Whats up with ISO buttons?

[Ed Herdman]

Wrote my last reply without seeing your last reply:

Another semantic issue here. You keep on talking about 'correct' this and 'correct' that, and 'correct' suggests something absolute and unarguable. So, you're liking 'correct exposure' and 'correct gain', but if you put these together, you get nothing at all, since the 'correct' exposure (if I understand what you mean by it) is, according to your view, controlled by the 'gain'. Therefore it would have to be 'correct' exposure for a given 'gain', not both 'correct'. In any case, please say what you mean by 'correct' in these contexts.
--
Bob

I think this issue washes out entirely, given what you've said and after my hopefully having absorbed it. The problem is, as we seem to agree, one of how the desired end result (if that makes more sense than saying "correct exposure") - a sufficiently bright display image - is achieved from an image without gain applied. When I said "correct exposure" I meant that the image, on current cameras, should appear bright enough in a direct review. Doing this with the more controlled mathematical fashion on the CPU seems to make more sense in terms of absolute image quality.

What's interesting to me is the apparent consequence that, since the sensitivity of a sensor is fixed, images that appear to be "saved" by a high gain setting must as a natural consequence always be achievable at the default ISO and adjusted for brightness in the CPU. On the other hand, images that cannot be achieved with a high ISO setting require the photographer to adjust aperture, or shutter, or both, to end up with the desired exposure - the only two variables that actually affect the exposure, as it turns out, since the same amount of light is hitting the sensor as before.

This sounds like an artifact of the days when you could simply swap out films of different speeds (which, to my understanding, also had their own noise issues, but I'd better not get any further off track for my own good because that introduces more factors).

In practical terms: On my camera, perhaps it'd be best (when time and lighting allows allows) to use ISO to get a "preview shot" and from there set it manually back to the base, assuming that I have all manual settings. The way that camera is built, however, it seems to require the gain-applied shot to get the desired exposure (well, I slipped back into that) - if it doesn't have a fast enough ISO available, it will start messing with the time value if aperture priority is selected, and in shutter priority mode it will start messing with the shutter speed. This is undesirable if the end result should be an image which works perfectly well as a "brightness pushed" image. I do get that some images will simply not come out without longer shutter speeds or wider apertures, regardless of the ISO setting, but my concern is in those marginal cases.

 

[bobn2]

Another semantic issue here. You keep on talking about 'correct' this and 'correct' that, and 'correct' suggests something absolute and unarguable. So, you're liking 'correct exposure' and 'correct gain', but if you put these together, you get nothing at all, since the 'correct' exposure (if I understand what you mean by it) is, according to your view, controlled by the 'gain'. Therefore it would have to be 'correct' exposure for a given 'gain', not both 'correct'. In any case, please say what you mean by 'correct' in these contexts.
--
Bob

I think this issue washes out entirely, given what you've said. The problem is, as we seem to agree, one of how the desired end result (if that makes more sense than saying "correct exposure") - a sufficiently bright display image - is achieved from an image without gain applied.

Any image can be made 'sufficiently bright'. Intrinsically, with a digital sensor and its linear brightness to response characteristic there is absolutely no reason to use exposure to control 'brightness'. Exposure does control noise, so one very much might want to control exposure to control noise, but brightness is entirely dependent on processing, except for the film emulation in the processing of current cameras.

When I said "correct exposure" I meant that the image, on current cameras, should appear bright enough in a direct review.

I suspect that you did, but this is fundamentally saying that the goal of exposure control is to achieve some desired brightness.

Doing this with the more controlled mathematical fashion on the CPU seems to make more sense in terms of absolute image quality.

What's interesting to me is the apparent consequence that, since the sensitivity of a sensor is fixed, images that appear to be "saved" by a high gain setting must as a natural consequence always be achievable at the default ISO and adjusted for brightness in the CPU.

In practice that depends on the quality of the capture chain. While the sensitivity of the sensor is fixed, it never 'gathers' any more light than it does at base ISO, the ADC systems used often have very high noise, and insufficient dynamic range to capture everything the sensor captures. That is the point of the variable gain, to boost the level of low-light signals coming from the sensor high enough to avoid the noise in the ADC system. A base ISO raw file from such a camera adjusted in post-digitisation processing looks very noisy in the shadows.
Cameras with better ADC systems don't need this. A raw fil

On the other hand, images that cannot be achieved with a high ISO setting require the photographer to adjust aperture, or shutter, or both, to end up with the desired exposure - the only two variables that actually affect the exposure, as it turns out, since the same amount of light is hitting the sensor as before.

