Jonathan Eastland's M9 review in BJP

[haroldp]

Advertisement (gone after registration)

Bo:

 

You should see what it's like when we disagree.

 

regards ... Harold

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[nhabedi]

Expecting a camera's processor to execute a process in 1/4 sec. that is equivalent to what a much more powerful computer does in 5-10 seconds is not reasonable, and processing shortcuts are always taken.

 

Not sure if this is true for the M8/M9, but as a general statement this is wrong. Your PC at home might be "much more powerful" as a general purpose machine, but things like JPEG compression can be done (and are done) in hardware and that's a lot faster than you might think.

 

E.g., the hardware JPEG engine built into the Maestro chip in the Leica S2 is rated at 160 MP per second which means it can render four S2 files (which are huge) to JPEG in one second! Can your PC do that?

[haroldp]

Not sure if this is true for the M8/M9, but as a general statement this is wrong. Your PC at home might be "much more powerful" as a general purpose machine, but things like JPEG compression can be done (and are done) in hardware and that's a lot faster than you might think.

 

E.g., the hardware JPEG engine built into the Maestro chip in the Leica S2 is rated at 160 MP per second which means it can render four S2 files (which are huge) to JPEG in one second! Can your PC do that?

 

It is not as simple as that.

 

The difference that 'processing in hardware' means, is that the instructions (software) are loaded from silicon (aka firmware), rather than a library on a disk (or silicon). It would never be 'hardwired' because it could then never be updated (except with a soldering iron).

 

It is ultimately an instruction set precessed by a cpu in either case.

 

Might that instruction set (aka software) be more efficient in a 'dedicated' processor engine ? certainly, but it is a difference in implementation, not in 'kind'.

 

Ultimately it is the results that matter, and every manufacturers computer based raw processor produces slightly better results than their best in-camera processing, most likely because they are time-limited in camera and therefore algorithm limited.

 

How slightly better ? varies, and does it matter ?, are highly personal questions.

 

For myself, I go to significant lengths and expense to get very high IQ, and shooting RAW is an easy and free way to get the last increment. It matters to me.

 

160 MP per second ?? doing what ?

 

What does that mean, is that a standard benchmark I never heard of ?.

( ISO reference please ).

 

My computer cannot render any S2 files in a second because I do not have an S2.

 

However my Nikon D3x can render 5 of it's 24 mp files in a second and my computer can do the same in 60 seconds. However it's results are better, and the work is not comparable. If I wanted to dumb down the software I am sure it could render 60, or 600, or 6,000 a second.

 

The only relationship these numbers have to each other is that they are numbers.

 

Apologies to the innocent victims of this crossfire.

 

Regards ... Harold

[nhabedi]

It is not as simple as that.

 

It is.

 

The difference that 'processing in hardware' means, is that the instructions (software) are loaded from silicon (aka firmware), rather than a library on a disk (or silicon). It would never be 'hardwired' because it could then never be updated (except with a soldering iron).

 

Please read again what I wrote: JPEG compression can be done (and is done) in hardware. Period. This is a hardwired algorithm and you can buy ready-made chips which do exactly that. There's also no reason to update the algorithm because the JPEG specification is unlikely to change...

 

However my Nikon D3x can render 5 of it's 24 mp files in a second and my computer can do the same in 60 seconds. However it's results are better, and the work is not comparable. If I wanted to dumb down the software I am sure it could render 60, or 600, or 6,000 a second.

 

So, by "dumbing down" the software a bit you'd be able to speed it up by a factor of 72,000? And still achieve results comparable to what comes out of the D3x? I'm sure there are a lot of companies out there waiting to hire you...

[haroldp]

nhabedi:

 

I am not sure we are really disagreeing on substance.

 

The quality issue is not jpeg compression, that is quite standard, and is very fast on all platforms.

 

Data comes off of the sensor in Raw form.

 

The quality issue and difference is in interpolating bayer patterns and in de-artifacting / aliasing, which is particularly important in Leica's, that do not use an AA filter. It is also very processor intensive. These algorithm's are updated regularly. These are where shortcuts are taken due to time limitations. All of this must be done before jpeg compression can even begin.

