dr_jon

Is this really the issue though, or does the CPU (presumably a SoC in reality) never get so unhappy it needs a heatsink (and they can run at 70C case temperature, 100C internally, probably way more than assorted other components in the camera body). (BTW 45C and above is not something you want to touch.) As there are YouTube videos showing you can record 4k HQ for 4 hours to an external recorder provided no cards are in the camera that suggests the issue is: (1) The super-whizzy cards get too hot and aren't rated to record reliably when hot, so the camera stops trying to record (2) The SoC gets a lot hotter when doing H.265 compression (3) Both Someone needs to record 4k HQ to a fast SD UHS II card with the other slot empty, IMHO...

I don't know about 2.8k upscaled but I have seen "3480 to 3840" upscaling mentioned, which is a 10% stretch. Also the maths works great, as it's every other pixel. So I suspect they read all the pixels (as hard not to along a line) and scale them, they could read every other line but that would break bayer, so possibly the approach is that they read the whole (16:9 crop of the) sensor and just do the simplest scaling option of using 2x2 bayer blocks to make pixels and upscaling by 10-ish% to give 4k. Note the 16:9 crop is 6960 x 3904, making 2x2 pixels gives 3480 x 1952, which is about 10% off 3840 x 2160. Plus the sensor/processor combination in the M6II can do 30fps 18MP Raws with AF/AE until it runs out of buffer, so reading the whole 16:9 area at 30fps is certainly within the realm, or trivial if some video processing is done on-sensor, as is quite common (the 16:9 crop being 50% more pixels than 18MP).

Surely the really big question is whether the 8k 444 Canon camera used at the launch is the next C300 with an 8k sensor option, or something much more expensive? 10:45 on...

The point about equivalent aperture isn't just DoF, it's noise too, as most of the noise comes in with the light and a physically smaller aperture (i.e. higher f/ equivalent number) lets in less of it. Up to a point you don't care, then you do... I like the point about using the crop Sigma on the S1, but then you can only shoot cropped video so more one for the pros with a budget.

I haven't read all 4 pages of replies, but have owned GH3, GH4, GH5 and 5D2 plus 5Dsr. What I will say about 1080p on the GH3 is it doesn't use all the pixels (it skips a bunch) which can lead to quite bad aliasing/moire on subjects that lead to that (the 5Dsr is actually a lot better and really doesn't do aliasing that I can detect). I would be tempted to look for deals on the Panasonic G80/85 (number depends on where you are in the World, it's the same camera, note G not GX) with the 12-60 kit lens (plus check you are happy with the viewfinder when panning). Another option if you have Canon glass is the EOS M50, which does a nice 1080p with Dual-Pixel AF (the 4k isn't as good) and can take your Canon lenses with an adaptor (which is sometimes thrown in).

Firstly if using the 12-60 Leica you'd want to keep to the wide end as the aperture falls rapidly... 2.8 - 12mm up to... 2.9 - 13mm 3.0 - 15mm 3.1 - 16mm 3.2 - 18mm 3.3 - 20mm 3.4 - 23mm (tested twice) 3.5 - 24mm (tested twice) 3.6 - 27mm 3.7 - 29mm 3.8 - 32mm 3.9 - 36mm 4.0 - 52mm A brighter lens would be good but you need to check you get the DoF as it's no use if you don't have enough stuff in focus. Also you can try a longer exposure. Rather than 180 degrees (so 1/50th for 25fps) try shooting up to 360 degrees (i.e. 1/25th) if really dark. Note not so good with much movement or panning. Oh and yes, 25 fps not 50. I like to shoot in VLogL (which is a paid-for upgrade and needs practising before using, also setting a viewfinder LUT which is a built-in option) when it's dark as it gives me the most flexibility on setting the light level I want in editing and not falling off a cliff, but that's for when I don't want a "standard exposure" but want the scene to look somewhat dark. The in-viewfinder histogram can be useful (once moved to a corner). If you want a zoom and don't mind manual focus you could consider using a f2.8 FF lens with a SpeedBooster, if you can rent them from somewhere (e.g. 16-35). Or just rent/buy a f1.2 (or faster) m43 lens. However you might find a Panasonic lens with OIS that supports Dual-IS2 is really useful (I tend to do that when hand-holding over a slightly faster lens with less good stabilisation). Do note that wanting a dark scene to look somewhat dark, rather than fully lit, will help with noise etc. Hope something there helps!

This really depends on the sensor and how well it copes with light coming in at very wide-angles. They are T1.3 with film. With the lower angles from the RF mount they should be somewhere in-between film and EF sensors. Surely all Canon need to do is make a fourth EF-RF adaptor that has Speed-Booster style optics for 4k. They could label it as a Cinema lens thingy to avoid confusing people. If the adaptor electronics is pass-thought I'd assume metabones and the others will have devices really rather soon.

As an aside I have a fairly early GH4 and when I sent it in for the Audio fix (which wasn't, BTW) they just changed the main board and I got it back with an earlier firmware version than it had when I sent it off (which had been 2.3)... P.S. I should say I don't think VLogL is a good idea with internal recording (I like CineD BTW), as it only uses 160 levels so you tend to get posterisation on areas of flat colour, like the sky for example. Presumably external recording won't help if you don't have 10-bit output.

