SPDTDL wrote:
Under the same Phillips hue LED lighting in my house, my Nikon ZR (not a stills camera but the point remains) got heavy banding, my Canon R5II in ES had no such issue.
ZR 14.4ms (according to AI search)
R5II 6.3ms
In fact, I have the R5II set to ES constantly, and have never noticed any issue. The ZR, even with its ability to fine tune shutter speeds to combat the flicker, was not good. I spent more time fiddling with the shutter speeds setting to try and tune out the flicker, and by then I’d missed the shot anyway.
I have lots of hue bulbs. Hue switches at very high frequency so you don’t really see banding.
With 3-4ms readout (A1), if the led flicker period is 2ms - 8ms it will be very easy to see the flicker
?si=kqoiEEzfaNmTXqKZ
In this video I vary the flicker period and display the simulated output.
If you have a led that flickers between 3-14ms you can test that with your R5ii
Regarding stacked sensors and readout rates...Many may not be aware but stacked sensors achieve faster readout rates via two complementary mechanisms...#1 faster on-sensor DRAM, which most know and #2 less known: multiple parallel row readouts, necessary to pull data off the sensor fast enough to feed the faster DRAM.
There are non-stacked sensors that use #2 without #1, for example Canon's R cameras. These achieve faster readout rates vs single-row readouts but obviously slower than fully-stacked sensors with DRAM.
Based on Richard Butler's initial article at DPReview, the A7RVI sensor is stacked but without DRAM, which means it only employs #2, the same as non-stacked sensors. If we assume the A7RVI has a similar SLVS-EC "slow" DRAM interface as their stacked-sensor cameras and same # parallel readouts, we can now estimate the perf gain from stacked DRAM:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A1: 5760 rows / 3.91ms read = 0.678us per row
So stacked DRAM = 4.33x perf gain
I wonder if forgoing stacked DRAM was a cost-cutting move due to the exorbitant price of memory in the market right now, owning to the demand for the higher-margin HB DRAM from AI. Also, there are reports the A7RVI's buffer is much smaller than the A7RV, even discounting for the A7RVI's higher burst rate - the latter might be from a reduction of total DRAM, including the slower, non-stacked DRAM.
Next...if we compare the readout speed of the AR7VI to the partially-stacked A7V:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A7V: 4672 rows / 15.1ms read = 3.23us per row
We see very similar performance, which indicates the current partially-stacked sensors likely omit DRAM as well. Perhaps it's safe to assume the logic on the partially-stacked layer is simply extra readout circuitry to support the additional row readout logic for #2 above. If so, maybe it's also safe to assume the A7RVI requires more of this logic, which is why it needs a fully-stacked wafer.
snapsy wrote:
Regarding stacked sensors and readout rates...Many may not be aware but stacked sensors achieve faster readout rates via two complementary mechanisms...#1 faster on-sensor DRAM, which most know and #2 less known: multiple parallel row readouts, necessary to pull data off the sensor fast enough to feed the faster DRAM.
There are non-stacked sensors that use #2 without #1, for example Canon's R cameras. These achieve faster readout rates vs single-row readouts but obviously slower than fully-stacked sensors with DRAM.
Based on Richard Butler's initial article at DPReview, the A7RVI sensor is stacked but without DRAM, which means it only employs #2, the same as non-stacked sensors. If we assume the A7RVI has a similar SLVS-EC "slow" DRAM interface as their stacked-sensor cameras and same # parallel readouts, we can now estimate the perf gain from stacked DRAM:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A1: 5760 rows / 3.91ms read = 0.678us per row
So stacked DRAM = 4.33x perf gain
I wonder if forgoing stacked DRAM was a cost-cutting move due to the exorbitant price of memory in the market right now, owning to the demand for the higher-margin HB DRAM from AI. Also, there are reports the A7RVI's buffer is much smaller than the A7RV, even discounting for the A7RVI's higher burst rate - the latter might be from a reduction of total DRAM, including the slower, non-stacked DRAM.
