[Eug]

Nikon D3s

Seems like a nice camera, but I wonder if it's CPU limited given some of the specs.

[mattyb]

Interesting that Nikon are going the high ISO route instead of more MP route.

Still, those that want really good video will be disappointed.

The high ISO shots are v. impressive.

[Eug]

Originally Posted by mattyb 

Interesting that Nikon are going the high ISO route instead of more MP route.

You don't need to have low MP to have high ISO. These two things are not mutually exclusive.

I fully expect the Canon 1D replacement for example (Oct. 20?) to have both high ISO and relatively high MP as well.

That's why I'm starting to wonder if Nikon choice of lower MP is partially due to CPU speed constraints.

[tooki]

All else held equal, higher MP = higher noise. That's simple physics. The overall noise levels in DSLRs continues to drop, but otherwise identical sensors of differing resolutions WILL have more noise in the higher-res sensor. Some cameras offer "bucketing" where multiple pixels are grouped to create a single "bucket" to create lower-noise, but lower-res images.

Given Nikon's history of putting exceptionally powerful CPUs in its DSLRs, I can't even begin to imagine that being a bottleneck.

P.S. I'd give up some useless pixels for higher ISO any day.

[Eug]

Originally Posted by tooki 

All else held equal, higher MP = higher noise. That's simple physics. The overall noise levels in DSLRs continues to drop, but otherwise identical sensors of differing resolutions WILL have more noise in the higher-res sensor. Some cameras offer "bucketing" where multiple pixels are grouped to create a single "bucket" to create lower-noise, but lower-res images.

And therein lies the important point. What's important is the final print/image. Higher MP does not increase noise in that context if all things are otherwise equal. However, higher MP can increase detail for cropped images.

Given Nikon's history of putting exceptionally powerful CPUs in its DSLRs, I can't even begin to imagine that being a bottleneck.

I find it very difficult to find the details on Nikon CPUs. Finding details on Canon CPUs is hard enough, but it's even harder for Nikon CPUs. However, I suspect Nikon cannot do 1080p24 video because its CPU is not fast enough to do 1080p24 Motion JPEG, and they haven't added the encoder yet for other formats either. I suspect that one of those two options will come with Nikon's next CPU.

P.S. I'd give up some useless pixels for higher ISO any day.

But that's the point. High ISO and high resolution are not mutually exclusive. It will be interesting to see what the 1D replacement from Canon offers in terms of ISOs, and ISO quality. I suspect they'll try match Nikon (or come very close to matching Nikon) in that regard, but at higher resolution.

[mattyb]

Originally Posted by Eug 

I find it very difficult to find the details on Nikon CPUs. Finding details on Canon CPUs is hard enough, but it's even harder for Nikon CPUs. However, I suspect Nikon cannot do 1080p24 video because its CPU is not fast enough to do 1080p24 Motion JPEG, and they haven't added the encoder yet for other formats either. I suspect that one of those two options will come with Nikon's next CPU.

Got any links for camera cpu stuff?

[Eug]

^^^ Not really, except for Wikipedia's summary for DIGIC (and the links contained there) and what I know about the cameras.

For example, the Canon 5D Mark II is incapable of fast fps despite having Canon's lastest DIGIC 4 CPU. It's a 21 MP camera limited to 3.9 fps. Still this is a remarkable improvement over the older 5D, which is a 12.8 MP camera limited to 3 fps. The 5D used a much older CPU.

So, how is the Canon 7D, an 18 MP camera, getting 8 fps with the same DIGIC 4 CPU? Easy, Canon put two of them in the 7D.

Unfortunately, Nikon is much less forthcoming on details it seems. They do have the D3X with 24.5 MP at 5 fps though. Nikon's fastest camera is 9 fps, but that's only at 12 MP. It's curious to note that their fastest camera is nearly twice as fast as their flagship camera which has twice the MP. This suggests to me that the CPU speed is the bottleneck here. This is evidenced by their last bunch of high end releases:

Nikon D3 - 12.1 MP x 9 fps = 108.9 MP/s
Nikon D3X - 24.5 MP x 5 fps = 122.5 MP/s
Nikon D3s - 12.1 MP x 9 fps = 108.9 MP/s

As you can see, the D3s, the successor to the D3, has the exact same speed and MP. This suggests to me they're probably using the same or a very similar chip in terms of processing capabilities.

