Slashcam test reveals Sony A7S 1080/60p softer than 24p mode

[maxotics]

The tests done with the a7S seem pretty indisputable to me, that in FACT, the camera has the best high-ISO performance of any camera.  Whether you need it in real-world situations is subjective question.  

 

The question most others are interested in is the tradeoff the camera makes between low noise at high ISO and less DR at base ISO.  All sensors have a base ISO at which they perform optimally.  Everything after that is degraded.  Apparently, and I don't fully understand this myself, higher sensel count cameras improve DR at base ISO.  In other words, the a7S will be better at low-light than the a7, but the a7 will have better DR at base ISO (say 100).  

 

It is not the low megapixel count that makes the a7S more sensitive to light, it is the LARGER sensel/pixel size on the sensor.  Each sensel is like a little telescope/radio dish with a colored filter in front of it.  As in astronomy, you can collect radiation through lots of little dishes, or one big dish.  In picking up faint objects, the bigger the better.  In short, larger pixel, less noise, period.  There is nothing temporary in this area of physics.  I do not see TV stations replacing their satellite dishes with 50 little dishes.

 

Again, it is NOT about the low megapixel count.  It is about the increased pixel size (more sensitivity/less noise) of the a7S that makes it unique.    Here is someone's calculation of pixel width

 

Approximate pixel pitch  (in microns)

 

Refer to the reservations  here  about calculating the "true" width and area of an individual pixel.

 

Pixel pitch in microns  = width of sensor in millimetres  divided  by  image width in pixels  multiplied by 1000

 

A77   =   3.917    (23.5 / 6000  x 1000)

A7S =     8.443    (35.8 / 4240 x 1000)

 

Relationship: A7S is approximately 116% greater than A77

 

http://www.robsphotography.co.nz/Sony-A77-SonyA7S.html

 

Would you rather have a 4 inch telescope or 8 inch telescope when peering into the night sky? 

[maxotics]

We've always had many s35 sensors with 2 megapixels (all 1080p video cameras) and we don't see these native, extremely low megapixel count-video cameras being much more sensitive than photographic sensor, in fact it's almost quite always the opposite

 

 

Hi Ebraham, can you be specific here?  This is quite a claim.  Are you sure those sensors aren't actually larger resolution sensors where the maker is only using 2 megapixels from, say 5, on-sensor pixel density?

[Andrew Reid]

Hmm. The FS100 had a S35 sensor with 2 megapixel and the low light performance of that was way better than a high resolution photographic sensor of the same size, i.e. 7D, even taking into account the downsampling effect on lessening noise per pixel.

 

You have to also take into account the architecture of the sensor, for example whether global shutter takes up room on the front side of the sensor that otherwise could be used to gather light. In case of BM Production Camera that made an 8MP APS-C sensor far less sensitive than it otherwise would have been. To only take into account sensor size and megapixel count is too simplistic.

 

For video a resolution matched as closely to the native res (2K or 4K) is beneficial.

 

4K actually only needs 8MP, and 12MP on the A7S can go lower still.

Wait till you see the Sony curved CMOS with purpose built fast prime and 8MP 4K video... Low light on that will be another game changer and blow the A7S out of the water.

 

It's a moving target, constantly.

 

A lower resolution sensor holds a massive advantage for video. You can downscale a 36MP still to 2MP and say "look how clean" but pixel quality is pixel quality, signal to noise is signal to noise, there's far fewer ways to fudge the facts when it comes to video.

[horshack]

The question most others are interested in is the tradeoff the camera makes between low noise at high ISO and less DR at base ISO.  All sensors have a base ISO at which they perform optimally.  Everything after that is degraded.  Apparently, and I don't fully understand this myself, higher sensel count cameras improve DR at base ISO.  In other words, the a7S will be better at low-light than the a7, but the a7 will have better DR at base ISO (say 100).  

 

It is not the low megapixel count that makes the a7S more sensitive to light, it is the LARGER sensel/pixel size on the sensor.  Each sensel is like a little telescope/radio dish with a colored filter in front of it.  As in astronomy, you can collect radiation through lots of little dishes, or one big dish.  In picking up faint objects, the bigger the better.  In short, larger pixel, less noise, period.  There is nothing temporary in this area of physics.  I do not see TV stations replacing their satellite dishes with 50 little dishes.