I struggle to see how splitting the Micro Four Thirds cameras off from the Professional Video Division (which seems an ideal fit, except the latter is profitable and the former not-so-much) and making people redundant is quite such a positive as they are trying to spin it... It will be interesting to see what happens, but IMHO "redoubled" seems like marketing spin to keep people buying the products. Not that I'm suggesting cameras are doomed, but I think the ILCs will get trimmed somewhat along the way...

I have a suggestion for how you might do a 6k photo mode... if You're using an anamorphic lens maybe it can extract stills from video and stretch them horizontally to make a "correct" picture with the same image height.

I have a small point/question about the comment that the lower-end cameras don't have a log profile - how about loading Technicolor Cinestyle in a custom slot? Or http://www.similaar.com/foto/flaat-picture-styles/index.html ? (I'd say "or Visionstyle" but they seem to have gone into a stealth mode. There's also Prolost Flat of course.)

Quite interestingly LensRentals say the metabones products are not reliable enough for professional use... https://www.lensrentals.com/blog/2016/06/a-look-at-the-new-sigma-mc-11-lens-adapter/ "The Metabones adapters give a tempting promise of being able to use the larger Canon, Nikon, and Sony A- mount on smaller mounts like the micro 4/3 and Sony E-mount. Many times they work and provide aperture control and support image stabilization, but we have routine problems with these adapters that can be quite frustrating. The most common being a loss or lack of electronic connection between lens and camera, sometimes creating errors and freezing in the camera. The adapters also regularly have screws coming loose and other functional issues. We state a warning to customers on the individual product pages discouraging against relying on the adapter for the success of the shoot."

Thanks for that, I was wondering what was happening...

Diffraction is an inherent property of light passing through an aperture. What you end up with is each part of the light turns into an Airy Disc with particular physical properties. The diameter of the disc only depends on the f-stop, nothing else. Hence the smaller the pixels the more of an issue it is. This is worth a read and includes a calculator (just ignore the camera section in the lower half): http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm (Astronomers are even more interested, as stars, being point light sources, really show the effect.) Generally (IMHO) diffraction blurring starts to become an issue at the pixel level when it hits two pixel widths and gets worse at three widths. Exactly how bad it is depends on stuff like the strength of the anti-aliasing filter (if any) and how much blurring the de-bayering algorithm adds (which is always some). It isn't a hard limit, it's always there and just gets worse, so people have to define the point they start to care about it, which varies somewhat depending on who you read. For still photography it has been taken, for a fair while now, that a 30um blur on a FF sensor (divide by crop factor for others) is the limit for good sharpness of uncropped still images (which comes from lots of perceptual testing of people and a bunch of assumptions, like using 8"x10" images at a particular viewing distance). At the pixel level the 2-3 pixel widths seems to be where people are. I personally think if you are at 2 pixels or less you can ignore diffraction effects at the pixel level (compared to other factors). (Edit) P.S. reverse engineering the graph on the previous page gives about 4.4um as the diffraction limit for f8 (10.7um Airy disc), which is 2.43 pixels and 5.8um for f11 (14.7um Airy disc), which is 2.53 pixels, so I suspect they used 2.5 pixels as their limit. It doesn't magically get really bad at this point, just gradually worse. P.P.S. Also remember as you move away from the plane of focus the image gets gradually blurrier, this is where the 30um value is most often used, as the limits of the depth of field. Diffraction softening just adds to this.

You were saying "my bad" so I threw in my version too... (Who knew it would be so tough to poke fun at myself...)

Hey, I swapped f2.8 and f1.4 in one of my posts (although no-one noticed, not that it stopped me going "Duh!" as I spotted it just outside the edit window), so I think I beat you ;-) Here you go, for people to read and laugh at my ability to go from brain to keyboard... "Not really, the light per unit area (which is what the f-stop is) will be 2 stops less, but there is 4x (i.e. 2 stops) the area at that illumination level, so the amount of light captured to make the image is the same between the f2.8 on m43 and f1.4 on FF. Also the noise in the light depends on the amount of light you capture, so will also be the same."

P.S. It seems to me like NHK showed a sub-m43 sized 8k sensor at ISSCC. It has 33MP (8k is 32MP, sensors always have a few more pixels than the active area) and can do full sensor read-out at 240fps (although you then have to get rid of 3W). The pixels are 1.1um (so about 9mm across which is about half m43).