With Canon:

Canon 5D Mark II - 21.1 MP x 3.9 fps = 82.3 MP/s (using one DIGIC 4 chip)
Canon 7D - 18.0 MP x 8 fps = 144 MP/s (using two DIGIC 4 chips)

Note however, that even the 5D Mark II can do 1080p30 video despite having a lower MP/s throughput rate than Nikon's cameras. Why? Cuz the DIGIC 4 chip in it has a dedicated built-in H.264 real-time encoder. I'm guessing if Canon had to use Motion JPEG it wouldn't be able to do 1080p30 either. As evidenced by their specs and releases, Nikon does not have dedicated H.264 encoder on their chip. They instead chose to use Motion JPEG, which presumably can be done with the existing chip tech, since the chip is already built for JPEG.

So why doesn't Nikon just simply put two EXCEED chips in their cameras? I dunno. Maybe it's not that simple with how EXCEED is designed, or maybe it's a cost issue. Either way, I'm guessing Nikon's next chip will either include some sort of 1080p real-time video encoder built-in, or will be fast enough to do Motion JPEG at 1080p.

Now, some may argue that 1080p30 or 1080p24 is actually a lower pixel throughput rate than 9 fps at 12 MP. That's true, but one has to remember that the sensor is still 12 MP, and there has to be some sort of pixel binning and/or downsampling algorithm utilized to change 12 MP to 1080p. That's got to take some overhead to do. Some have suggested that pixel skipping can be used, but Canon has already categorically denied using pixel skipping. (Some think doing too much pixel skipping can introduce more aliasing artifacts.)

[mattyb]

I'm not sure what its like on Canon gear, but did you know that you can boost the fps rate with a grip for the D300s for example?

I know much less about cameras than it seems you do, but I don't understand why by the addition of more power in the form of a grip, one can have higher fps if its the cpu that is somehow the limiting factor.

[Eug]

Actually you don't need the grip to get 8 fps on the D300s. The fps limitation is a due to intentional crippling through software. Adding the grip just removes the software block. You can actually get 8 fps through a hack without the grip.

The main point here though is that the max fps is still limited to 12.3 MP x 8 fps = 98.4 MP/s, which is actually slower and a lower pixel throughput rate than the D3s.

[OreoCookie]

Originally Posted by Eug 

And therein lies the important point. What's important is the final print/image. Higher MP does not increase noise in that context if all things are otherwise equal. However, higher MP can increase detail for cropped images.

That depends on what you mean by `all things are otherwise equal.' Noise is directly related to pixel size. So if Canon had used the same sensor technologies it has put into the 7D and made a, say, 12 MP sensor, it would certainly have less noise. That's physics and there is no way around it.

However, one point is that in most situations, even sensor size is moot: if you use normal ISO ranges (100-400), you will notice very little difference in practice. For higher ISOs, crop sensors fare well. To my knowledge, I have never seen a film rated higher than ISO 3200 (perhaps there were some, but even ISO 1600 films were not that easy to get by). So unless you're looking at extremes, noise is not that relevant anyway. Canon decided to trade improvements in resolution to (better) improvements in (very) high ISO behavior while Nikon chose the opposite route. Besides, as a photographer you (not you personally) should know the limits of your equipment and work with it. I'm not limited by the 10 megapixels I have. YMMV, though.

Originally Posted by Eug 

So why doesn't Nikon just simply put two EXCEED chips in their cameras? I dunno.

I guess because it's not necessary. Achieving higher frame rates than ~8 is difficult due to sensor readout speeds, mirror blackout and af performance and one EXCEED chip seems to cope with the camera. Why should Nikon put in another EXCEED chip if the one they put in can cope with the frame rate? You could say for video, but that could be put on a newer version of the EXCEED chip. In any case, the Nikons aren't limited by throughput, but by the lack of a video encoder. Personally, I don't care, I take pictures with my dslr, not videos, but hey, whatever floats your boat.

[Eug]

Originally Posted by OreoCookie 

That depends on what you mean by `all things are otherwise equal.' Noise is directly related to pixel size. So if Canon had used the same sensor technologies it has put into the 7D and made a, say, 12 MP sensor, it would certainly have less noise. That's physics and there is no way around it.

Overall image noise is a function of sensor size, not pixel size per se, which is my main point.

That's why full frame sensors consistently outperform crop sensors of similar generation in terms of noise, regardless of the pixel density. That's also why Canon sensors and Nikon sensors of the same size are usually in the same ballpark in terms of overall image quality.

Is the Canon 7D more noisy than the Nikon D300s for overall picture quality? Maybe, maybe not. Both are in the same ballpark. Yet the Canon 7D has almost 50% more pixels.

