7D mark 2 crop vs full frame

[flyingSquirrel]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

 

[dgatwood]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

You're crazy.



Sorry, couldn't resist.

 

[Big Tex]

It depends on what you shoot. I shoot a lot of HS sports, and love using my 70-200 2.8. With the crop it gives me beyond 300MM in reach. I am much more mobile than the guys I see using full-framers and the big 300 & 400MM lenses. I can move all over the field and get down all the way to the turf if I want.

Can't wait for the new 7D Mark ii. Am hoping to get great noise improvement at 6400 ISO.

 

[flyingSquirrel]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

You're crazy.



Sorry, couldn't resist.

Hehe, no problem, I fully expected someone to take that one. Besides, I like it when people say I'm crazy. It lets me know I'm doing something right

 

[Mt Spokane Photography]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Since it will have likely have dual pixel technology, you could claim that its 50mp, but the dual pixel output is combined before saving to a memory card. There is a possibility that those dual pixels will do more than autofocus, like improving DR, but the specs will be in the low 20 mp range.

 

[Don Haines]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Since it will have likely have dual pixel technology, you could claim that its 50mp, but the dual pixel output is combined before saving to a memory card. There is a possibility that those dual pixels will do more than autofocus, like improving DR, but the specs will be in the low 20 mp range.

I'd argue with you about the amount of data to process being too much....

Look at the 60D... it has a DIGIC4 and saves to SD cards. We now have DIGIC6 which has 20 times the processing power.... When the 60D came out with it's SDHC cards the maximum transfer speed was 25Mbytes per second and now we have SDXC-UHS2 cards which support up to 2Tbytes of storage and up to 312Mbytes per second storage.. this improvement in technology is easily enough to double the frame rate plus double the pixel count of a 60D so handling a 36Mpixel sensor at 10 frames per second is not a hard task.

 

[Mt Spokane Photography]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Since it will have likely have dual pixel technology, you could claim that its 50mp, but the dual pixel output is combined before saving to a memory card. There is a possibility that those dual pixels will do more than autofocus, like improving DR, but the specs will be in the low 20 mp range.

I'd argue with you about the amount of data to process being too much....

Look at the 60D... it has a DIGIC4 and saves to SD cards. We now have DIGIC6 which has 20 times the processing power.... When the 60D came out with it's SDHC cards the maximum transfer speed was 25Mbytes per second and now we have SDXC-UHS2 cards which support up to 2Tbytes of storage and up to 312Mbytes per second storage.. this improvement in technology is easily enough to double the frame rate plus double the pixel count of a 60D so handling a 36Mpixel sensor at 10 frames per second is not a hard task.

Don,
My comment had to do with the internal readout of the sensor and the processor. That's why the 7D and 1D series use dual processors, to get the data flow that's needed. A 50MP raw image before it is compressed by the processor will be well over 200MB. Jpeg images are very highly compressed, but the processor must work very hard to compress a 200MB+ image to jpeg size. After that, write speeds are not the issue.

SDXC UHS cards are still limited to slow write speeds after their first use and a normal in camera format like most users do. Some are misled by reading specifications, they are for new cards only, or for those that have undergone the slow process of erasing. Just doing a regular format in camera does not erase a card, so a SD card must erase a whole block of used sectors before they can be written to. That's why they say "up to" in the spec.

Hopefully, a 7D MK II would have dual CF slots, which will not be a write speed limitation.

 

[Don Haines]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Since it will have likely have dual pixel technology, you could claim that its 50mp, but the dual pixel output is combined before saving to a memory card. There is a possibility that those dual pixels will do more than autofocus, like improving DR, but the specs will be in the low 20 mp range.

I'd argue with you about the amount of data to process being too much....

Look at the 60D... it has a DIGIC4 and saves to SD cards. We now have DIGIC6 which has 20 times the processing power.... When the 60D came out with it's SDHC cards the maximum transfer speed was 25Mbytes per second and now we have SDXC-UHS2 cards which support up to 2Tbytes of storage and up to 312Mbytes per second storage.. this improvement in technology is easily enough to double the frame rate plus double the pixel count of a 60D so handling a 36Mpixel sensor at 10 frames per second is not a hard task.

Don,
My comment had to do with the internal readout of the sensor and the processor. That's why the 7D and 1D series use dual processors, to get the data flow that's needed. A 50MP raw image before it is compressed by the processor will be well over 200MB. Jpeg images are very highly compressed, but the processor must work very hard to compress a 200MB+ image to jpeg size. After that, write speeds are not the issue.