Next...if we compare the readout speed of the AR7VI to the partially-stacked A7V:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A7V: 4672 rows / 15.1ms read = 3.23us per row
We see very similar performance, which indicates the current partially-stacked sensors likely omit DRAM as well. Perhaps it's safe to assume the logic on the partially-stacked layer is simply extra readout circuitry to support the additional row readout logic for #2 above. If so, maybe it's also safe to assume the A7RVI requires more of this logic, which is why it needs a fully-stacked wafer. ...Show more →
Is the buffer size different or mostly the number of frames fewer because the ARW files are larger due to image size and noise? The cost of system RAM is not all that much compared to this grade of camera. The a7rV was so slow it did not buffer out much. It seems that the slightly lossy compression and 20FPS would be needed to get a decent burst length. But we needed lossy compression just to get 10FPS mechanical on the a7rV so it's better with the a7rVI anyways.
snapsy wrote:
I wonder if forgoing stacked DRAM was a cost-cutting move due to the exorbitant price of memory in the market right now, owning to the demand for the higher-margin HB DRAM from AI. Also, there are reports the A7RVI's buffer is much smaller than the A7RV, even discounting for the A7RVI's higher burst rate - the latter might be from a reduction of total DRAM, including the slower, non-stacked DRAM.
Here's an interesting counterpoint to that posted in the comments on SAR:
DRAM prices doesn't affect this. Sony will not solder normal DRAM chips to the back of their sensors. They will make custom chips specifically designed for their sensor, with the right connections and everything. Those are most probably made in the same fab as the sensor. Sony doesn't sell RAM so they cannot use their factories to make RAM instead of sensors to make more profit. The sensors probably sell for more than RAM chips anyway. So RAM prices has no impact on memory integrated directly into sensors or CPUs.
I'm keeping my A1. The faster stacked sensor is the main differentiator. I love to shoot silently, and the A1 can do that in 99% of the cases. It's also able to sync flash at 1/200s in e-shutter mode. I think the new A7R VI cannot sync flash in e-shutter mode?
The only things I really miss on the A1 are the pre-capture and speed boost. Both things that could probably be done with a firmware update... I'm longing for a Magic Lantern A1, so much untapped potential here...
May 14, 2026 at 10:32 AM
Steve Spencer Offline • • • • • • • Upload & Sell: On
snapsy wrote:
Regarding stacked sensors and readout rates...Many may not be aware but stacked sensors achieve faster readout rates via two complementary mechanisms...#1 faster on-sensor DRAM, which most know and #2 less known: multiple parallel row readouts, necessary to pull data off the sensor fast enough to feed the faster DRAM.
There are non-stacked sensors that use #2 without #1, for example Canon's R cameras. These achieve faster readout rates vs single-row readouts but obviously slower than fully-stacked sensors with DRAM.
Based on Richard Butler's initial article at DPReview, the A7RVI sensor is stacked but without DRAM, which means it only employs #2, the same as non-stacked sensors. If we assume the A7RVI has a similar SLVS-EC "slow" DRAM interface as their stacked-sensor cameras and same # parallel readouts, we can now estimate the perf gain from stacked DRAM:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A1: 5760 rows / 3.91ms read = 0.678us per row
So stacked DRAM = 4.33x perf gain
I wonder if forgoing stacked DRAM was a cost-cutting move due to the exorbitant price of memory in the market right now, owning to the demand for the higher-margin HB DRAM from AI. Also, there are reports the A7RVI's buffer is much smaller than the A7RV, even discounting for the A7RVI's higher burst rate - the latter might be from a reduction of total DRAM, including the slower, non-stacked DRAM.
Next...if we compare the readout speed of the AR7VI to the partially-stacked A7V:
A7RVI: 6656 rows / 19.6ms read = 2.94us per row
A7V: 4672 rows / 15.1ms read = 3.23us per row
We see very similar performance, which indicates the current partially-stacked sensors likely omit DRAM as well. Perhaps it's safe to assume the logic on the partially-stacked layer is simply extra readout circuitry to support the additional row readout logic for #2 above. If so, maybe it's also safe to assume the A7RVI requires more of this logic, which is why it needs a fully-stacked wafer. ...Show more →
Thanks snapsy. This is very helpful as always.
So it turns out we have four primary types of sensor architecture:
1) Global shutter in which the whole sensor is read out at once. We see this in the Sony A9 III
2) On sensor fast DRAM and multiple parallel row readouts (Your #1 & #2). We see this in the Sony A1/A1 II, A9/A9 II, Nikon Z8/Z9, Canon R1/R3/R5 II.
3) Multiple parallel row readouts (your #2). We see this in the Sony A7 V, A7r VI, Nikon Z6 III, Canon R8, Canon R6 III, and Canon R5.
4) Simply readout the rows one at a time (neither your #1 or #2). We see this in the Sony A7r V, Nikon Z7 II and Z5 II, and the Canon RP.