So unless you're looking at extremes, noise is not that relevant anyway.

To many photographers, high ISO performance is the holy grail. It should be noted that it's even more important to some people when it comes to video.

In fact, I love the fact that ISO 1600 is very usable on my 7D. It is nowhere near as good on the 20D. Same goes for similar generation Nikons. ISO 1600 is very good with current generation cameras, but wasn't very good with Nikons from 5 years ago.

I guess because it's not necessary. Achieving higher frame rates than ~8 is difficult due to sensor readout speeds, mirror blackout and af performance and one EXCEED chip seems to cope with the camera. Why should Nikon put in another EXCEED chip if the one they put in can cope with the frame rate?

Because it's precisely because it can't cope. Remember, (it seems to me anyway) the only way they were able to put 24.5 MP in their sensor was to decrease the frame rate to 5 fps. ie. It's not even the 8 fps you mention.

I suspect that Nikon's marketing dept. determined that even 8 fps was not necessary for this particular camera, so it felt it didn't need to go the extra mile here, despite the cost of the camera.

Canon has now taken a different marketing approach though, with high resolution and high fps in the same cameras... but has had to make the compromise of utilizing dual CPUs (ie. more $) to achieve that.

I'm sure the next version of Canon's DIGIC and Nikon's EXCEED will be fast enough though. They will both be fast enough to do 8 fps at 20+ MP resolutions in a single chip. And when that happens, I expect to see Nikon's sensors make a significant jump in pixel density.

[OreoCookie]

Originally Posted by Eug 

Overall image noise is a function of sensor size, not pixel size per se, which is my main point.

No, it's pixel size/pixel density: if you crop a full frame sensor to a lower resolution, you get the same noise (trivially). Larger frame sensors allow you to either pack more pixels while keeping the pixel density constant or to use larger pixels.

Originally Posted by Eug 

Is the Canon 7D more noisy than the Nikon D300s for overall picture quality? Maybe, maybe not. Both are in the same ballpark. Yet the Canon 7D has almost 50% more pixels.

I honestly don't care about noise charts or megapixels these days. Of course, for people with special needs, high ISO behavior may still be an important point, but I'd prefer a better 12 megapixel sensor than an 18 megapixel sensor with higher pixel density. In practice, this means you need even better lenses to make use of the additional megapixels.

Again, I reiterate what I've said before: the 7D would be a wonderful camera also for Nikon shooters if it had something like an APS-H-sized sensor. I would like to see full frame sensors to become affordable, i. e. available at the price point of the D300/50D. Sony is trying to do something here by offering the Alpha 850, clearly a crippled Alpha 900, but still, it's a very interesting offering. (Although I'd like a full frame sensor purely for reason that with equivalent focal lengths, I get a smaller depth of field on full frame sensors.)

Originally Posted by Eug 

Because it's precisely because it can't cope. Remember, (it seems to me anyway) the only way they were able to put 24.5 MP in their sensor was to decrease the frame rate to 5 fps. ie. It's not even the 8 fps you mention.

How do you know the frame rate is limited by the cpu speed? And why single out Nikon?
It could just as well be limited by read out speed of the sensor. One indication that this may be the case is that if you compare the frame rates of 20+ megapixel dslrs, they are all at around 4~5 fps -- including Canons (5 fps in case of the EOS 1Ds Mark III, the Sony Alpha 900 and the Nikon D3x, 3.9 fps for the 5D Mark II (and that's with two DIGIC 4 processors)). But honestly I again think this is another non-issue unless you like to argue specs to people over the internet. 10, 15 years ago 3 fps was considered fast among ambitious amateurs.

Originally Posted by Eug 

Canon has now taken a different marketing approach though, with high resolution and high fps in the same cameras... but has had to make the compromise of utilizing dual CPUs (ie. more $) to achieve that.

Even if you take the 7D into the mix, the throughput of the 7D compared to the D3x is only 15 % higher. That's not really a lot. I doubt it even necessitates to add a second processor. However, if it were the sensor, then this means, Canon has managed to build a sensor with a large throughput (which is a good thing™).

Originally Posted by Eug 

I'm sure the next version of Canon's DIGIC and Nikon's EXCEED will be fast enough though. They will both be fast enough to do 8 fps at 20+ MP resolutions in a single chip. And when that happens, I expect to see Nikon's sensors make a significant jump in pixel density.