SDXC UHS cards are still limited to slow write speeds after their first use and a normal in camera format like most users do. Some are misled by reading specifications, they are for new cards only, or for those that have undergone the slow process of erasing. Just doing a regular format in camera does not erase a card, so a SD card must erase a whole block of used sectors before they can be written to. That's why they say "up to" in the spec.

Hopefully, a 7D MK II would have dual CF slots, which will not be a write speed limitation.

I think that if they wanted to, that they could read out a large pixel count sensor more than 10 times per second.... and yes, dual DIGICs really helps a lot. After all, I have a 60Ghz spectrum analyzer sitting on the bench next to my desk.... you can design some insanely fast electronics if you really want to.

Although most of us think in terms of bytes, I/O knows no such bounds. Reading a sensor is a massively parallel operation and instead of reading one byte or word at a time, they can be designed to be read hundreds of bits at a time, and even rows at a time into an internal buffer on the sensor. ultimately it comes down to design decisions and architecture.

However we slice it, even with dual cFast cards, the real bottleneck is going to be dumping it to storage. A good buffer can help a lot there, but ultimately, the buffer fills and things slow down.

 

[AprilForever]

So many reasons it needs to be a crop sensor...

 

[bdunbar79]

I think we ALL know deep down that it is NOT going to be either FF or APS-H.

 

[dgatwood]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Only because Canon's custom silicon is abysmally slow. If the specs from Magic Lantern are correct (and if I read the spec sheets correctly), then the top-of-the-line, current-generation 64-bit ARM chips are at least 200x as fast as the DIGIC 5, and maybe an order of magnitude more than that. (I don't know which version of the CPU they based their chip on.) All their custom DSP hardware is just working around the CPU itself being a total dog.

The FPS of DSLRs is not primarily limited by the CPU. It is primarily limited by the speed of the mechanical shutter and secondarily by the size of the buffer and the speed at which the flash parts can write the data usefully. The CPU speed is almost certainly chosen to be just fast enough to handle the needed throughput.

My comment had to do with the internal readout of the sensor and the processor. That's why the 7D and 1D series use dual processors, to get the data flow that's needed.

I suspect they use dual processors because it's cheaper to add more existing, slower parts than it is to design newer, faster ones.

A 50MP raw image before it is compressed by the processor will be well over 200MB. Jpeg images are very highly compressed, but the processor must work very hard to compress a 200MB+ image to jpeg size. After that, write speeds are not the issue.

The big headache with data that big is actually the power consumption of the RAM needed to store it. JPEG processing is borderline trivial, CPU-wise, and is almost infinitely parallelizable. You could throw a separate CPU core at each DCT block if you were so inclined, or even even parallelize the DCT itself (though I don't remember precisely how you do the latter without duplicating parts of the computation). Even if you're doing scaling to obtain a lower quality resolution, it's still trivially parallelizable. Just throw a decent 64-bit, 4-core ARM CPU at the problem, and you're likely in the ballpark.

But write speeds are going to be a big part of the problem. Lots of people shoot RAW, and a 200 MB file size represents nearly a factor of ten increase. And JPEG, assuming the quality setting remains the same, and assuming you're shooting at full resolution, is likely to also increase in size roughly proportional to the number of pixels, so that's going to balloon by close to a factor of ten as well.

 

[jrista]

I posted this in another 7D mk II thread but wanted to mention it here as well. I have decided that I would be willing to bet money (if I had any money) that this camera will have 30+ MP, and that, in Q1 of 2015, a camera will be released from Canon which will have 40-50 MP. Call me crazy.

It might go 23-24mp, anything higher in a APS-C would cause big technical issues trying to achieve a 10 fps speed. The amount of data to process is too much.

Only because Canon's custom silicon is abysmally slow. If the specs from Magic Lantern are correct (and if I read the spec sheets correctly), then the top-of-the-line, current-generation 64-bit ARM chips are at least 200x as fast as the DIGIC 5, and maybe an order of magnitude more than that. (I don't know which version of the CPU they based their chip on.) All their custom DSP hardware is just working around the CPU itself being a total dog.

The FPS of DSLRs is not primarily limited by the CPU. It is primarily limited by the speed of the mechanical shutter and secondarily by the size of the buffer and the speed at which the flash parts can write the data usefully. The CPU speed is almost certainly chosen to be just fast enough to handle the needed throughput.

