Posted by David Kjaer on February 17, 2010 at 7:00 PM
It appears Canon has given up on low noise and are just interested in pixels. Read the latest test review from DXO Mark the new Mk1V has worse noise performance than the 6 year old 1Ds MkII and is hardly better than the 1D MkIII. It isn't even in the same 'ball park' as the new Nikon D3s. I certainly would have preferred if Canon had kept the 1D Mk1V to say 12.5 MP with better noise performance!
Posted by David Bostedo on February 17, 2010 at 7:59 PM
David - Be careful in how you interpret the DxOMark scores. They're just a tool and there are a few things to keep in mind that show you can't simply look at the "Low Light ISO Ranking" and draw too much of a conclusion, since resolution is not factored in. I'd say Canon still thinks noise is very important, and continues to make improvements, however small.
1) Sensor size generally trumps everything.
Overall noise in final images is largely a factor of the whole sensor's efficiency in capturing light. What I mean is, a larger sensor has an inherent noise advantage in prints/screen images even if the individual pixels are noisier. Add in the fact that you can generally make larger pixels on a larger sensor and the advantage for larger sensors grows.
The 1D MkIV has a smaller sensor than the 1Ds series, and as such I'd expect that even higher res 1Ds sensors would be better noise-wise. It's not anything to do with Canon's attitude toward noise as much as the 1Ds size sensor capturing more light.
In fact, the 1D MkIV is the highest ranking camera with a sensor smaller than "full frame". And ALL of the "full frame" cameras, including the Nikons, Canons, and Sonys, score higher than any smaller sensor camera. (The medium format digitals are a special case, and they actually AREN'T designed for high ISO shooting.)
2) Noise in a fixed print size is helped by increased resolution even when the per-pixel noise increases
A lot of people compare pixel against pixel which isn't usually a very valid comparison. Smaller pixel are inherently more noisy as they collect less light. If you being to factor in your final output, however, the higher res sensors usually have an advantage. For instance, if you assume an 11x14 output at 300 ppi, or a fixed size digital image, a high-res sensor will oversample each pixel which will cancel out the amount of per-pixel noise.
So an image from a 1D MkIII with less noisy pixels will likely still be noisier than the same sized print from a 1D MkIV.
3) The DxO Mark data is not normalized for a fixed output size unless you use the "Print" selection in the comparison tool.
The data in this case is put on a level playing field by down sizing all output to 8 MP. That way, the resolution advantage for noise is accounted for, and you see things evening out amongst the cameras.
--------------------------------------------
As an example of all of this, Nikon has released a couple of stunning cameras for high ISO in the D3 and D3S. Despite accomplishing this, the D300 and D300S are only OK by most people's estimation (and DxO Mark) when it comes to noise. And you wouldn't say that Nikon doesn't care about noise! This is largely a sensor size issue, and there's no magic bullet to getting the same low noise on smaller sensors.
Now comparing apples to apples, the 1Ds MkIII has much noisier pixels (as you'd always expect) than the Nikon D3. Yet when you factor in resolution, the 1Ds MkIII is less than 1/2 stop behind the D3 in high ISO performance - and has a lot more resolution allowing for much more detail at lower ISOs.
Now this ignores the fact to some extent that some people buy higher res sensors so that they can print larger. In that case, you have a trade-off between print size and noise, and resolution. Nothing's perfect, and companies choose different ways to approach this -- Canon by giving people higher res at the cost of more per pixel noise (not necessarily more print noise) and Nikon going the other way.
Posted by Dan Borzynski on February 17, 2010 at 10:19 PM
In every day, real world usage, every currently manufactured DSLR by Nikon and Canon has no noise issues at reasonable print sizes (which is now extending to 16"x24" and beyond) and at reasonable ISO levels (under 1000).
I shot for years on Velvia 50. The whole concept of getting something useful at ISO 800 is simply mind boggling to me. And by useful, I am looking at both publication and art print sales, and is completely sharp, color accurate and without having grain/noise the size of golf balls.