The question then is, what determines the desired exposure, and really it is the amount of noise that the photographer will put up with, or is serious enough to compromise the DOF or motion blur requirements.

This sounds like an artifact of the days when you could simply swap out films of different speeds (which, to my understanding, also had their own noise issues, but I'd better not get any further off track for my own good because that introduces more factors).

Film had grain issues, which were different from noise issues, and also an S shaped sensitometric curve - that made it rather important to choose your exposure to hit the right part of the curve for he film you had loaded. With positive/negative processes (especially monochrome, where you could change the contrast of the paper) there was still quite a lot of leeway, even from frame to frame, but nothing like what there is with digital.
--
Bob

 

[Ed Herdman]

Hmm - is there a way to tell, before the capture, when it is better (in terms of overall image quality) to boost brightness after the fact, and take the noise from the ADC system, versus introducing some noise from a higher gain setting? It seems to me - perhaps this is a fantasy - that the necessity of the gain, which you have outlined, is currently at odds with its main use, which is overusing it to achieve a "bright enough" image before CPU capture. To split that apart would seem to require the camera itself decide when to simply apply gain, and when to further brighten the image on the CPU (which I suppose further could be adjusted later just like white balance - in that it's an artificial enhancement from no known reference point but rather a best guess). I suppose this might have some benefit with current live view systems, for instance.

 

[bobn2]

Hmm - is there a way to tell, before the capture, when it is better (in terms of overall image quality) to boost brightness after the fact, and take the noise from the ADC system, versus introducing some noise from a higher gain setting?

So far as noise goes, increasing gain is only good. Where it's bad is risking blowing highlights, since each doubling of gain halves the highlight headroom. So, for a camera which uses or needs variable gain, the general rule if you're really trying to minimise noise is to set the ISO (and with it the gain) as high as possible for any given exposure, but avoiding clipping the highlights. A rather more sophisticated approach is to know at which ISO the camera stops raising the gain. You can see this from the read noise curves on my sensorgen.info site.
For instance, on a D3s



the read noise flattens at 3200 ISO (meaning no more analog gain past that ISO), so for a D3S, no point going above 3200, you are just risking the highlights. Similarly, for a 5DII



don't bother with anything above 1600.
Or a D3X



above 400
On a D2X



You might as well set it on 100, same for a D7000

It seems to me - perhaps this is a fantasy - that the necessity of the gain, which you have outlined, is currently at odds with its main use, which is overusing it to achieve a "bright enough" image before CPU capture. To split that apart would seem to require the camera itself decide when to simply apply gain, and when to further brighten the image on the CPU (which I suppose further could be adjusted later just like white balance - in that it's an artificial enhancement from no known reference point but rather a best guess). I suppose this might have some benefit with current live view systems, for instance.

That's pretty much what Auto ISO does, which I think makes it a pretty useful feature, if properly executed, for non 'ISOless' cameras. If cameras were designed around the 'exposure centric' (as opposed to 'ISO centric') method of operation, then the in-camera JPEG processor could determine the right 'ISO' to use simply by evaluating the data in the capture, which would be more complete and accurate than any meter.
--
Bob

 

[Ed Herdman]

That's precisely what I was looking for, thanks.

Interestingly, the 50D (similar, if not the same, sensor as the T1i) curve flattens out exactly where my field testing suggested it would be - at 800. 1600 and above (expecially the extended modes) become noisier for no apparent gain (using the word finally with its common meaning). I'm a bit impressed to see that the 550D seems much improved in this area, pushing pretty close to ISO 3200 before it flattens out (3200 is an improvement over 1600, apparently Canon was stretching the noise curve there a bit, understandably so - I notice that coincides with DR being somewhat smaller at that setting than it was on the 50D, for example).

A couple points - is the 1D Mark IV's 56K ISO result a good result? There are a couple plateau steps before it, but it still appears there's DR being squeezed out at that range.

Secondly - you mention the JPEG processor being used to evaluate data in the capture for a more accurate reading than a meter. My supposition - the through-lens metering may be more precise, but not more accurate, then. The big point I wanted to ask was if here you are using the JPEG processor interchangeably with the camera's ability to automatically adjust the brightness of an image, including with RAW format?