 

By definition, dumbing down the code produces inferior results, that is why speed can only be compared when the processing results are identical. The easiest way to make anything go faster is to do less.

 

Many companies have in fact hired me. I am a 35 year software professional and have built, IPO'd and sold software companies. I also hold several patents in the areas of transaction and rules processing algorithm's. You use my software every day.

 

Regards .. Harold

[jl4069]

"Applying some judicious highlight and shadow recovery in Raw processing comes closer to the film's capability in this regard, but of course will not perfectly emulate it (nor would I want it to)."

 

Harold,

Just curious here, what would you ultimately like highlights and shadows to look like? I'm assuming for you digital is quite a ways as well? Do you like the Fuji in the S5?

thanks, Jeremy

[zeitz]

Advertisement (gone after registration)

These reviews serve a great purpose. We out here in reader land would have never figured out that the M9 is expensive on our own. Are the lenses expensive too? He doesn't say and I don't know how to figure that out myself.

[marknorton]

It's certainly the case that dedicated silicon can do much better than a conventional stored program approach but there's no evidence at all that the M9 uses such a thing. That might be one advantage of adopting the Maestro chip set but the M9 is a bit of a slug when it comes to processing if you want both DNG and JPEG. Even more than with the M8, I think of the M9 as a DNG-only camera.

[lars_bergquist]

Gentlemen, re the statement that all DSLR camera make excellent in-camera JPEGs, I would like to draw your attention to page 49 in the current (November) issue of LFI, where JPEGs by both the M8 and the M9 are tested against the Canon 5D Mk II. The M9 is clearly better at this than the M8, and also in some respects better than the Canon. I think that Jonathan Slack's statement in the previous issue stands justified, that with correct exposure, a JPEG from the M9 is not really inferior to a DNG.

 

I would argue for a more even-handed attitude toward JPEG files. Yes, I have been told that real men with hair on their chest do DNG, but I disagree: Real men with hair on their chest make their pictures inside caves. Exposure? WB? Focusing? Composition? To hell with it -- I'll fix it in Photoshop. Wham goes the flint axe.

 

The old man from the Age of Collodion Wet Plates

[nhabedi]

Many companies have in fact hired me. I am a 35 year software professional and have built, IPO'd and sold software companies. I also hold several patents in the areas of transaction and rules processing algorithm's. You use my software every day.

 

OK, to get this out of the way first, I have a Ph.D. in mathematics, I've worked as a software professional for more than twelve years, and I wrote my first programs (in assembler) almost thirty years ago. Furthermore, one of my best friends is a physicist who designs chips with hardwired algorithms for picture and movie processing every day. I think that makes us kind of even...

 

As to the point we're discussing, I think we just have to agree that we disagree. I wasn't discussing the quality of in-camera JPEG creation vs. what PC-based RAW converters can do. I just took issue with your argument that RAW converters must be better because they work significantly longer on the picture although the hardware they're running on is faster ("much more powerful" in your own words). That essentially means a PC doing the same thing the in-camera engine does would be faster than the camera.

 

I stand by my original claim that this simply isn't true. Top-notch digital cameras like the Nikon you're mentioning use special hardware which is optimized for the tasks they're doing and the computation-intensive parts are hardwired. If a PC would be doing the very same thing, the camera's board would run circles around it.

 

PS: And when did you ever hear that a firmware update for a camera included a new algorithm for the JPEG engine? They tweak the user interface or adjust parameters, that's all.

[nhabedi]

It's certainly the case that dedicated silicon can do much better than a conventional stored program approach but there's no evidence at all that the M9 uses such a thing.

 

Right. That's why I initially said that I'm "not sure if this is true for the M8/M9."

[jonoslack]

Gentlemen, re the statement that all DSLR camera make excellent in-camera JPEGs, I would like to draw your attention to page 49 in the current (November) issue of LFI, where JPEGs by both the M8 and the M9 are tested against the Canon 5D Mk II. The M9 is clearly better at this than the M8, and also in some respects better than the Canon. I think that Jonathan Slack's statement in the previous issue stands justified, that with correct exposure, a JPEG from the M9 is not really inferior to a DNG.