At this point in the development of DSLRs, unless ISO needs are in the thousands range or higher, noise is shouldn't be a consideration in the purchase of a camera body.
Posted by David Bostedo on February 17, 2010 at 10:51 PM
...unless ISO needs...
Dan - For a lot of people, I think they are more ISO wants than needs...including for myself. Do I WANT to be able to use ISO 12800 and still get nice prints? Definitely yes. It would let me shoot easier in some situations where my current camera barely passes muster. But as a pure amateur, I don't NEED it (which is one reason I own an Olympus camera).
I also have nothing film related to put me in that mindset, so I've definitely noticed situations where I'd like to be able to go at least to ISO 3200 if not higher.
Posted by Nate Chappell on February 17, 2010 at 11:08 PM
I agree with David Kjaer about being disappointed with Canon in the noise department. I've heard that the Mark IV is a little better than the Mark III - about 1/3 of a stop to 2/3 of a stop. This is certainly not enough of an improvement for me to want to go spend 5K to upgrade. I will wait and see what the Mark V looks like and get by with my Mark III and maybe a 7D in the meantime. This is a small improvement and they are a long way behind Nikon in this respect. The D3 and even the D700 are quite a better - although as David B noted they are full frame cameras.
When photographing birds and animals in tropical forests in countries like Ecuador and Thailand believe me it would be nice to have a camera like the D3 that produces little noise up to 2000 iso or more.
Posted by David Bostedo on February 18, 2010 at 1:05 AM
...they are a long way behind Nikon in this respect.
Nate - I don't really think they are very far behind at all, technology-wise. If you compare the D3X to 1Ds MkIII, for instance - a pretty head's up, apples to apples comparison, they are essentially equivalent. The D3X has a bit more resolution so is ever so slightly better for it's pixel size.
It's just that, for whatever reason, Canon hasn't decided to make a low pixel count full-frame sensor to compete with the D3. If they did, I'm willing to bet it would, at worst, be just a small notch below the D3. The D3 (and D3s) aren't magic, as wonderful at high ISO as they are. Low-res, large pixel, large sensors are MUCH easier to make cleaner at high ISO.
As you pointed out, though, Canon also hasn't done anything to make a big leap forward and get you to buy a Mark IV (for noise anyway). But they DID manage to add 5 more MP of resolution into the sensor and STILL improve the pixel level noise - and that's not trivial.
Posted by David Kjaer on February 18, 2010 at 5:26 AM
Hi David
Thank you for your detailed reply I was aware that even if the noise performance was the same as the old 1D MkIII in DXO's test - in practice the Mk1V would be close to 60% better due to the higher resolution. That said I still feel say 12MP on a 1.3 crop (assuming most people buy the Mk1V for action images) is a high enough resolution for for most application and the increase in noise performance would have been a better trade off.
I agree that Canon clearly are capable of making lower res. large pixel sensors with better high iso performance (which as you point out are much easier to make & cheaper) I just wish they would and give you the choice as Nikon do with their users.
Re. the Nikon D3X it's pixel pitch is quite close to that of the Mk1V but has quite a marked all round improvement in sensor performance.
However on a brighter note a few of my friends already own the Mk1V and are delighted with AF performance.
David
Posted by David Bostedo on February 18, 2010 at 9:30 AM
Re. the Nikon D3X it's pixel pitch is quite close to that of the Mk1V but has quite a marked all round improvement in sensor performance.
That's true, but for purely noise (SNR), they are pretty much equivalent (per pixel, not including resolution). In fact, according to DxO, the Canon actually has better dynamic range at ISO 800 and above, which is pretty impressive given the competition. The Nikon seems tuned for lower ISOs, as it's metrics drop off much quicker than the Canon as the ISO rises.
I'm curious in this case about the wide disparity between DxO's "Low Light ISO" ratings for these cameras. That rating is :
"...the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9EVs and a color depth of 18bits."
But if you look at the component graphs, the 1D MkIV looks equal or better in some categories, and maybe half a stop worse in color depth. I suppose that's the main difference then, and makes their "Low Light ISO" numbers 1992 and 1320. Seems like a really big difference compared to the individual graphs - but it's a pretty complex metric.