I doubt it. Nikon has differentiated itself already with a low-res high ISO model and a high-res model. I think you can reasonably expect a D700x or so, but I prefer to have a choice.

[Eug]

Originally Posted by OreoCookie 

No, it's pixel size/pixel density: if you crop a full frame sensor to a lower resolution, you get the same noise (trivially). Larger frame sensors allow you to either pack more pixels while keeping the pixel density constant or to use larger pixels.

However, overall image noise depends on overall sensor size. Remember, basically when you print or resize the image for display, you've effectively pixel-binned the image for real-world image usage.

I honestly don't care about noise charts or megapixels these days.

Judging by your posts, you seem to care very much about noise. A vs B noise charts may not interest you as much, but my whole argument was that effective overall image noise between competing high density and mid-density sensors is similar when the sensor size is the same.

Of course, for people with special needs, high ISO behavior may still be an important point, but I'd prefer a better 12 megapixel sensor than an 18 megapixel sensor with higher pixel density.

I too would prefer a better 12 MP sensor than a lower quality 18 MP sensor. However, if the overall noise is basically the same, I'd be very happy to get the higher MP.

In practice, this means you need even better lenses to make use of the additional megapixels.

Not really. The good lenses from 15 years ago or whatever still easily outresolve high density sensors. So if you had reasonably good lenses to begin with with, you're already good to go.

How do you know the frame rate is limited by the cpu speed?

I don't know for sure, because I don't work at Canon or Nikon. However, given the numbers, that seems like the simplest conclusion. The most obvious example is Canon's. Their highest end CPU is in the Canon 5D II, yet it can "only" achieve 21 MP at 3.9 fps. They wanted to improve on that... so the added a second CPU in the 7D to be able to accomplish it. Suddenly, with the second CPU, the pixel throughput rate almost doubled. Seems like pretty simple assumption to make that pixel throughput was limited by CPU speed here.

And why single out Nikon?

I talked about both Canon and Nikon. However, what I did say is that it seems from the numbers, Nikon's bean counters are not willing to go with the second CPU to speed up frame rates.

One indication that this may be the case is that if you compare the frame rates of 20+ megapixel dslrs, they are all at around 4~5 fps -- including Canons (5 fps in case of the EOS 1Ds Mark III, the Sony Alpha 900 and the Nikon D3x, 3.9 fps for the 5D Mark II (and that's with two DIGIC 4 processors)).

Incorrect. The 5D Mark II has only one DIGIC 4 processor. As for the 1Ds Mark III, it uses an older generation DIGIC chip... and the 1Ds Mark III is ripe for an update. Some say the 1D Mark III will get an update first though, possibly as early as next week.

But honestly I again think this is another non-issue unless you like to argue specs to people over the internet. 10, 15 years ago 3 fps was considered fast among ambitious amateurs.

No, 10 years ago I thought 3 fps was slow. I just couldn't afford anything else (in terms of the camera cost and the film cost).

P.S. 10 years ago I was also using a Pentium II class CPU, too. We've moved on since then.

Even if you take the 7D into the mix, the throughput of the 7D compared to the D3x is only 15 % higher. That's not really a lot. I doubt it even necessitates to add a second processor. However, if it were the sensor, then this means, Canon has managed to build a sensor with a large throughput (which is a good thing™).

Canon says in the marketing blurbs that they have two DIGIC 4 CPUs for speed. Not that it's surprising:

Dual DIGIC 4 Image Processors ensure that images are captured, processed and saved with remarkable speed. The EOS 7D's ability to capture and process data of images shot at 18.0 Megapixels at 8 fps, as well as Face Detection Live Mode, Full HD video recording, Auto Lighting Optimizer and Lens Peripheral optimization are all possible thanks to the Dual DIGIC 4 Image Processors.

I doubt it. Nikon has differentiated itself already with a low-res high ISO model and a high-res model. I think you can reasonably expect a D700x or so, but I prefer to have a choice.

I betcha this is a combination of marketing decisions and practical CPU limitations.

[OreoCookie]

Originally Posted by Eug 

Judging by your posts, you seem to care very much about noise. A vs B noise charts may not interest you as much, but my whole argument was that effective overall image noise between competing high density and mid-density sensors is similar when the sensor size is the same.

I know about noise, yes. And since I'm a scientist, I care about numbers. So I can and might start arguing numbers. However, for me personally, because (I think*) I know the numbers and the way I take pictures, I can say that personally my camera's performance is good enough for me.