And on that point, veering into the realm of chip design unfortunately, is the JPEG processor (which I understand just to be a set portion of the CPU, perhaps even microcode) repurposed with RAW capture to deal with ISO balancing (a bit like, to use perhaps an incomprehensible analogy, unused data pins on expensive purposed microprocessors or microcontrollers were often repurposed to achieve other tasks in old computer hardware design, such as arcade games and some proprietary computer systems), or is that another part of the chip? The term "JPEG engine" seems to refer, as it is most often used, to much more than simply the JPEG compressor, but also the balance of what other data is prioritized or enhanced - at least in older DPReview reviews this appears to be the case.

 

[bobn2]

That's precisely what I was looking for, thanks.

Interestingly, the 50D (similar, if not the same, sensor as the T1i) curve flattens out exactly where my field testing suggested it would be - at 800. 1600 and above (expecially the extended modes) become noisier for no apparent gain (using the word finally with its common meaning).

Thanks, I appreciate those little bits of info, when what's seen in the graphs is confirmed by someone's real-world experience.

--
Bob

 

[Ed Herdman]

If that's useful, perhaps my most recent setup should be described: Images of fireflies at night with some sky visible, 1/6 to 1/8 second (manually selected), f/1.4 (focus distance manually estimated, confirmed from the photo review) on the Canon 50mm prime lens. The sky was too bright for the balance of the photograph but didn't come close to clipping. I had used the same aperture and shutter values for earlier photographs in slightly better light, when I had the ISO set to 400. I didn't take ISO 1600 images in that set, and part of my reasoning for not pushing it higher (aside from noise) was that I was already experiencing hot pixels - which I now think may have been mistaken reasoning. I need to experiment with the "long ISO noise reduction" feature, which (so I am told) subtracts a black frame from the eventual images. Personally, I feel that a few nasty red spots here and there are preferable to noise throughout the whole image distorting everything. Even in the center of the frame they usually are hidden by other texture well enough.

 

[AlfBundy]

Hello,

What's the point of manually changing ISO when Auto ISO does it for you ?

Aperture is important for DOF control,
Shutter speed is important for non-stab lenses (1/FL) or fast moving subjetcs.
Then you want the lowest possible ISO. And that's exactly what Auto ISO does.

On my Sony NEX, Auto ISO does an even better job than I can do. Because Auto ISO is able to select semi value (250,320,500,640,...) while manual ISO only allows me to select full stop ISO (200,400,800,...).

Don't you always want the lowest possible ISO according to your Aperture and Shutter speed settings ?

 

[glanglois]

Hello,

What's the point of manually changing ISO when Auto ISO does it for you ?

Aperture is important for DOF control,
Shutter speed is important for non-stab lenses (1/FL) or fast moving subjetcs.
Then you want the lowest possible ISO. And that's exactly what Auto ISO does.

On my Sony NEX, Auto ISO does an even better job than I can do. Because Auto ISO is able to select semi value (250,320,500,640,...) while manual ISO only allows me to select full stop ISO (200,400,800,...).

Don't you always want the lowest possible ISO according to your Aperture and Shutter speed settings ?

Sounds as if NEX has a bit of a flaw in manual ISO.

 

[tom sugnet]

F-button can be your ISO button, if you wish so,
on any entry level Nikon.

--
Feel free to visit my photo sites:
http://tom.st , http://www.foto.tom.st

 

[Ed Herdman]

Hello,

What's the point of manually changing ISO when Auto ISO does it for you ?

I wouldn't publicly admit that a camera always selects Auto ISO better than I do...but then again, I haven't used the NEX so it may be as good as you say. But in some situations, the camera may well think that you want a nice bright image when you don't, and pushes up the ISO beyond what you would select. And on cameras with iffy Auto ISO (I put the Canon T1i in that group) it's unfortunately a fact of life that it has to be manually limited much of the time. This is part of the reason why there's been a call on the Canon Rumors forum from time to time (which I support) of allowing Auto ISO with an optional top limit, instead of only having two choices - fully automatic selection and manual. From my discussion above, however, it sounds as if the newest cameras have an even better solution available (only applying gain up to the limit of the sensor and then altering brightness with the CPU).

Sounds as if NEX has a bit of a flaw in manual ISO.