Posted by Larry Fischer on February 18, 2010 at 5:48 PM
David,
I own a D300 and D700.
The D300 is noisy and I don't like the prints as much as I do the D700.
One can talk pixel density, but the fact is on screen the noise between these two cameras is significant. Bostedo is the guru here so I would follow his expertise.
All I can say is I do a lot of low light photography and that was the impetus for the D700. Bright light and low ISO the D300 does fine.
the fisch
Posted by David Bostedo on February 18, 2010 at 8:52 PM
Bostedo is the guru here so I would follow his expertise.
Well, thanks...but I think that's a bit overstated. I just like to read a lot and muck about with the technical side of photography. I think it helps that I design computer chips for a living. I wish I had as much practical experience actually using all this pricey equipment as I do at reading data on it.
If you want to learn more about noise, I can highly recommend this article by Emil Martinec, physics professor and avid nature photographer. It's a bit technical, but very readable. He'd qualify as a guru I suspect.
Discussion
9 repliesAs a Canon user I am very disappointed.
Look here:http://www.dxomark.com/index.php/eng/DxOMark-Sensor/Camera-rankings
David
1) Sensor size generally trumps everything.
Overall noise in final images is largely a factor of the whole sensor's efficiency in capturing light. What I mean is, a larger sensor has an inherent noise advantage in prints/screen images even if the individual pixels are noisier. Add in the fact that you can generally make larger pixels on a larger sensor and the advantage for larger sensors grows.
The 1D MkIV has a smaller sensor than the 1Ds series, and as such I'd expect that even higher res 1Ds sensors would be better noise-wise. It's not anything to do with Canon's attitude toward noise as much as the 1Ds size sensor capturing more light.
In fact, the 1D MkIV is the highest ranking camera with a sensor smaller than "full frame". And ALL of the "full frame" cameras, including the Nikons, Canons, and Sonys, score higher than any smaller sensor camera. (The medium format digitals are a special case, and they actually AREN'T designed for high ISO shooting.)
2) Noise in a fixed print size is helped by increased resolution even when the per-pixel noise increases
A lot of people compare pixel against pixel which isn't usually a very valid comparison. Smaller pixel are inherently more noisy as they collect less light. If you being to factor in your final output, however, the higher res sensors usually have an advantage. For instance, if you assume an 11x14 output at 300 ppi, or a fixed size digital image, a high-res sensor will oversample each pixel which will cancel out the amount of per-pixel noise.
So an image from a 1D MkIII with less noisy pixels will likely still be noisier than the same sized print from a 1D MkIV.
DxO Mark themselves have an article on exactly this
3) The DxO Mark data is not normalized for a fixed output size unless you use the "Print" selection in the comparison tool.
The data in this case is put on a level playing field by down sizing all output to 8 MP. That way, the resolution advantage for noise is accounted for, and you see things evening out amongst the cameras.
--------------------------------------------
As an example of all of this, Nikon has released a couple of stunning cameras for high ISO in the D3 and D3S. Despite accomplishing this, the D300 and D300S are only OK by most people's estimation (and DxO Mark) when it comes to noise. And you wouldn't say that Nikon doesn't care about noise! This is largely a sensor size issue, and there's no magic bullet to getting the same low noise on smaller sensors.
Now comparing apples to apples, the 1Ds MkIII has much noisier pixels (as you'd always expect) than the Nikon D3. Yet when you factor in resolution, the 1Ds MkIII is less than 1/2 stop behind the D3 in high ISO performance - and has a lot more resolution allowing for much more detail at lower ISOs.
Now this ignores the fact to some extent that some people buy higher res sensors so that they can print larger. In that case, you have a trade-off between print size and noise, and resolution. Nothing's perfect, and companies choose different ways to approach this -- Canon by giving people higher res at the cost of more per pixel noise (not necessarily more print noise) and Nikon going the other way.