* Nobody is perfect

Originally Posted by Eug 

Suddenly, with the second CPU, the pixel throughput rate almost doubled. Seems like pretty simple assumption to make that pixel throughput was limited by CPU speed here.

They also added a completely new sensor. All I'm saying is that we don't know what is limiting the performance. It could be sensor read out rather than throughput in the `development' stage. See for example in this older review of the Canon 350D (emphasis mine):

The 20D sensor can read data much faster, allowing it to deliver a 5fps continuous shooting mode, whereas the 350D sensor's slower throughput limits it to 3fps. This is still an improvement over the 300D's continuous shooting speed of 2.5fps though, and the buffer has also been increased considerably - see later.

I think this is particularly relevant for cameras with cmos sensors where the AD converters are built right onto the chip. You can find similar hints in Sony's marketing speak on the 24 megapixel sensors (although I don't know why the initial 6.3 fps haven't been reached; perhaps that was a theoretical RAW throughput figure and not the realistic effective throughput).

All I'm saying is that you don't know whether in case of the D3x (or 1Ds Mark III or Alpha 900) the speed is limited by the cpu or by the sensor. And judging from the numbers, it seems that Nikon's EXPEED processors are faster (on a per chip basis) than Canon's DIGIC4. I conjecture that the 7D's which was designed for speed necessitated the use of a second processor, because one DIGIC4 processor couldn't keep up with the throughput of the newly designed sensor. In my opinion, especially given the price tag of the D3x (or other top-end models), I don't think no company needs to limit the performance artificially. Both, Canon, Nikon and Sony would be happy to offer their most expensive camera (e. g. Canon's EOS 1Ds Mark III or Nikon's D3x) with all the features it could get. I think throughput in those cases is limited by sensor throughput of the sensor rather than processing power.

There are marchitecture cameras (e. g. the 5D Mark II, the 7D, low-end Nikons without built-in focus motors or Sony's Alpha 850) or cameras that won't be built because of marchitecture -- right away, at least (the much-rumored D700x comes to mind).

Originally Posted by Eug 

Incorrect. The 5D Mark II has only one DIGIC 4 processor. As for the 1Ds Mark III, it uses an older generation DIGIC chip... and the 1Ds Mark III is ripe for an update. Some say the 1D Mark III will get an update first though, possibly as early as next week.

I stand corrected.

Originally Posted by Eug 

No, 10 years ago I thought 3 fps was slow. I just couldn't afford anything else (in terms of the camera cost and the film cost).

… which is why I've added `ambitious amateur' as in double-digit Canon bodies (EOS 50 or EOS 30), Nikon F80 or similar bodies. Even the cheapest bodies these days can match their performance in terms of AF system and raw fps. When I switched from 1 fps to 2.5 or 3 (not sure anymore), I found it plenty fast. Especially when all you have are 36 exposures.

Of course, you're absolutely right that technological progress allows us to get more performance for the price. But such as with many things, there is only limited return: if you increase the fps from 3 to 4.5 (as in the case of the D80 --> D90), many people will be happy to read this. And in quite a few cases, this is an incentive for users to upgrade or for a buyer to choose this camera over its competition, but in most cases, it's just like quoting the power output of your car engine. I rarely switch to continuous shoot mode and I don't think I've ever wanted more than 3 fps (when I shoot some motion or a group of people, I switch to continuous mode to be able to pick the best shot afterwards). But again, that's just me. Perhaps there are others (who like to take pictures of skateboarders or whatnot) who benefit from increased fps. And sure, I won't object if my next camera is faster


To be honest (on a side note), I'm still playing with the idea of getting an analog body to get the slowness back into my workflow (plus, a smaller DOF if I use equivalent focal lengths). I've noticed that very often, I put in more effort into my pictures, knowing that I only have, say, 7 rolls à 36~38 exposures with me on a vacation. For similar reasons, I think of getting a 35~50 mm equivalent prime: I want to limit myself to enhance my creativity and learn how to use my equipment better.

[tooki]

Back to the topic of MP-vs-ISO: as a photographer, high-ISO performance expands the creative performance of the camera in ways more MP never do.

Yes, if I have to crop, the extra MP can be useful. Or I can just frame the shot better to begin with.
(I am a tad of a photo purist, in that I prefer to just take the photo right to begin with, rather than relying on the crutch of post.)

In any event, having a camera with usable ISO 3200 or higher would significantly expand the situations in which I can shoot in ambient light without a tripod. That I’d pay money for. That I will pay money for when it’s time for a new camera.