My comment had to do with the internal readout of the sensor and the processor. That's why the 7D and 1D series use dual processors, to get the data flow that's needed.

I suspect they use dual processors because it's cheaper to add more existing, slower parts than it is to design newer, faster ones.

A 50MP raw image before it is compressed by the processor will be well over 200MB. Jpeg images are very highly compressed, but the processor must work very hard to compress a 200MB+ image to jpeg size. After that, write speeds are not the issue.

The big headache with data that big is actually the power consumption of the RAM needed to store it. JPEG processing is borderline trivial, CPU-wise, and is almost infinitely parallelizable. You could throw a separate CPU core at each DCT block if you were so inclined, or even even parallelize the DCT itself (though I don't remember precisely how you do the latter without duplicating parts of the computation). Even if you're doing scaling to obtain a lower quality resolution, it's still trivially parallelizable. Just throw a decent 64-bit, 4-core ARM CPU at the problem, and you're likely in the ballpark.

But write speeds are going to be a big part of the problem. Lots of people shoot RAW, and a 200 MB file size represents nearly a factor of ten increase. And JPEG, assuming the quality setting remains the same, and assuming you're shooting at full resolution, is likely to also increase in size roughly proportional to the number of pixels, so that's going to balloon by close to a factor of ten as well.

Your comparing general-purpose processors to special-purpose DSPs designed to handle, in hardware, the specific processing needs to a specific camera, or small set of cameras. The two, a general-purpose ARM and a DIGIC DSP, are NOT directly comparable. The clock rate of a DIGIC may be "abysmally slow", however it's IPC is extremely high compared to the ARM. The ARM may have a ridiculously high clock rate, but it's IPC is very low. Canon's DSP's aren't any slower really than the competitions. DIGIC 5/5+ is already a bit older now, DIGIC 6 is really what should be compared to the competition. In that respect, DIGIC 6 is on par with Sony BIONZ X (the DSP used in their A7s), as they both do similar processing, all in hardware (although Canon's is currently only used for some of their compact cameras.)

Regarding memory, I wouldn't say that it's just the memory that consumes power...because the IPC of the DIGIC chips is high, they ARE doing a LOT of work, regardless of the clock rate. Despite that, the primary power consumer is unlikely to be either the memory nor the DSP. Moving physical components requires more power...flapping a mirror @ 12fps, moving large focus groups in lenses, those are going to consume more power. If your shooting action, those things are going to consume a lot more power. With tiny transistors these days, its easy to build low-power electronics...but the force required to move a physical object will always be the same.

I don't think "throwing an ARM at the problem" is a solution. The IPC of an arm is low, they are GENERAL purpose processors, so they will require far more cycles to perform the kind of image processing necessary to handle the information coming off the sensor. A specially-designed DSP that has the necessary logic built into the hardware will perform image processing a lot faster for less power, as it's a SPECIAL purpose device. That's why we have GPUs in our computers...they are specially designed to tackle the problem of pixel processing in a more efficient manner than a CPU ever could.

 

[mrsfotografie]

It will not be a mark 2 of the original 7D if it wasn't an APS-C sensor.

True.

I personally don't care much about APS-C DSLR's anymore. When I get the chance I will get a full frame mirrorless too (at a budget price), but then, full frame is compatible with my type of photography.

Of course for the birders and such I wish the 7D2 will have an excellent aps-c sensor

 

[TexPhoto]

It will not be a mark 2 of the original 7D if it wasn't an APS-C sensor.

True.

I personally don't care much about APS-C DSLR's anymore. When I get the chance I will get a full frame mirrorless too (at a budget price), but then, full frame is compatible with my type of photography.

Of course for the birders and such I wish the 7D2 will have an excellent aps-c sensor

If you shoot a variety of images, or even just a variety in one nitch, an APS-C camera is a nice compliment to a FF camera. Shooting sports I will often use a fisheye and my 400 with extenders in the same shoot. The difference in the look and reach of the 2 cameras can be nice. The 5DII + 7D combo because very popular for this reason.

 

[mrsfotografie]

It will not be a mark 2 of the original 7D if it wasn't an APS-C sensor.

True.

I personally don't care much about APS-C DSLR's anymore. When I get the chance I will get a full frame mirrorless too (at a budget price), but then, full frame is compatible with my type of photography.