But they are. A FF f2.8 lens will capture the same amount of light over the image as a m43 f1.4 lens (both wide-open). That also means you get the same shot noise (which is most of the noise in the image). You need to increase the ISO on the FF camera by 2 stops to get the same exposure. However that doesn't affect the noise as (i) most of the noise is in the light and changing the ISO won't change that and (ii) the sensor read noise will usually fall a bit with increasing ISO, so the total noise will be lower if anything. (Think of it this way... at the same illumination level if a f1.4 lens puts 1M photons onto a m43 sensor then a f2.8 lens will put 1M photons onto a FF sensor. Of those photons on average 1,000 will be the shot noise, in both cases.) Increasing the ISO isn't a problem as the FF sensor will be able to capture around 4x the electrons of the m43 sensor due to its greater size. Increasing ISO by two stops knocks about a factor of 4 off that so it will be about the same as the m43 sensor and not saturate if the m43 sensor doesn't. Also remember a 50mm f2.8 lens has the same diameter entrance pupil as a 25mm f1.4 lens (17.86mm) so the DoF will be the same at the widest aperture. The FoV will also be the same. Diffraction softening will also occur at the same DoF (it goes with DoF, regardless of format). Hence the minimum and maximum amounts of available DoF are the same.

Am I reading this incorrectly? It seems to agree with me to a decent approximation. The f8 and m43 lines intersect at about 13MP, which is about the 8MP (in a 16:9 crop) needed for 4k.

If you're delivering a 4k image, as the article suggested (downscaling 6k, not producing 6k) diffraction isn't that much of a problem. At f8 (so equivalent to f16 DoF on FF - that's plenty, yes?) the blur caused by diffraction will be 10.7um diameter (regardless of format, plus generally, although not evenly, declining towards the edges). This will begin to get noticeable at pixel sizes under about 5.35um, but will only increase slowly (it'll be worse under 3.5um). Taking 4k from a 17.3mm wide (m43) sensor means each video pixel is effectively 4.2um wide. You'll see some diffraction softening, but not much - and that's with a lot of DoF. (You get more softening from debayering the sensor than minor amounts of diffraction.)

Not really, the light per unit area (which is what the f-stop is) will be 2 stops less, but there is 4x (i.e. 2 stops) the area at that illumination level, so the amount of light captured to make the image is the same between the f2.8 on m43 and f1.4 on FF. Also the noise in the light depends on the amount of light you capture, so will also be the same.

I'm afraid I struggle to see the business case here. Panasonic aren't doing that well selling cameras (in Japan they are now selling less mirrorless than Canon) and new lenses just trickle out. I doubt anyone inside Panasonic sees throwing lots of money into an all-new camera system as likely to be a winner, especially in a generally declining market. I think they only sell a decent number of GH4s as it offers a lot for the money and is stills and video. No way will videographer sales (if they effectively ditch the stills side by only having 8MP with m43 lenses) get them back that money at similar prices on a new format. I think if they go to a different sensor size they will have to use 3rd party lenses (developing their own being an unlikely investment), so whose mount would be possible - Canon EF maybe using the V35 sensor (4096x2160)? Won't be any good for stills. Plus sounds tricky to avoid damaging higher-margin V35 LT sales... I can't see it. To do your suggestion at a minimum they'll need someone else to have designed a sensor just for this camera and to be happy about low sales volumes for that sensor (which usually means prices that drive 5-figure camera pricing, so they've pretty much just designed the V35 LT again). P.S. I read your whole article before coming to the forum for my first post on the subject.

It's a tricky one. They can improve the DR by lowering the read noise and 3.1 electrons is a fair bit these days. To reduce the shot noise they need to capture more Photons, but the lens f-ratio controls how many fall on the sensor - same lens, same number of Photons. You can improve the Quantum Efficiency, but 58% is really good. You can weaken the colour filters to let a bit more of other colours through. However if you make more electrons you need to make a bigger pixel to store them, which gets harder for the same surface area. I can see them improving the DR (but I don't think they will, as they will want 50p/60p which mean running it faster, which means more noise, so other improvements will get eaten up), but I think for overall noise all they can do is what people have mostly been doing, which is implement ever cleverer noise reduction in their processor. It's doable, but I don't think it'll get you to where Neat Video can get you now. Just my opinion though. Well, except the physics bits. (Edit) Should have said, they can maybe improve the DR a bit if they use the Arri/Canon way of splitting the pixels up, although I suspect that would run into a lot of patents.

Provided someone designed a 32MP FFs sensor that is fast enough to shoot 8k video (which is very fast indeed). I find it hard to see how they would make their money back as lots of the people shooting 8k are also the people using cine lenses that generally don't have an FF image circle. Sony might, but it would be hard to keep cool. At ISO 800 a GH4 pixel can hold 5091 electrons and has a read noise of 3.1 electrons. That means the noise in the light will exceed the read noise once you get to about 17 electrons (it's 56% efficient converting photons to electrons and the noise in light is the square root of the number of photons). So anytime the pixel is much more than 3.3% illuminated the sensor read noise isn't that big a thing. For example at 10% illumination (509 electrons) you have on average 17 electrons of noise and the rest is signal. Even if you go from 3.1 electrons of read noise to 0.5 it isn't making a significant difference (btw noise combines as less than straight addition).

I can't see you getting ISO 3200 as good as the GH4's 800 unless it's by more noise reduction, as it's a sensor area issue, most of the noise coming in with the light rather than being due to the sensor. I don't know if you'll get 10-bit but it's a definite maybe. I think the rest looks likely. I'd also like FHD modes that don't skip 30% of the pixels, so the aliasing/moire goes away. I suspect they will try to be close to the Ultra HD Blu-ray spec, including gamut.