I don't know, it works that way on at least many DSLRs, too. Do other cameras allow you to choose arbitrary ISO values? There are even fixed steps you can choose on most cameras between different time and aperture values (on the T1i, again, I have a choice between half-stop increments, probably for faster switching between high and low values, and the default third-stop steps).

 

[jrtrent]

At least Pentax seem to understand that ISO is now simply a third exposure parameter.

It's not a third exposure parameter. Exposure is simply the density of light times time at the sensor, and ISO doesn't change that. What it is is the right way to change output image brightness if you're using in-camera processing.

Changing the ISO setting on a camera is one way to change the image brightness, along with aperture, shutter speed, the application of neutral density filters, and probably others that I'm not aware of or don't use; it is not necessarily the right way. There are numerous articles referring to exposure as a triangle of ISO, aperture, and shutter speed, explaining the different effects each of these controls might have on the final image, and thinking of exposure control in that way can be helpful.

Why have Olympus and Nikon forgotten that?

Because it wasn't ever true. It's an idea that also has its roots in misinformed ideas about what exposure was from people used to lab processed film.

I'm not sure I understand you here. Rather than being rooted in film-based thinking, considering ISO as a third exposure variable would seem a product of digital technology, at least according to this footnote in the DPReview glossary entry on exposure:

"For the sake of simplicity, as is the case in this article, Light Value is often referred to as "exposure value", grouping aperture, shutterspeed and sensitivity in one familiar variable. This is because in a digital camera it is as easy to change sensitivity as it is to change aperture and shutterspeed. Many digital cameras even offer an auto-ISO mode. Although sensitivity will not change the amount of light entering the camera, it changes the effect of it and is therefore a third variable that can be adjusted to achieve an exposure that matches what is measured by the camera's light meter." http://www.dpreview.com/learn/?/key=exposure

 

[bobn2]

If that's useful, perhaps my most recent setup should be described: Images of fireflies at night with some sky visible, 1/6 to 1/8 second (manually selected), f/1.4 (focus distance manually estimated, confirmed from the photo review) on the Canon 50mm prime lens. The sky was too bright for the balance of the photograph but didn't come close to clipping. I had used the same aperture and shutter values for earlier photographs in slightly better light, when I had the ISO set to 400. I didn't take ISO 1600 images in that set, and part of my reasoning for not pushing it higher (aside from noise) was that I was already experiencing hot pixels - which I now think may have been mistaken reasoning. I need to experiment with the "long ISO noise reduction" feature, which (so I am told) subtracts a black frame from the eventual images. Personally, I feel that a few nasty red spots here and there are preferable to noise throughout the whole image distorting everything. Even in the center of the frame they usually are hidden by other texture well enough.

Yes, with a 50D you should have done better at 1600 ISO but the same exposure. Remember that raining the ISO doesn't cause noise, it is the reduction in exposure that often, but not always, goes with it. So 1600, 1/6, f/1.4 would have been better than 400, 1/6, f/1.4, so long as the light bits of sky didn't blow out. This is a dilemma you wouldn't have had on a D7000, you could just have set 100 or 200, 1/6, f/1.4 and shot away without risk of any highlights going.

As for long exposure noise, you can use long exposure NR or take a dark frame at your intended settings with the lens cap on, then subtract that in processing.
--
Bob

 

[glanglois]

Hello,

What's the point of manually changing ISO when Auto ISO does it for you ?

I wouldn't publicly admit that a camera always selects Auto ISO better than I do...but then again, I haven't used the NEX so it may be as good as you say. But in some situations, the camera may well think that you want a nice bright image when you don't, and pushes up the ISO beyond what you would select. And on cameras with iffy Auto ISO (I put the Canon T1i in that group) it's unfortunately a fact of life that it has to be manually limited much of the time. This is part of the reason why there's been a call on the Canon Rumors forum from time to time (which I support) of allowing Auto ISO with an optional top limit, instead of only having two choices - fully automatic selection and manual. From my discussion above, however, it sounds as if the newest cameras have an even better solution available (only applying gain up to the limit of the sensor and then altering brightness with the CPU).

Sounds as if NEX has a bit of a flaw in manual ISO.

I don't know, it works that way on at least many DSLRs, too. Do other cameras allow you to choose arbitrary ISO values? There are even fixed steps you can choose on most cameras between different time and aperture values (on the T1i, again, I have a choice between half-stop increments, probably for faster switching between high and low values, and the default third-stop steps).

