kingslayer: $9000? lol. If this thing was priced around $3500, it would sell very well.
I doubt that the sensor costs $4k unless they get a really pathetic yield. The processed wafer might be $4k perhaps.
It's a really nice looking camera though. Beats the heck out of the mahogany Sony thing they first tried. BTW, how much was that?
But then they couldn't call it a Hasselblad :-)
The statement made in the ISO invariance section: "it minimizes the need to amplify the sensor's signal in order to keep it above that noise floor (which is what ISO amplification conventionally does)." is misleading.
The ISO amplifier amplifies both the signal AND the noise coming from the sensor elements equally. What the so-called "ISO invariant" camera actually does is to yield a constant signal to noise ratio (SNR) in the RAW file regardless of the ISO setting. It is able to do this because the predominant noise source in the signal chain is the sensor pixel noise itself (or at least comes from stages before the ISO gain is set).
In so-called "Non-ISO invariant" designs there is significant system noise coming from stages after the ISO gain stage. This noise becomes dominant at low ISO settings with the result that the SNR is NOT constant across the ISO range.
IMO, a better and more descriptive term for this characteristic might be be "SNR invariant".
Looks like Canon's traditional architecture you can clearly see in frame 9 a four channel Analog Devices AFE/ADC on the back of the Sensor module. Looks like the PN is AD80246. This doesn't show in their catalogue -- must be a custom PN for Canon.
justmeMN: The copper and silver plates are a nice touch. Somebody at Canon is thinking about details that the average consumer will never see.
Shielding to keep their analog circuitry quiet.
Mikael Risedal: It seems that many people have difficulty to understand that D5 is optimised sport camera and to be able to take many images per second and therefore uses a different read out than the other Nikon cameras
Yea, based on Bill's data this new Nikon is just as "bad" as any Canon. It probably won't make a bit of difference to those using it, because it will probably work just as "good" as any Canon. I can't wait to hear some of the Nikon fanboys try to defend this though -- too funny. I would like to see how the Canon equivalent measures up.
Can you (or someone) put up the link to Bill's site? For some reason my old one doesn't work. Thanks in advance.
Never mind, I found it. Wow! I had not seen this. It clearly goes darker but the low ISO read out noise guys aren't going to like it too much. I wonder what they are doing different?
OK thanks. I was having some trouble getting on his site the other day. I'll look again and see if I can see it.
I think it behooves everyone to wait to see some actual measurement data from DxO or Bill Claff on these things to see what is going on. I doubt that Nikon made a bad camera particularly in the D3, D4, D5 "flagship" class. The tests here are not done the same way and it is a bit like comparing apples and oranges in a way. We know what typical DxO DR curves look like for both Canon and Nikon, it will be simple enough compare the new ones with the old ones to see if the changes are significant or not.
Of course, DxO or Bill have to get their hands on the equipment which ain't cheap and I don't think either company is in the habit of sending these guys test samples.
Holy Sh... , OMG, is this Mikael saying DR isn't important in a camera? Woops, looks like my phone is ringing... be right back...
Wow, that was the devil something about having a cold spell down there Hmmm....
Sorry Mikael, just couldn't resist :-)
David Hull: It is not shortening the electronic path or even the integration of the converter that achieves the improvement. It is the fact that they have increased the effective number of bits (ENOB) of the camera by going to entirely different converter approach. The ramp converter approach is very accurate, but very slow so in order to use it they need to use a lot of them and the easiest and least expensive way to achieve this is to put them on the same die with the sensor elements.
Shortening the path does bring benefits but this was never the root cause of the issue people have been whining about. The root cause always was that they put an ADC with an ENOB of 12.5 bits in what they claimed was a 14 bit camera. The reason that they had to do that was that converters that were fast enough to digitize just a few analog lanes from the sensor design they had, were pipeline designs that sacrifice accuracy to achieve speed.
I just picked the most recent camera as an example. We know that they are using QUAD ADC packages (based on one of the tear-downs and their white papers) so it is a matter of taking a guess as to which one it might be.
Without a schematic and BOM, everything is speculation.
I am very confident in the system analysis that I sent you in the PM, take a look. I think it tells the story pretty well.
One problem with going to a 16 bit converter, would be the noise sensitivity at it's input. This would be a situation where the signal lengths would probably be a problem as Rishi alluded in the header.
Probably the best part of the new architecture is that they don't have to worry about shielding and routing as much.
As you point out, there are a number of plausible approaches, the trick is to find one that meets the cost and power budget and still does the job.
David Hull: This seems like a great thing but I wonder what it means for the future of the applications?
I had my editing computer die a couple months ago and I e-mailed NIK to see how they handled things since I was an owner and needed to re-install. I couldn't see where to put the codes in in the latest version. The lady I was in contact with just sent me a link to download -- no codes necessary.
Hopefully they will keep investing in these apps.
Yea, that's what I am afraid of.
newe: Adobe should do this with CC
Yea, right :-). They probably figure that is what they have done at $9.95.
This seems like a great thing but I wonder what it means for the future of the applications?
Rather than "speculate" you can simply "Calculate". Let's look at the 5Ds. This camera has a firing rate of 5 frames per second at 50x10^6 pixels per frame. That means that you are going to have to make 250 million conversions per second therefore needing a digitizer that can support that conversion rate. You decide on a 14 bit converter and you decide to use four quad ADC packages for a total of 16 converters (we know this from the canon white paper). So you need the converters to be capable of 250/16 = 16 MSPS. So you pick something like the Burr Brown (Now TI) ADC3441. This is a 25 MSPS Quad 14 bit ADC that can do the job.
If you pull up that data sheet on the TI site you will see that it has an ENOB = 12 bits (despite being a 14 bit design). If you calculate ENOB = SNR(dB)/6 you get 74/6 = 12.33 ENOB.
So call it "Educated Speculation".
TI DS: http://www.ti.com/lit/ds/sbas671c/sbas671c.pdf
It is not shortening the electronic path or even the integration of the converter that achieves the improvement. It is the fact that they have increased the effective number of bits (ENOB) of the camera by going to entirely different converter approach. The ramp converter approach is very accurate, but very slow so in order to use it they need to use a lot of them and the easiest and least expensive way to achieve this is to put them on the same die with the sensor elements.
David Hull: Wow, is the radio not built into the flash? That seems a bit backward at this point.
That makes sense.
How does this differ, though from what you can already purchase after market?
Yea the external module does have that benefit. I am just surprised that they didn't integrate the whole thing like CaNikon did. I am not sure what this thing brings to the game that the after market guys like Photix are not already doing. However, I am not a Sony user so perhaps there is something I am missing here.
Good to see radio control taking hold though, I love my 600EX RT's.