Nikon Z6 III Sensor Readout Speed Measurements
ฝัง
- เผยแพร่เมื่อ 20 ก.ค. 2024
- My full sensor readout speed measurements of the Nikon Z6 III.
Results page for my open-source rolling shutter measurement project:
horshack-dpreview.github.io/R...
Project homepage:
github.com/horshack-dpreview/...
0:00 Introduction
0:20 Photo readout speeds
1:46 Video readout speeds
3:40 Comparing results
6:39 Methodology - วิทยาศาสตร์และเทคโนโลยี
![THE TOYS x NONT TANONT - ดอกไม้ที่รอฝน (spring) [Official MV]](http://i.ytimg.com/vi/pPa1d5cC8M4/mqdefault.jpg)
There's an important aspect to fully-stacked sensors I forgot to mention in my video. None of the fully-stacked sensors currently on the market can use their fast, on-chip stacked memory in video modes. It's presumed this is due to thermal constraints running that fast memory continuously for long stretches of time. A big clue about this came from the datasheet for the stacked sensor in the Sony A9, which states that none of its video modes can use the stacked memory due to heating concerns. This is why the sensor in the Z8/Z9 has slower readout speeds for video than the Z6 III, even though the Z8/Z9 have almost 4x faster readout for still images vs the Z6 III.
This is a gold video for all the camera fans! Beats almost every other analysis video except perhaps Gerald Undone.
Your rolling shutter project Is stunning.
Just wow, very impressive and useful compilation!
You sir just combined my two favorite subjects in one video. My love for cameras (Z6 especially) and my nerd side for statistics :D A value pack video indeed!
Awesome, this is really interesting! Thanks for your work on this project.
Thank you very much. So one take away for me is that to speed up the readout time and lessen the risk of banding, shooting in DX mode gets you a useful speed up of the readout.
Shooting in DX crop mode to reduce sensor readout time does not reduce rolling shutter artifacts (including banding in artificial lighting) because the per-row timing is unchanged - the camera is simply reading out fewer rows, so only the total readout time is lower. The DX rows that are read will be subject to the same inter-row readout timing, and thus subject to the same rolling shutter artifacts. However, shooting in JPEG/HEIF will reduce rolling shutter artifacts because it uses a faster 12-bit sensor readout, which reduces the per-row readout time.
@@testcams Ah. Thanks for that explanation. Like quite a few photographers I am plagued with banding issues on my Z8's even with antiflicker on. I work in theatre and film sets all the time. Film sets are fine but a lot of theatres have a huge assortment of LED lights and some, especially it seems when run at low level or predominately one of the RGB channels, generate banding on the Z8. I had hoped the Z8 would be an answer to banding alas not always. I have Z6lll on order so that at least one camera is mechanical shutter. My original Z6's are fine with mech shutter but the AfF is problematic for action. Thanks again for the clarification.
Unfortunately even a very fast stacked sensor like the Z8/Z9/A1 aren't immune to artificial-lighting banding since many LED lights employing PWM cycle at very high rates. For example I have a smartbulb at home that cycles at 2000Hz for all brightness levels below 100%. Have you tried the high-frequency flicker reduction, which lets you set a fractional shutter speed that can better align to the banding frequency? Unfortunately Nikon requires you to determine the shutter speed manually - Canon's implementation actually detects and selects the frequency for you.
@@testcams Unfortunately it is just not feasible to select a high frequency fractional shutter in my theatre work. By the time I had determined it the scene or the lighting will have changed. In a steady state lighting setup it would work but not in my usual scenario. Perhaps you can answer this question for me? I believe that if I get banding at say a 1/100th sec moving it to say 1/250th sec, a much more desirable shutter speed, isn't ever going to solve it is it? It will just be worse? I realise 1/100.2 might help but as a shutter speed that is just as compromised as a 1/100th. It's a fast shutter speed I need. I am pretty sure I know the answer but I think you might be able to confirm it one way or the other. Many, many thanks for taking the time to answer. It's really appreciated. Subscribed. Best wishes.
Faster shutter speeds don't change the number of bands since that is a function of the light frequency vs sensor readout time. However faster shutter speeds do change the appearance of those bands - the faster the shutter speed, the sharper the edges, which subjectively makes the banding even more objectionable. I have a section in my GitHub project that discusses this: github.com/horshack-dpreview/RollingShutter?tab=readme-ov-file#effect-of-shutter-speed-on-banding
Excellent work. And. In photographic jargon, the generic term for what you call "rolling shutter" is "time parallax". Rolling shutter is a legacy of movie cameras with a rotating shutter near the focal plane. In the rotating shutter mechanism, two discs with open slits in them could be adjusted relative to each other and the angle between them determined exposure time, provided we knew FPS.