 

Again, it is NOT about the low megapixel count.  It is about the increased pixel size (more sensitivity/less noise) of the a7S that makes it unique.    Here is someone's calculation of pixel width

 

Approximate pixel pitch  (in microns)

 

Refer to the reservations  here  about calculating the "true" width and area of an individual pixel.

 

Pixel pitch in microns  = width of sensor in millimetres  divided  by  image width in pixels  multiplied by 1000

 

A77   =   3.917    (23.5 / 6000  x 1000)

A7S =     8.443    (35.8 / 4240 x 1000)

 

Relationship: A7S is approximately 116% greater than A77

 

http://www.robsphotography.co.nz/Sony-A77-SonyA7S.html

 

Would you rather have a 4 inch telescope or 8 inch telescope when peering into the night sky? 

 

 

Larger pixels are not more sensitive to light - the A7s's sensitivity ("quantum efficiency") is about equal to that of the D800/A7 - this is why midtone noise on the A7s is the same as the D800/A7r when compared on a per-area basis (via downsampling). What larger pixels do provide, at least with current technology limits, is better Higher ISO dynamic range, which is due to the total lower cumulative read noise from fewer pixels having to be read over the same area vs a higher MP sensor. This translates to lower noise in the shadows. This larger pixel advantage turns into a disadvantage at base ISO because of the read noise penalty from holding a larger charge (full-well capacity). You can read more details at my dpreview thread here: http://***URL removed***/forums/post/53860364

[freiheit]

here's my test with a7S's APS-C mode. I've testing a7S with Sony Japan few month back and seems 1080P60fps has different sensor readout and it makes softer, moire-oriented image.

[andrgl]

Larger pixels are not more sensitive to light

"Without compensating technologies, smaller pixels have lower dynamic range, lower fill factor, worse low light sensitivity, higher dark signal, and higher non-uniformity."

Source: Standford Research Paper

[horshack]

"Without compensating technologies, smaller pixels have lower dynamic range, lower fill factor, worse low light sensitivity, higher dark signal, and higher non-uniformity."

Source: Standford Research Paper

 

At the individual pixel level they do. But pixel-level metrics aren't relevant since images are rendered by area irrespective of the number of pixels that area contains. On a per-area basis the best technology high MP sensor (D800/A7r) has the same or better quantum efficiency as the best technology lower MP sensor (A7s/D4/D4s/Df/D3s/1DX), and have better base ISO dynamic range.

[maxotics]

Larger pixels are not more sensitive to light - the A7s's sensitivity ("quantum efficiency") is about equal to that of the D800/A7 - this is why midtone noise on the A7s is the same as the D800/A7r when compared on a per-area basis (via downsampling). What larger pixels do provide, at least with current technology limits, is better Higher ISO dynamic range, which is due to the total lower cumulative read noise from fewer pixels having to be read over the same area vs a higher MP sensor. This translates to lower noise in the shadows. This larger pixel advantage turns into a disadvantage at base ISO because of the read noise penalty from holding a larger charge (full-well capacity). You can read more details at my dpreview thread here: http://***URL removed***/forums/post/53860364

 

HI horshack, I did read that thread but had difficulty understanding.  It seems misleading when you say, "cumulative read noise".  PLEASE correct me.  I would think that on a set of 4 pixels (pre de-bayering), you would get more sensitivity at high ISO, and less noise, then a similar block of 16 pixels fit into the same space.  I can see how you could say the cumulative noise is less on the 4 pixels, then 16 pixels, but isn't that obscured the prime difference between different size pixels?  I can understand that experimentally, it makes sense to look at cumulative read noise, but in explaining the difference to filmmakers, isn't it better that they think in pixels?

 

On the base ISO issue, it seems that comparing 4 pixels against 16, or 1 against 4, allows one to visualize the difference between say having a dish microphones vs an omnidirectional.   

[horshack]

HI horshack, I did read that thread but had difficulty understanding.  It seems misleading when you say, "cumulative read noise".  PLEASE correct me.  I would think that on a set of 4 pixels (pre de-bayering), you would get more sensitivity at high ISO, and less noise, then a similar block of 16 pixels fit into the same space.  I can see how you could say the cumulative noise is less on the 4 pixels, then 16 pixels, but isn't that obscured the prime difference between different size pixels?  I can understand that experimentally, it makes sense to look at cumulative read noise, but in explaining the difference to filmmakers, isn't it better that they think in pixels?