So, we do seem to have 4 classes of cameras based on four different types of sensor architecture and with this camera Sony has moved the A7r series from the fourth type with nothing to speed up sensor read out to the third type with multiple parallel row readouts to speed out sensor readouts. But that is still quite a bit different from they second type when fast DRAM included on the sensor.
johnvanr wrote:
Oh, sure. I’m not saying these new cameras shouldn’t be produced. I’m just amazed how many people seem to still be in the upgrade path when I don’t see that need anymore. And that goes for every brand I know of, not just Sony.
I’m with you. I still shoot the A7r3. Only camera that really interests me is the next version of the A7cr. I’m hoping for a better viewfinder. I wear glasses and found it hard to use.
I was all set to ignore this round and now I am all gassy.
I have always been more landscape and wildlife and had the A1 and R4 for their roles thinking I would up to the R6 someday and replace the R4. Then I started travelling more and fell in love with small street type work so picked up a Cii which is so fun with a cv 21 fe40g and viltrox 85 small bag fun kit. So the R4 got relegated to just landscapes and just is sitting. Still no reason for me to jump on this right I have 3 of the most amazing bodies in the camera world and everything in an A1ii or R6 is just minor bump. But that damn GAS.
I really am seeing where I should sell the A1 / R4 and just get the R6 to simplify my world back to two cameras then hope that they come out with a C size 9iii global sensor camera to upgrade the Cii with.
But I dont have to.....everything is covered now. Oh my weak will. I hate Sony for keeping me on the edge.
My wife has been using the Sony 24-50mm f/2.8 on an A7Cr, with minor complaints regarding the viewfinder. The A7Rvi will sacrifice a bit of weight and size, but not necessarily a bad thing with the zoom. The fact that APSc mode results in 75mm and 28Mpix makes this quite an appealing combination. Pretty much a one and done travel setup.
SPDTDL wrote:
Under the same Phillips hue LED lighting in my house, my Nikon ZR (not a stills camera but the point remains) got heavy banding, my Canon R5II in ES had no such issue.
ZR 14.4ms (according to AI search)
R5II 6.3ms
In fact, I have the R5II set to ES constantly, and have never noticed any issue. The ZR, even with it’s ability to fine tune shutter speeds to combat the flicker, was not good. I spent more time fiddling with the shutter speeds setting to try and tune out the flicker, and by then I’d missed the shot anyway.
Unless you have GS you are guaranteed to see flicker at the right frequency
?si=T9_r8DWnPfvyalMf
This simulates how the expected banding varies as the led period changes.
To make it simple,
With a 3ms sensor you can see banding at 6ms, but it’s harder and harder to see the banding as the flickering gets slower. Perhaps at 12ms (4x readout) it’s sufficiently hidden
Same thing for all other sensors, if the readout is 6.3ms (R5ii) then it’s hard to see banding if the flicker is slower than 25ms
EB-1 wrote:
Is the buffer size different or mostly the number of frames fewer because the ARW files are larger due to image size and noise? The cost of system RAM is not all that much compared to this grade of camera. The a7rV was so slow it did not buffer out much. It seems that the slightly lossy compression and 20FPS would be needed to get a decent burst length. But we needed lossy compression just to get 10FPS mechanical on the a7rV so it's better with the a7rVI anyways.
EBH
Buffer depth is a topic that's surprising complicated. What may at first seem like a simple matter of image size vs DRAM is really just the first stage of image capture. When an exposure is completed the contents of the sensor are read and dumped into a DRAM buffer. This occurs at the rate of the sensor readout. But before that data can be written to media it has to undergo a long sequence of image processing steps, and it's the performance of these steps which has the large effect on buffer depth.
All cameras can dump the contents of the sensor into a buffer at rates much faster than that data can be processed and made ready to write to media. It's the backlog of images awaiting processing that causes the out-of-buffer condition that slows down an ongoing burst. The speed at which that processing backlog can be handled is a function of the image processing ASIC performance. Sony names their ASIC image processor "BIONZ".
The ASIC has to perform a long sequence of image processing steps, some on the raw data and many more on the resulting JPG from that raw data, which is generated even for raw images (the embedded JPG in raws is used by the camera for the fast image review/playback, as well as by the image viewing/processing tools on the computer in lieu of the raw data being processed by their unique algorithms). Some of these image processing steps can be done in parallel but many have to be sequenced serially. Multiple images can be processed in parallel at various stages, esp on systems that used multiple ASICs (more common in past cameras when ASIC performance was more limited).
Image processing includes raw noise reduction, black level adjustments, massaging data for any sensor errata, optional raw compression, then on to the JPG-specific processing, including debayering, tone curve/profile application, lens corrections, noise reduction, etc.. etc... until the last step, compression.