I agree on the first part. I like being able to put both maximum and minimum limits on my ISO. I use that a great deal.

Clearing up my comment on the NEX: I was referring to the limitation (as I read the post) on the steps allowed for manual ISO. I like having the choice between 1/2 and 1/3 stop. I may have a full stop option but I haven't looked into it.

 

[bobn2]

At least Pentax seem to understand that ISO is now simply a third exposure parameter.

It's not a third exposure parameter. Exposure is simply the density of light times time at the sensor, and ISO doesn't change that. What it is is the right way to change output image brightness if you're using in-camera processing.

Changing the ISO setting on a camera is one way to change the image brightness, along with aperture, shutter speed, the application of neutral density filters, and probably others that I'm not aware of or don't use;

None of those things, apart from changing the ISO setting, determine the image brightness. All they do is determine how many photons are counted at each pixel. That is not translated to a 'brightness' until you do the processing. If you use in-camera processing, it is purely the ISO setting that determines image brightness.

it is not necessarily the right way. There are numerous articles referring to exposure as a triangle of ISO, aperture, and shutter speed, explaining the different effects each of these controls might have on the final image, and thinking of exposure control in that way can be helpful.

I think it is very unhelpful, because it leads people along two faulty lines of thought:

• that raising the ISO is in some way doing something similar to increasing the amount of light projected onto the sensor. It isn't.
• that the amount of light projected onto the sensor has a direct relationship with the brightness of the final image. It doesn't.

There are two steps in digital image forming, counting the photons in each pixel and translating those values to a range of image tones. Exposure affects the first stage only and ISO affects the second stage only, the effects are fundamentally different.

Why have Olympus and Nikon forgotten that?

Because it wasn't ever true. It's an idea that also has its roots in misinformed ideas about what exposure was from people used to lab processed film.

I'm not sure I understand you here. Rather than being rooted in film-based thinking, considering ISO as a third exposure variable would seem a product of digital technology,

ISO is not a third exposure variable, so while thinking of it that way might be very modern and digital, it would also be very wrong.

at least according to this footnote in the DPReview glossary entry on exposure:

"For the sake of simplicity, as is the case in this article, Light Value is often referred to as "exposure value", grouping aperture, shutterspeed and sensitivity in one familiar variable.

There's a mistake straight away, the sensitivity of a digital camera never changes.

This is because in a digital camera it is as easy to change sensitivity as it is to change aperture and shutterspeed.

That is wrong also, it is impossible to change sensitivity.

Many digital cameras even offer an auto-ISO mode. Although sensitivity will not change the amount of light entering the camera, it changes the effect of it

no it does not, so far as the pixels are concerned the effect is just the same. The processing (analog and digital) is different

and is therefore a third variable that can be adjusted to achieve an exposure that matches what is measured by the camera's light meter." http://www.dpreview.com/learn/?/key=exposure

And that is just confused. Adjusting the ISO doesn't change the exposure, what it does do is adjust the target exposure that the meter sets. So, by adjusting the ISO, you never change exposure, you change the exposure the camera is expecting to the one you have. That is a very useful thing to do, but completely undermines the idea of 'correct' exposure.

In the end though, if you want to choose a different exposure, you can, you don't need an ISO control to do it for you.
--
Bob

 

[Ed Herdman]

Yes, with a 50D you should have done better at 1600 ISO but the same exposure. Remember that raining the ISO doesn't cause noise, it is the reduction in exposure that often, but not always, goes with it. So 1600, 1/6, f/1.4 would have been better than 400, 1/6, f/1.4, so long as the light bits of sky didn't blow out. This is a dilemma you wouldn't have had on a D7000, you could just have set 100 or 200, 1/6, f/1.4 and shot away without risk of any highlights going.

As for long exposure noise, you can use long exposure NR or take a dark frame at your intended settings with the lens cap on, then subtract that in processing.
--

Yes, I need to set up a dark frame soon and see about applying it to my images (I don't know if DPP or the Gimp allows it...I'm somewhat constrained to free software at the moment). I'm actually using a T1i, aka the 500D. I will try ISO 1600 again, but I don't expect it to work better than ISO 800 in similar lighting because of the sensorgen.info data which predicts it won't work out well - which, as I mentioned before, already seemed to be beyond the limit where images still worked well. If I need brightness I can try getting some of it back afterwards in post.