Of course for the birders and such I wish the 7D2 will have an excellent aps-c sensor

If you shoot a variety of images, or even just a variety in one nitch, an APS-C camera is a nice compliment to a FF camera. Shooting sports I will often use a fisheye and my 400 with extenders in the same shoot. The difference in the look and reach of the 2 cameras can be nice. The 5DII + 7D combo because very popular for this reason.

I worked with that combination for a long time, but the image quality of the 7D was disappointing compared to the 5D MkII so in the end that's what caused me to sell the 7D in favor of a 5D MkIII. The Mk II and Mk III work much better together; but I prioritize the MkIII for low light and sports, obviously.

Last weekend I did a motorsports shoot with the 70-200 on the MkIII and my 35mm Sigma on the Mk II. The results were staggeringly good, and the bodies and lenses a seamless match, with matching sharpness and colors.

 

[dgatwood]

Your comparing general-purpose processors to special-purpose DSPs designed to handle, in hardware, the specific processing needs to a specific camera, or small set of cameras. The two, a general-purpose ARM and a DIGIC DSP, are NOT directly comparable. The clock rate of a DIGIC may be "abysmally slow", however it's IPC is extremely high compared to the ARM.

I'm area of the difference between normal CPUs and dedicated DSPs. I'm also aware that modern 64-bit ARM CPUs have vector engines and graphic chips that are fast enough to almost certainly make that DSP hardware completely unnecessary. The reason you use DSPs is because the CPUs can't handle the processing. The CPU in an iPhone 5S is faster than a single-processor 2 GHz G5 Mac from just a few years ago. And Canon RAW image rendering seems to be pretty close to instant on an iPhone 5 using just the CPU, as far as I can tell from my Safari experiments, so I would expect that a CPU comparable to the one in the 5S ought to be able to handle a DSLR's image processing without breaking a sweat. Granted, converting to JPEG takes extra work, but not that much extra work.

You should be able to get by with a small amount of dedicated hardware to control the ADC sweep across the CMOS part and shove the data into a small chunk of dual-port RAM so the CPU can then copy it into normal RAM using NEON instructions. Mind you, I could be wrong—I'd have to actually write the code before I could say with absolute certainty—but I'm pretty sure we're either past the point where those DSPs are unnecessary or at least rapidly approaching it.

BTW, I'm not sure what you mean by "IPC". To me, that means interprocessor communication, which isn't relevant here. Do you mean IOPS?

Regarding memory, I wouldn't say that it's just the memory that consumes power...because the IPC of the DIGIC chips is high, they ARE doing a LOT of work, regardless of the clock rate. Despite that, the primary power consumer is unlikely to be either the memory nor the DSP. Moving physical components requires more power...flapping a mirror @ 12fps, moving large focus groups in lenses, those are going to consume more power. If your shooting action, those things are going to consume a lot more power. With tiny transistors these days, its easy to build low-power electronics...but the force required to move a physical object will always be the same.

35mm cameras used to run for months on a tiny button cell. So I'd expect that, compared with the CPUs, LCD panels, RAM, the mechanical bits should pretty much be lost in the noise, power-wise (though that may not be true with mirror lock-up—not sure). Then again, people took fewer shots in those days, so maybe that's not a fair comparison.

I don't think "throwing an ARM at the problem" is a solution. The IPC of an arm is low, they are GENERAL purpose processors, so they will require far more cycles to perform the kind of image processing necessary to handle the information coming off the sensor. A specially-designed DSP that has the necessary logic built into the hardware will perform image processing a lot faster for less power, as it's a SPECIAL purpose device. That's why we have GPUs in our computers...they are specially designed to tackle the problem of pixel processing in a more efficient manner than a CPU ever could.

It doesn't really matter how many cycles the processing takes. What matters is the clock time and, to a lesser extent, the power consumption. If the general-purpose CPUs can handle the processing in the required time, it makes a lot more sense to use those rather than custom DSP hardware, because in the downtime between photos, you can repurpose that extra CPU power for other useful tasks, unlike DSP hardware, which is pretty much a one-trick pony.

 

[Don Haines]

Your comparing general-purpose processors to special-purpose DSPs designed to handle, in hardware, the specific processing needs to a specific camera, or small set of cameras. The two, a general-purpose ARM and a DIGIC DSP, are NOT directly comparable. The clock rate of a DIGIC may be "abysmally slow", however it's IPC is extremely high compared to the ARM.