 

I would argue for a more even-handed attitude toward JPEG files. Yes, I have been told that real men with hair on their chest do DNG, but I disagree: Real men with hair on their chest make their pictures inside caves. Exposure? WB? Focusing? Composition? To hell with it -- I'll fix it in Photoshop. Wham goes the flint axe.

HI Lars

Quite agree . . and I stand by what I said, however I always shoot DNG, not because the jpgs aren't fine, but because if you use one of the DAM's there's no real processing overhead - and because it's better to change white balance / exposure after the event with RAW than it is with JPG.

[jaapv]

Gentlemen, re the statement that all DSLR camera make excellent in-camera JPEGs, I would like to draw your attention to page 49 in the current (November) issue of LFI, where JPEGs by both the M8 and the M9 are tested against the Canon 5D Mk II. The M9 is clearly better at this than the M8, and also in some respects better than the Canon. I think that Jonathan Slack's statement in the previous issue stands justified, that with correct exposure, a JPEG from the M9 is not really inferior to a DNG.

 

I would argue for a more even-handed attitude toward JPEG files. Yes, I have been told that real men with hair on their chest do DNG, but I disagree: Real men with hair on their chest make their pictures inside caves. Exposure? WB? Focusing? Composition? To hell with it -- I'll fix it in Photoshop. Wham goes the flint axe.

 

The old man from the Age of Collodion Wet Plates

It is, to my mind not a matter of inferiority now, Lars. It is just that I do not see the sense of throwing away information. Information that may be needed in the future to get something better than we have now. And a DNG certainly has more latitude in postprocessing, even at the present day. Btw, real men with hair on their chest need to clean their sensor more often...

[bill]

I thought I'd read all the nonsense I ever needed to read about the M8 until I read this... Holy cr@p. The cover was too slippery for Bill, apparently. Oh my god. Seriously. The cover was too slippery. Seriously. Too slippery. Oh my gawd. I really don't know - this apparently contributed to it being "a doorstop" in Bill's "humble" opinion. (I love that "humble")

 

...and yet another thread is dragged down by a bit of snide bigotry...

 

Y'know, Mani, I thought better of you. Sadly it seems I was mistaken. You know damn well - I said it often enough in the M8 Ghetto - that I had three issues with the M8 that stopped me ever wanting one. 1. crop sensor 2. filtration 3. form factor. The M9 addresses 2 of the three and from my albeit limited handling of one the breeze-block characteristics of the M-Digital shape have been ameliorated by the change in the covering.

 

Now, if you want to have a fit of the vapors like some Edwardian maiden whose corsets are too tight every time someone expresses an opinion that doesn't fit into Maniworld, that's up to you; personally I thought you were a bit more of a grown-up.

 

Ian, on the other hand, I didn't think better of and I have not been disappointed. I do admire your single-mindedness though; it must make your life so much more simple.

 

Regards,

 

Bill

[haroldp]

As to the point we're discussing, I think we just have to agree that we disagree. I wasn't discussing the quality of in-camera JPEG creation vs. what PC-based RAW converters can do. I just took issue with your argument that RAW converters must be better because they work significantly longer on the picture although the hardware they're running on is faster ("much more powerful" in your own words). That essentially means a PC doing the same thing the in-camera engine does would be faster than the camera.

 

I stand by my original claim that this simply isn't true. Top-notch digital cameras like the Nikon you're mentioning use special hardware which is optimized for the tasks they're doing and the computation-intensive parts are hardwired. If a PC would be doing the very same thing, the camera's board would run circles around it.

 

PS: And when did you ever hear that a firmware update for a camera included a new algorithm for the JPEG engine? They tweak the user interface or adjust parameters, that's all.

 

The issue is not whether a camera's dedicated processor is faster than an outboard computer (some of us do not use PC's), it is that the inboard processor is always time limited for practical reasons (some of us like to get the next shot ), and accomodate this by limiting their processing. JPEG engine / conversion has nothing to do with this and can be identical in output on all platforms.

 

Bayer interpolation, noise reduction, de aliasing/artifacting is where the processing time goes.

 

That is why, even with 'as shot' parameters, outboard raw conversion produces better results at the margin, because it can take as long as it needs.