The total time it takes to fully process a single raw is the latency between the exposure ending and when that image can be written to media. It's also the latency before which that image can be displayed by image review or playback, at least when viewed at full quality with all the image processing completed. Only when the image is fully processed and written to media can the buffer(s) it occupies be recycled and made available for a new incoming sensor frame dump.
Naturally the more DRAM a camera has, the deeper the backlog of images to be processed can be. But since DRAM is expensive, even more so today, it's actually better to make the ASIC faster, since that reduces pressure against the DRAM buffer and allows it to be smaller while maintaining desirable buffer depths.
This is not to say that media card speed has no impact on buffer depth and buffer clearing time...but if you look at the total computational and memory bandwidth consumed during image processing, media performance is a rather some percentage, esp for fast PCIe-based media solutions available today in CFE.
quantumloop wrote:
Here's an interesting counterpoint to that posted in the comments on SAR:
DRAM prices doesn't affect this. Sony will not solder normal DRAM chips to the back of their sensors. They will make custom chips specifically designed for their sensor, with the right connections and everything. Those are most probably made in the same fab as the sensor. Sony doesn't sell RAM so they cannot use their factories to make RAM instead of sensors to make more profit. The sensors probably sell for more than RAM chips anyway. So RAM prices has no impact on memory integrated directly into sensors or CPUs....Show more →
The argument I would make against this is that excess semiconductor wafer production capacity is extremely limited today due to AI demand, so any capacity committed to fabricate DRAM is capacity that has to be stolen from other uses, many of which generate much higher margins, including other types of DRAM itself (for other uses).
I haven't cared about buffer since getting a Z9 and getting Prograde cards. I honestly don't even care about the buffer on my XH2 or 2S. GFX is it's own thing, but nobody shoots that for speed. RoamingScott wrote:
The year is 2026 and we are still worried about buffer? Oof.
Steve Spencer wrote:
Thanks snapsy. This is very helpful as always.
So it turns out we have four primary types of sensor architecture:
1) Global shutter in which the whole sensor is read out at once. We see this in the Sony A9 III
2) On sensor fast DRAM and multiple parallel row readouts (Your #1 & #2). We see this in the Sony A1/A1 II, A9/A9 II, Nikon Z8/Z9, Canon R1/R3/R5 II.
3) Multiple parallel row readouts (your #2). We see this in the Sony A7 V, A7r VI, Nikon Z6 III, Canon R8, Canon R6 III, and Canon R5.
4) Simply readout the rows one at a time (neither your #1 or #2). We see this in the Sony A7r V, Nikon Z7 II and Z5 II, and the Canon RP.
So, we do seem to have 4 classes of cameras based on four different types of sensor architecture and with this camera Sony has moved the A7r series from the fourth type with nothing to speed up sensor read out to the third type with multiple parallel row readouts to speed out sensor readouts. But that is still quite a bit different from they second type when fast DRAM included on the sensor....Show more →
That's a good summary Steve. The only note I'd add is to #1 - you likely already know this but global shutters don't technically read out their contents instantaneously, at least by what I term "read out". They shift their contents instantaneously into a pixel storage node. But that node must then be read out and transferred to DRAM, which uses similar techniques applied to rolling shutters and so has some of the same performance constraints, at least based on my limited knowledge of global shutters.
A minor point but I just wanted it to be clear for anyone else reading.
I have shot heavily futzed around with in post bus/subway ads with that format so it's more than enough for most/all professional needs. arbitrage wrote:
Sadly so. Unless you shoot a Z8/Z9 in HE*.....then fire away!!
I am not sure why people are so bent about the a7r6vi It is an R series camera. It is not primarily a video camera nor is it a high speed camera. If you need high speed capture get an a9iii. If you shoot primarily video buy a video centric camera.
One camera doesn't satisfy all needs. The a7rvi is a truly amazing upgrade, mine is on order to replace my a7riii.
As an aside I get amazing motorsports images from my old slow a7riii, it does great for birds and wildlife as well as landscapes and portraits. I couldn't care less about video, so I don't have a clue about it's video capabilities.
Folks you all need to lighten up, this is going to ba an amazing camera, if it isn't for you then buy the one that is.
PS. I am delighted that the sensor design landed at 66.8mp. 80 or 100 would just be crazy overkill!!!
JadedWriter wrote:
I have shot heavily futzed around with in post bus/subway ads with that format so it's more than enough for most/all professional needs.
HE* is all I ever shot with my Z9s.
Just like with Canon using CRAW and Sony using Lossy Compressed RAW.
I have no issues using those formats to maximize speed and buffer depth.
I don't feel IQ suffers compared to the better RAW formats for what I shoot.