I'm area of the difference between normal CPUs and dedicated DSPs. I'm also aware that modern 64-bit ARM CPUs have vector engines and graphic chips that are fast enough to almost certainly make that DSP hardware completely unnecessary. The reason you use DSPs is because the CPUs can't handle the processing. The CPU in an iPhone 5S is faster than a single-processor 2 GHz G5 Mac from just a few years ago. And Canon RAW image rendering seems to be pretty close to instant on an iPhone 5 using just the CPU, as far as I can tell from my Safari experiments, so I would expect that a CPU comparable to the one in the 5S ought to be able to handle a DSLR's image processing without breaking a sweat. Granted, converting to JPEG takes extra work, but not that much extra work.

You should be able to get by with a small amount of dedicated hardware to control the ADC sweep across the CMOS part and shove the data into a small chunk of dual-port RAM so the CPU can then copy it into normal RAM using NEON instructions. Mind you, I could be wrong—I'd have to actually write the code before I could say with absolute certainty—but I'm pretty sure we're either past the point where those DSPs are unnecessary or at least rapidly approaching it.

BTW, I'm not sure what you mean by "IPC". To me, that means interprocessor communication, which isn't relevant here. Do you mean IOPS?

Regarding memory, I wouldn't say that it's just the memory that consumes power...because the IPC of the DIGIC chips is high, they ARE doing a LOT of work, regardless of the clock rate. Despite that, the primary power consumer is unlikely to be either the memory nor the DSP. Moving physical components requires more power...flapping a mirror @ 12fps, moving large focus groups in lenses, those are going to consume more power. If your shooting action, those things are going to consume a lot more power. With tiny transistors these days, its easy to build low-power electronics...but the force required to move a physical object will always be the same.

35mm cameras used to run for months on a tiny button cell. So I'd expect that, compared with the CPUs, LCD panels, RAM, the mechanical bits should pretty much be lost in the noise, power-wise (though that may not be true with mirror lock-up—not sure). Then again, people took fewer shots in those days, so maybe that's not a fair comparison.

I don't think "throwing an ARM at the problem" is a solution. The IPC of an arm is low, they are GENERAL purpose processors, so they will require far more cycles to perform the kind of image processing necessary to handle the information coming off the sensor. A specially-designed DSP that has the necessary logic built into the hardware will perform image processing a lot faster for less power, as it's a SPECIAL purpose device. That's why we have GPUs in our computers...they are specially designed to tackle the problem of pixel processing in a more efficient manner than a CPU ever could.

It doesn't really matter how many cycles the processing takes. What matters is the clock time and, to a lesser extent, the power consumption. If the general-purpose CPUs can handle the processing in the required time, it makes a lot more sense to use those rather than custom DSP hardware, because in the downtime between photos, you can repurpose that extra CPU power for other useful tasks, unlike DSP hardware, which is pretty much a one-trick pony.

Gentlemen, perhaps the solution is not A or B, but A and B....

 

[dgatwood]

Gentlemen, perhaps the solution is not A or B, but A and B....

True. And after thinking about it further, I've concluded that they probably will still need some DSP hardware for video scaling and compression, if only for power consumption and heat reasons.

 

[Don Haines]

Gentlemen, perhaps the solution is not A or B, but A and B....

True. And after thinking about it further, I've concluded that they probably will still need some DSP hardware for video scaling and compression, if only for power consumption and heat reasons.

Some things are easier to do in hardware and some are easier to do in software. Almost all of the test equipment I use is a hybrid of the two, and some of it runs at insane speeds like satellite modems with 50Mhz wide bandwidth and spectrum analyzers that can run at 60Ghz... It makes sense to do some of the work in hardware if it means reducing heat and thereby improving battery life and noise....

 

[jrista]

Your comparing general-purpose processors to special-purpose DSPs designed to handle, in hardware, the specific processing needs to a specific camera, or small set of cameras. The two, a general-purpose ARM and a DIGIC DSP, are NOT directly comparable. The clock rate of a DIGIC may be "abysmally slow", however it's IPC is extremely high compared to the ARM.

I'm area of the difference between normal CPUs and dedicated DSPs. I'm also aware that modern 64-bit ARM CPUs have vector engines and graphic chips that are fast enough to almost certainly make that DSP hardware completely unnecessary. The reason you use DSPs is because the CPUs can't handle the processing. The CPU in an iPhone 5S is faster than a single-processor 2 GHz G5 Mac from just a few years ago. And Canon RAW image rendering seems to be pretty close to instant on an iPhone 5 using just the CPU, as far as I can tell from my Safari experiments, so I would expect that a CPU comparable to the one in the 5S ought to be able to handle a DSLR's image processing without breaking a sweat. Granted, converting to JPEG takes extra work, but not that much extra work.