While our digital cameras in the Bayer paradigm (using analogue and colour blind sensors) started as scanning photosite by photosite and doing AD conversion in a discrete chip, for decades, TV/video cameras also scanned, but remained in the analogue domain.
What is called "focal plane shutter" (erroneously) in English, also has time parallax. In other languages, your focal plane shutter may be called curtain shutter or slit shutter.
The race car images shot by Lartigue in the Interbellum show time parallax as slanted wheels, thus suggesting high speed. While an artefact following from shutter dumbness, the concept was adopted e.g. by cartoon makers and many humans have these slanted wheels engrained in their brains as suggestive of high speed since early childhood.
Lartigue's shutter was an after-market device placed over the front of the lens.
Note that "leaf shutters" - like a second aperture (iris) mechanism that however can completely open and closed in a timed manner - have no time parallax (but do have a vignetting effect). Not having time parallax directly impacts flash (strobe, speedlight) sync times.
As to "shutter speed" note that about all shutters run at constant and consistent speed in their parts all the time. If "speed" is meant to relate to how fast the closing is after the opening, fine, but the speed of components stays the same.
Word, words, words - yes, but essential to deeper understanding and that's what this is all about.
Excellent work. And. While the milliseconds taken by the camera for sensor readout matter, I would correlate it to another thing: MB/s or Gbps bandwidth. To figure this out, we need a few columns more data: (1) the gross sensor rows and columns, not just what is called "effective"; (2) real bit precision. You may set your camera to 14 bits depth, but the sensor, c.q. the stacked chip, may relay 16 bits to the frame cache.
Example. My 45.4 MP camera has a readout time of 0.003731 second. At 45.4 MP, we are effectively using 8,256*5,504 photosites. But the sensor has 52.37MP "gross". It uses the additional rows and columns outside the effective area for two things: (1) defeat the requirement for an edge algorithm in deBayerisation and (2) IBIS in software with zero latency in the image processing pipeline of motion in the sensor plane.
If we want to know the sensor stack's bandwidth, we need the gross number and total bits. While we shoot at 14 bits depth, some sensor fab/foundry advertises their inhouse sensor designs as "16 bits" and then in a footnote say, but 14 are photographically meaningful. So the 16 don't matter to images , and we don't see them back in the camera's menus. But they are relevant in bandwidth. While my camera's sensor and stacked chips got (logically) designed by the camera company, and I don't how my camera is doing the 2 additional bits or not, let's assume that it uses two bits to indicate the Bayer filter colour with each photosite's data element. That may be relevant in the case of moving (IBIS) sensors and moving effective crops out of the total sensor image area.
Now we get to the 8,863*5,909=52.37 MP ballpark gross resolution and at 16 bits this means each gross frame ~equates to 837.9 Mbits (or 0.837861 Gbits) that in 0.003731 sec readout mean the camera/sensor works at 224.55 Gbps (224,550 Mbps). And that's before raw bits are polished, or any bits get compressed and deBayerised for the eVF. And, and.
Yes, readout time is important for time parallax, but bandwidth reveals the state of technology under the hood/bonnet. Or, while a Z 6iii has very fast readout, it by far does not have the bandwidth of a Z 8 - and considering price that was totally expected.
Word, words, words - yes, but essential to deeper understanding and that's what this is all about.
Great work! i wonder if the camera has oversampled 1080p in DX mode? the z8 and z9 does this, whereas the 1080p in FX is binned….
Excellent work - thanks.
well done man.
Hell yeah! Thanks!
YOUR CHANNLE IS UNDERATED ....
Incredible work! Love it. Do you work in the Camera industry?
Excellent work! I have been looking for such a detailed analysis for a long time now. I am wondering if you ever tried to estimate how fast mechanical shutter slits are traveling in modern mirrorless cameras? Is it quicker than 1/500 of the second? Otherwise, one or two cycles should be visible (like with the super-fast Nikons Z8/Z9). Can you share any thoughts on that matter? And maybe any test images?
Mechanical shutters can be measured using this same technique. I've actually measured a few of the MILC in my database. If you go to my live results page at horshack-dpreview.github.io/RollingShutter/ and do a text search for "mechanical" you'll see the measurements.
@@testcams Thank you so much! I've been looking for such accurate measurements for many months.
Good detailed test results.
If I may make a suggestion, please stay with the data. Rocking a still photo of a Z6III back and forth from 0:19 to 3:40 is not only unnecessary, but is distracting and annoying.
Keep up the good work!
Thanks. There's actually a funny functional reason I include motion in all my frames. I have a smart tv whose remote I sometimes accidentally press mute or pause. I can quickly determine which by looking at the content - if there's motion then it's mute, otherwise it's pause. If there are long stretches of silence or stasis in the content I'm watching then my strategy doesn't work. I like to assume my experience of the world isn't unique so just in case there are others like me I try to keep something moving at all times.