
May 3, 2016
500
544
To tide you over, here is an imperfect, non-tripod, slightly different angle quick test shot of R5 and R3 at 50k ISO underexposed a couple stops and raised one stop. I expect that the more controlled tests will show similar results....
It was evident after only a short while of shooting with the R3 that it is excellent in low light. I prefer this to the massive void in the ISO range that you are presented with when shooting on some dual gain systems.
 
Upvote 0

docsmith

CR Pro
Sep 17, 2010
1,223
1,109
50k iso underexposed by one stop is equivalent to iso 100k. The official iso range of the R5 is up to 51k wheres the R3 and the R6 both go up to 102k, so you really are taking this to beyond the extreme range of the R5. It would be interesting to see how the R3 does vs the R6. If you are working at these very high isos, then the resolution of a high density sensor is wasted in any case as the noise wipes out fine detail.
Is it really a difference? If both are native ISO 51,200, and then bumped a stop in post, it is the same.

By the R3/R6 being "Native" to ISO 102k, that is just saying ISO gain is added on the analog side, before the ADC, whereas with the R5, it is now added post ADC, or the gain is added to the digital signal.

But for tiggy's tests, they were both added within the respective native ISOs, then then had a stop added in post. The fact that the R3 could have been native at ISO 102k is not relevant.

Please correct me if I am wrong here.

Tiggy...btw, thanks for sharing...this is exactly where I am interested in seeing if the R3 has an advantage over the R5...higher ISO, images scaled to the same size (Manny Ortiz just posted a video where he concluded the R3 was much better than the R5, but didn't scale...granted, the difference was apparent enough, I am not sure scaling would have mattered).
 
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,355
22,528
Is it really a difference? If both are native ISO 51,200, and then bumped a stop in post, it is the same.

By the R3/R6 being "Native" to ISO 102k, that is just saying ISO gain is added on the analog side, before the ADC, whereas with the R5, it is now added post ADC, or the gain is added to the digital signal.

But for tiggy's tests, they were both added within the respective native ISOs, then then had a stop added in post. The fact that the R3 could have been native at ISO 102k is not relevant.

Please correct me if I am wrong here.

Tiggy...btw, thanks for sharing...this is exactly where I am interested in seeing if the R3 has an advantage over the R5...higher ISO, images scaled to the same size (Manny Ortiz just posted a video where he concluded the R3 was much better than the R5, but didn't scale...granted, the difference was apparent enough, I am not sure scaling would have mattered).
Actually, iso is not a gain, it is a measure of the ceiling level of the signal that is being collected. The noise is in the signal itself, not the recorded iso. If the iso is set at 51k but is underexposed by a stop, the signal and its noise are those which would have been if the iso had been set at a 102k. The noise in the signal is a combination of the noise in the photon flux, the quantum efficiency and the circuitry, and the circuitry of the R3 is designed to operate to lower levels of light, ie to higher iso. DxOmark‘s measure of the low level iso depends on a calculation of the S/N, and was equivalent to a 0.42 stop advantage to the R3 over the R5.
 
Upvote 0
DxOmark‘s measure of the low level iso depends on a calculation of the S/N, and was equivalent to a 0.42 stop advantage to the R3 over the R5.
Where are you getting a 0.42 stop difference in SNR? That would correspond to a ~2.5dB difference in DxO's measurements when what they've presented is more like a 0.5dB difference.
 
Upvote 0
Thanks, so I can keep using my old LR license for organisation purposes while moving to another program(eyeing ACDsee or On1 or something else) for editing purposes.
Based purely on my preference and experience, that is what I do with a couple of exceptions
I really like the RAW engine PL has, for all my bird photos, I'll import them to LR, tag and rate and do the admin bit, then the chosen ones go to PL (there is an option to export from LR to PL as a project) and I edit the RAW there. From there to Photoshop if it needs further work.
For non bird photos (around 1% of the total and mainly of my daughter) I simply use LR/Photoshop because I don't "need" the level of detail/sharpening PL provides.

So I'm as niche as it gets and as bad a reference as you can imagine :LOL:

I should probably spend more time navigating PL to come to terms with the UI and stop complaining.
 
  • Like
Reactions: 1 user
Upvote 0
Jan 22, 2012
4,473
1,329
The one on the right (file name ending in 22909) is from the R5. I'll do a proper series in the next day or so on a tripod with over and under exposures at two or three high ISOs (maybe 12k, 25k, 50k).

I am biased to my own sort of work, which is very light-limited, forcing me to underexpose in order to get a shutter speed that doesn't ruin the shot. It's this that I'm most interested in testing.
I can see your need! Absolutely can.
 
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,355
22,528
However, I wonder if underexposing is better or using higher ISO. :)
That's a very good a question. It depends how iso invariant the sensor is in the iso range you are working. You can check this by going to https://www.photonstophotos.net/Charts/PDR_Shadow.htm
Here are the plots for the R5 and the 5DIV. You can see the plot is close to being flat between iso 800 and 40,000 for the R5. This means that there is no disadvantage to underexposing in RAW at isos greater than 800 and increasing in postprocessing to the equivalent of iso 40,000. For 5DIV, the plot slopes up from iso 800, which means you would get better results by getting the iso right and not underexposing.
Screenshot 2021-12-07 at 09.40.13.png
 
  • Like
  • Wow
Reactions: 4 users
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,355
22,528
Where are you getting a 0.42 stop difference in SNR? That would correspond to a ~2.5dB difference in DxO's measurements when what they've presented is more like a 0.5dB difference.
Go back to my post number 18 where DxOmark's numbers are in the image. The R3 has a low light iso of 4086 calculated by DxOmark from their S/N data, the R5 is 3042. The difference in stops (ev) is log(base2) of 4086/3042 = 0.4256.
Note the iso for the R6 is 3394, only 20% poorer than that of the R3, and for which I calculated 0.27 ev. I think my calculations must be consistent with DxOmark as they write: "A difference in low-light ISO of 25% equals 1/3 EV and is only slightly noticeable." https://www.dxomark.com/dxomark-camera-sensor-testing-protocol-and-scores/#sports
 
  • Like
Reactions: 2 users
Upvote 0

tron

CR Pro
Nov 8, 2011
5,222
1,616
Many thanks Alan for this. It seems more obvious if someone decides to put round numbers like 1000 and 4000 iso.

Log(base2) (4000/1000) = Log(base2) 4 = Log(base2) 2^2 = 2 stops <--- It's a nice way to explain to others and it kind of puts things into perspective. For example using your formula I found out that my R5 is about 1/5 stop better at low light than my D850 = practically equal.
 
  • Like
Reactions: 1 user
Upvote 0