 

On the base ISO issue, it seems that comparing 4 pixels against 16, or 1 against 4, allows one to visualize the difference between say having a dish microphones vs an omnidirectional.   

 

The typical analogy used is buckets of water. Four small 8oz buckets placed side-by-side hold the same amount of water as a single 32oz bucket. Since the smaller buckets are side-by-side you would expect some loss of water when pouring into them for the gaps in between the buckets; this is where sensor microlenses come in. They act as a funnel to keep water from spilling around the sides of the extra boundaries/edges of the smaller buckets (ie, the funnel light into the photosensitive portion of each pixel, preventing the light from reflecting off the additional edges of smaller pixels). You'd be surprised to learn that the highest-efficiency sensors available right now for any commercial camera are tiny P&S cameras, where they convert 75% of the light they receive, vs 56% for the best full-frame sensors.

[maxotics]

The typical analogy used is buckets of water. Four small 8oz buckets placed side-by-side hold the same amount of water as a single 32oz bucket. Since the smaller buckets are side-by-side you would expect some loss of water when pouring into them for the gaps in between the buckets; this is where sensor microlenses come in. They act as a funnel to keep water from spilling around the sides of the extra boundaries/edges of the smaller buckets (ie, the funnel light into the photosensitive portion of each pixel).

 

I would think another significant problem would be the decreased efficiency of the micro-lenses in front of each pixel.  The larger the micro-lens the more efficient?  Or, the smaller the micro-lens, the less efficiency, especially in low light.  

[maxotics]

Hmm. The FS100 had a S35 sensor with 2 megapixel and the low light performance of that was way better than a high resolution photographic sensor of the same size, i.e. 7D, even taking into account the downsampling effect on lessening noise per pixel.

 

 

Interesting. It looks like you can get an FS100 used for the same price as an a7S.  Thanks!

[horshack]

A lower resolution sensor holds a massive advantage for video. You can downscale a 36MP still to 2MP and say "look how clean" but pixel quality is pixel quality, signal to noise is signal to noise, there's far fewer ways to fudge the facts when it comes to video.

 

That's only because sensors and sensor procesisng ASICs aren't yet fast enough to process higher MP streams for video. It's not a pixel quality issue since those same high MP sensor produce equal or better IQ than their lower MP bretheren for stills, for everything except the very extreme High ISOs (+12,800 for FF).

[horshack]

I would think another significant problem would be the decreased efficiency of the micro-lenses in front of each pixel.  The larger the micro-lens the more efficient?  Or, the smaller the micro-lens, the less efficiency, especially in low light.  

 

One issue with the microlenses is that they have an associated max aperture, like normal lenses. Producing very large aperture microlenses is difficult and expensive apparently, so commercial sensors have to compromise a bit. The downside of this compromise is loss of light when using large aperture lenses, mostly at f/1.4 and larger. You can read it about it here: http://www.dxomark.com/Reviews/F-stop-blues

[mjfan]

the article is saying that from 24-60 fps in aps-c mode it is still the same level of sharpness correct?

[Michael1]

The downside of this compromise is loss of light when using large aperture lenses, mostly at f/1.4 and larger. You can read it about it here: http://www.dxomark.com/Reviews/F-stop-blues

What I am not understanding from the article is it implies that at large apertures, the camera is automatically compensating by increasing the ISO, but the user is unaware of the ISO increase? 

 

 

“Accordingly, the photographer has no way to detect that he/she has not benefited from the increase of light transmitted. Of course this increase in ISO translates into other downfalls – mainly in accrued noise. But these remain unknown to the operator,

 

So the camera says, for instance, ISO 1600, and the camera is really at ISO 3200?

 

Michael

[horshack]

What I am not understanding from the article is it implies that at large apertures, the camera is automatically compensating by increasing the ISO, but the user is unaware of the ISO increase? 

 

 

So the camera says, for instance, ISO 1600, and the camera is really at ISO 3200?

 

Michael

 

Yes, on a noise basis that is effectively what happens. On an exposure basis it matches the user-indicated ISO.