 

[Ed Herdman]

Clearing up my comment on the NEX: I was referring to the limitation (as I read the post) on the steps allowed for manual ISO. I like having the choice between 1/2 and 1/3 stop. I may have a full stop option but I haven't looked into it.

On the T1i, Custom Function I: Exposure's first entry is "Exposure level increments." By default it is at "0: 1/3-stop." (The little display at the bottom of the screen with alternating 0s and 1s for custom functions reminds me a bit of DIP switches.) Changing to the half step option makes the dial increments change like this:

For aperture with a f/2.8 lens:
F2.8 - F3.5 - F4 - F4.5 - F5.6 - F6.7 - F8

Back to the default 1/3 stop increments:
F2.8 - F3.2 - F3.5 - F4.0 - F4.5 - F5 - F5.6 - F6.3 - F7.1 - F8

For shutter speed at 1/2 stop increments:
1/8 - 1/10 - 1/15 - 1/20 - 1/30 - 1/45 - 1/60 - 1/90

Back to the default 1/3 stop increments:

1/8 - 1/10 - 1/13 - 1/15 - 1/20 - 1/25 - 1/30 - 1/40 - 1/50 - 1/60 - 1/80 - 1/100

The 1/2 stop modes, as expected, have a few odd numbers that don't turn up in the 1/3 stop mode.

While I wrote this comment I did something amusing: I turned off ISO Expansion (so top ISO is 3200, instead of 6400 and then "H" which is 12800) and turned long exposure noise reduction On (setting 2 - setting 1 is "Auto," so I'm not sure if it will try to apply it for all frames yet, will have to take a look). Where it gets amusing is that I then took a 30 second frame (with the cap on the lens), and afterward I hit the Play button to see that frame. Surprisingly, it didn't show a noisy black frame, but instead my previous image! I tried zooming in, and then zooming out, and also switching images with the wheel. Camera appeared frozen, but the constant red light on the back indicated that it was silently reading the sensor for the subtraction frame. I let that finish, and when it was done the previous photo (the play screen was still up) zoomed in a little, and then out, and then returned control of the camera to me. Overall, that reminds me a bit of how a personal computer might respond to queued commands when it's doing something and can't respond right away, without doing something unexpected. Something to watch out for but not totally horrible really; it's just waiting to do what you asked it to.

 

[bobn2]

Yes, with a 50D you should have done better at 1600 ISO but the same exposure. Remember that raining the ISO doesn't cause noise, it is the reduction in exposure that often, but not always, goes with it. So 1600, 1/6, f/1.4 would have been better than 400, 1/6, f/1.4, so long as the light bits of sky didn't blow out. This is a dilemma you wouldn't have had on a D7000, you could just have set 100 or 200, 1/6, f/1.4 and shot away without risk of any highlights going.

As for long exposure noise, you can use long exposure NR or take a dark frame at your intended settings with the lens cap on, then subtract that in processing.
--

Yes, I need to set up a dark frame soon and see about applying it to my images (I don't know if DPP or the Gimp allows it...I'm somewhat constrained to free software at the moment).

dcraw will do a dark frame subtraction in raw conversion.

I'm actually using a T1i, aka the 500D. I will try ISO 1600 again, but I don't expect it to work better than ISO 800 in similar lighting because of the sensorgen.info data which predicts it won't work out well - which, as I mentioned before, already seemed to be beyond the limit where images still worked well. If I need brightness I can try getting some of it back afterwards in post.

--
Bob

 

[goetz48]

For good image quality I take a longer exposure time in order to get lower ISO. What are all the stabilization things good for if you (and your camera) ignores them. I often shoot 1/10s freehands in tele range.

 

[CriticalI]

Hello,

What's the point of manually changing ISO when Auto ISO does it for you ?

Aperture is important for DOF control,
Shutter speed is important for non-stab lenses (1/FL) or fast moving subjetcs.
Then you want the lowest possible ISO. And that's exactly what Auto ISO does.

On my Sony NEX, Auto ISO does an even better job than I can do. Because Auto ISO is able to select semi value (250,320,500,640,...) while manual ISO only allows me to select full stop ISO (200,400,800,...).

Don't you always want the lowest possible ISO according to your Aperture and Shutter speed settings ?

Yes, but if I have a zoom lens with stabilisation how does auto-ISO know how low I can go, or if I have a moving subject?

Auto ISO is like an auto-gearbox, fine for shopping, not much use for racing.

--
Regards,
Steve