I shot for years on Velvia 50. The whole concept of getting something useful at ISO 800 is simply mind boggling to me. And by useful, I am looking at both publication and art print sales, and is completely sharp, color accurate and without having grain/noise the size of golf balls.
At this point in the development of DSLRs, unless ISO needs are in the thousands range or higher, noise is shouldn't be a consideration in the purchase of a camera body.
Dan - For a lot of people, I think they are more ISO wants than needs...including for myself. Do I WANT to be able to use ISO 12800 and still get nice prints? Definitely yes. It would let me shoot easier in some situations where my current camera barely passes muster. But as a pure amateur, I don't NEED it (which is one reason I own an Olympus camera).
I also have nothing film related to put me in that mindset, so I've definitely noticed situations where I'd like to be able to go at least to ISO 3200 if not higher.
When photographing birds and animals in tropical forests in countries like Ecuador and Thailand believe me it would be nice to have a camera like the D3 that produces little noise up to 2000 iso or more.
Nate - I don't really think they are very far behind at all, technology-wise. If you compare the D3X to 1Ds MkIII, for instance - a pretty head's up, apples to apples comparison, they are essentially equivalent. The D3X has a bit more resolution so is ever so slightly better for it's pixel size.
It's just that, for whatever reason, Canon hasn't decided to make a low pixel count full-frame sensor to compete with the D3. If they did, I'm willing to bet it would, at worst, be just a small notch below the D3. The D3 (and D3s) aren't magic, as wonderful at high ISO as they are. Low-res, large pixel, large sensors are MUCH easier to make cleaner at high ISO.
As you pointed out, though, Canon also hasn't done anything to make a big leap forward and get you to buy a Mark IV (for noise anyway). But they DID manage to add 5 more MP of resolution into the sensor and STILL improve the pixel level noise - and that's not trivial.
Thank you for your detailed reply I was aware that even if the noise performance was the same as the old 1D MkIII in DXO's test - in practice the Mk1V would be close to 60% better due to the higher resolution. That said I still feel say 12MP on a 1.3 crop (assuming most people buy the Mk1V for action images) is a high enough resolution for for most application and the increase in noise performance would have been a better trade off.
I agree that Canon clearly are capable of making lower res. large pixel sensors with better high iso performance (which as you point out are much easier to make & cheaper) I just wish they would and give you the choice as Nikon do with their users.
Re. the Nikon D3X it's pixel pitch is quite close to that of the Mk1V but has quite a marked all round improvement in sensor performance.
However on a brighter note a few of my friends already own the Mk1V and are delighted with AF performance.
David
That's true, but for purely noise (SNR), they are pretty much equivalent (per pixel, not including resolution). In fact, according to DxO, the Canon actually has better dynamic range at ISO 800 and above, which is pretty impressive given the competition. The Nikon seems tuned for lower ISOs, as it's metrics drop off much quicker than the Canon as the ISO rises.
I'm curious in this case about the wide disparity between DxO's "Low Light ISO" ratings for these cameras. That rating is :
"...the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9EVs and a color depth of 18bits."
But if you look at the component graphs, the 1D MkIV looks equal or better in some categories, and maybe half a stop worse in color depth. I suppose that's the main difference then, and makes their "Low Light ISO" numbers 1992 and 1320. Seems like a really big difference compared to the individual graphs - but it's a pretty complex metric.
I own a D300 and D700.
The D300 is noisy and I don't like the prints as much as I do the D700.
One can talk pixel density, but the fact is on screen the noise between these two cameras is significant. Bostedo is the guru here so I would follow his expertise.
All I can say is I do a lot of low light photography and that was the impetus for the D700. Bright light and low ISO the D300 does fine.
the fisch
Well, thanks...but I think that's a bit overstated. I just like to read a lot and muck about with the technical side of photography. I think it helps that I design computer chips for a living. I wish I had as much practical experience actually using all this pricey equipment as I do at reading data on it.
If you want to learn more about noise, I can highly recommend this article by Emil Martinec, physics professor and avid nature photographer. It's a bit technical, but very readable. He'd qualify as a guru I suspect.