 

These algorithms are updated in firmware regularly. For example early M8 firmware had updates to correct / adjust for the IR problems and I expect M9 upgrades to correct some of the red fringing issues with non retrofocus wides that I am seeing in this forum.

 

Nikon firmware upgrades often include new noise reduction and other image related parameters.

 

The processors in My nikon D3/D3x's are probably faster than in my computers, but they are still time limited to support frame rates.

 

You are correct that the JPEG processing does not change, that is a basic transform, but the processing that feeds the JPEG engine sometimes does.

 

To be clear, the issue with in-camera jpeg has nothing to do with the actual jpeg compression engine, but with the camera's raw processor which feeds it.

 

Regards ... Harold

[haroldp]

It is, to my mind not a matter of inferiority now, Lars. It is just that I do not see the sense of throwing away information. Information that may be needed in the future to get something better than we have now. And a DNG certainly has more latitude in postprocessing, even at the present day. Btw, real men with hair on their chest need to clean their sensor more often...

 

Jaap raises an excellent point, and it is real.

 

For example, Nikon Raw processing software (capture NX2) now corrects for geometric distortion, and both lateral and axial chromatic abberations. This works with older as well as current raw files, so that Raw files I captured several years ago can be re-processed if I want to.

 

Most Raw processors today have some degree of shadow / highlight recovery which they did not have years ago, those who shot Raw can benefit from it.

 

Noise reduction gets smarter with every generation of code.

 

Regards .. Harold

[LondonM]

I said this in the "good enough" thread and I'll say it here. Unless you want jpegs from the camera, just shoot raw and let your computer take care of the rest. The camera will do what it can in firmware, and that will include (fewer and fewer with every generation) shortcuts. But as pointed out, those shortcuts are needed for the moment until processing speed catches up to megapixel count and bit depth. A NVIDIA Tesla C1060 GPU in a Dell computer will produce a better jpeg than any in-camera processing, and will do it much, much faster. Of course that GPU costs $1600 and by the time you configure a computer with the right PCIe v2.0 x16 slots etc, you've spent 8 grand. And something better will come along in a year or two. As long as you shoot raw, or as long as you keep your real film negative, you can always go back and reprocess. As if any of us will.

[nhabedi]

The issue is not whether a camera's dedicated processor is faster than an outboard computer

 

That exactly is the issue because that's how our discussion started.

 

JPEG engine / conversion has nothing to do with this and can be identical in output on all platforms.

 

Bayer interpolation, noise reduction, de aliasing/artifacting is where the processing time goes.

 

Can you give us pointers to articles comparing the computational complexities of these algorithms or do we have to take your word for this? Where does your information come from? Being an expert in transaction and rules processing doesn't necessarily make you an expert in digital image manipulation algorithms.

 

These algorithms are updated in firmware regularly. For example early M8 firmware had updates to correct / adjust for the IR problems and I expect M9 upgrades to correct some of the red fringing issues with non retrofocus wides that I am seeing in this forum.

 

I already said that I don't think the M8/M9 uses custom chips with hardwired algorithms. That the issues you mentioned were addressed doesn't prove anything, though. It could still be possible (and I think it is very likely) that they "just" adjusted parameters while the algorithm itself remained unchanged. Again, do you have any inside information or are you guessing?

[Advertisement]

Hallo,

Look here: Leica M

[blatent liar]

Bo sums this up. "Shaking Head -- deep sigh... " ...on the heels of having owned M8 and M8.2, and whilst still operating my lovely M7, and daily rediscovering my M9, the grip may boil down to preference and perhaps what filth is left on my hands from lunch. Adding a grip to a digital M is best left to accessories, Thumbs up, Luigi case, stock handle, etc... Perhaps Eastland only mentioned this to plug the Thumbs-up (a bit of imbedded advertising). Good gadget either way.

 

What about JPegs? JPegs look like APS when compared to my Raw files swept through Light-room... True the M9 isn't a JPEG camera, suits me fine. My M9 files (to me) look more like Phase One backs. The files from my M9 are truly stellar. And nod to Jono if you can't focus a RF, you can't focus a RF, M9, MP, M7, M8... head shake, deep sigh.

 

-Max