You should be able to get by with a small amount of dedicated hardware to control the ADC sweep across the CMOS part and shove the data into a small chunk of dual-port RAM so the CPU can then copy it into normal RAM using NEON instructions. Mind you, I could be wrong—I'd have to actually write the code before I could say with absolute certainty—but I'm pretty sure we're either past the point where those DSPs are unnecessary or at least rapidly approaching it.

BTW, I'm not sure what you mean by "IPC". To me, that means interprocessor communication, which isn't relevant here. Do you mean IOPS?

IPC = Instructions Per Clock. It's a well-known term when describing how much work any kind of processor does in one clock cycle. Longer, more complex pipelines (such as in a purpose-built DSP) usually have much higher IPC. So, while they often have a lower clock rate, they do just as much if not more work than a general purpose device.

I don't think we'll see DSPs disappear from DSLRs any time soon. Purpose-built processors allow manufacturers to very finely and precisely tune the processor to the capabilities of the camera, optimize power usage, etc. If anyone was going to jump the DSP ship first, it would have been Sony, however instead, they built the Bionz X DSP for the A7s, and it is pretty kick-ass.

Regarding memory, I wouldn't say that it's just the memory that consumes power...because the IPC of the DIGIC chips is high, they ARE doing a LOT of work, regardless of the clock rate. Despite that, the primary power consumer is unlikely to be either the memory nor the DSP. Moving physical components requires more power...flapping a mirror @ 12fps, moving large focus groups in lenses, those are going to consume more power. If your shooting action, those things are going to consume a lot more power. With tiny transistors these days, its easy to build low-power electronics...but the force required to move a physical object will always be the same.

35mm cameras used to run for months on a tiny button cell. So I'd expect that, compared with the CPUs, LCD panels, RAM, the mechanical bits should pretty much be lost in the noise, power-wise (though that may not be true with mirror lock-up—not sure). Then again, people took fewer shots in those days, so maybe that's not a fair comparison.

I don't think any film SLR cameras were ever cranking out 12fps, nor moving the lens focus group of the great whites like the 300mm through 600mm lenses that frequently. The 1D X requires a higher voltage to supply the necessary power to the lens to support fast AF. Moving the mirror that fast, reliably and accurately on a consistent basis takes the right kind of power/signal.

I also don't think that a large pro-grade film SLR was running off of a button battery. IIRC, Canon has a specific NiMH battery used in the 1V and EOS 3 cameras...last I saw, it looked pretty similar to the 1D X batteries in terms of shape and size.

I don't think "throwing an ARM at the problem" is a solution. The IPC of an arm is low, they are GENERAL purpose processors, so they will require far more cycles to perform the kind of image processing necessary to handle the information coming off the sensor. A specially-designed DSP that has the necessary logic built into the hardware will perform image processing a lot faster for less power, as it's a SPECIAL purpose device. That's why we have GPUs in our computers...they are specially designed to tackle the problem of pixel processing in a more efficient manner than a CPU ever could.

It doesn't really matter how many cycles the processing takes. What matters is the clock time and, to a lesser extent, the power consumption. If the general-purpose CPUs can handle the processing in the required time, it makes a lot more sense to use those rather than custom DSP hardware, because in the downtime between photos, you can repurpose that extra CPU power for other useful tasks, unlike DSP hardware, which is pretty much a one-trick pony.

What matters is how much work is done in any given unit time. That's what affects power consumption. That's where IPC comes into play. A high IPC/low clock part can do the same amount or more work (same amount of power consumption) as a low IPC/high clock part.

As for repurposing extra processing power for other "useful tasks"...what useful tasks? Were talking about cameras here, not smartphones. I am happy that my immensely capable smartphone has a camera, but it's also basically a general purpose pocket PC. It's a small, mobile computer. My DSLR is just a camera...it only really has one specific task. I'm not going to be doing image editing on it's microscopic 3.2" screen, I'm not going to be dialing up my buddies or playing games or listening to music out in the field. I bought a DSLR so I would have a very powerful, very capable camera with a high frame rate, highly accurate focus, and the ability to use a wide variety of lenses. That's it's purpose. It's a SPECIAL purpose. I don't know what I'd use a whole lot of extra processing horsepower for in a camera... I use my Lumia for any general purpose tasks, as it is far better suited to it.