fredmiranda.com Login
This topic is in the 2024 archive. You can read it and like its posts here. It is read only.

Archive 2024Sony A9III CFE-a Memory Card Performance Data

  
 
dclark
Online
Details
p.5 #1Sony A9III CFE-a Memory Card Performance Data
snapsy wrote:
dclark wrote
snapsy wrote
dclark wroteOne point that I think is ignored is that some of the curves have two break points, which would be characteristic of a simple two sequential buffer model. It is also clear that sometimes the second buffer causes a break and sometimes it does not, even though in a simple model it should. The cameras seem to change modes of operation.
When the data doesn't fit the conclusion then it's best to reconsider the conclusion.
Conclusion? What conclusion is in conflict with what data?
You wrote the data indicates a "simple two sequential buffer model", then commented how it sometimes causes a break and sometimes does not, then commented how the model indicates it should. The data then does not fit the conclusion about the model.

I did not say that a simple two sequential buffer model was "indicated". My comment was to the point that a simple model does not work! "sometimes it does not, even though in a simple model it should".

I have not drawn any conclusions. In fact I have been consistently skeptical that anyone has offered a model that explains all the observations and have been more than skeptical about people that take memory card specs as an indication of how the camera will perform. I believe I have been consistent in stating that I only believe the data, not the theories. I believe the comments by @Daran and @j4nu have been interesting and reasonable and they have been appropriately hedged.

This debate is not interesting or useful. I am going to Torrey Pines to see what the Peregrines are doing.



May 17, 2024 at 11:10 AM
snapsy
Offline
Details
p.5 #2Sony A9III CFE-a Memory Card Performance Data
The publicly available sensor datasheets provide all the information one needs to establish a good working model of how this all works. When formulating theories it's best to start with the available documentation and work backwards rather than speculating from data observations alone.

For example, here's a screenshot of the datasheet for the stacked sensor that likely shares the same design as the A9:

Sony stacked-sensor datasheet

"On-Chip DRAM function is available by readout data into DRAM using fast 12-parallel ADC prior to output still pictures via SLVS-EC to reduce readout-time of each frame"

To decode that, the data is read off the sensor in parallel 12-rows at a time and deposited into the fast stacked memory colocated on the sensor complex. There is no logic on the sensor to perform any of the image processing steps that have been speculated in this thread. All those operations are the sole domain of the image ASIC, which for Sony cameras is named BIONZ. Since the datasheet clearly describes how the data is deposited into memory, the subsequent image processing steps performed are then performed on that memory. That memory area is the "buffer".
May 17, 2024 at 11:23 AM
nhmorgan
Offline
Details
p.5 #3Sony A9III CFE-a Memory Card Performance Data
Fascinating thread that I'd missed before I started a new post. All I can add is that I found in trying to do panning shots with the a9iii that it can actually hit 30fps with a 1/30s shutter speed provided that the camera is set to 60fps or 120fps. If the camera is just set to 30fps it gets around 15fps at 1/30s.

This is a solution to an issue I ran into with the Z8/Z9 when I found that those cameras couldn't hit 20fps until 1/250s shutter. Similarly, when I shot the A1 I found that it also hit around 15fps at 1/30s. The a9iii also hits this limitation unless the camera is in H+. Then, it will hit as many frames as the shutter speed will allow up to 120fps.
Jun 18, 2024 at 08:57 AM
Hodie
Offline
Details
p.5 #4Sony A9III CFE-a Memory Card Performance Data
UKBob wrote:
Shuttermuse has an interesting article about the best card for an A9iii and he recommends one with the highest sustained write speed ie the ProDigital Gold Type A card with a sustained write speed of 600mb/secBest memory cards for the A9iii -I was wrong
Can I ask what capacity cards would be recommended for an exciting day of wildlife action and how do you configure slots 1 and 2
For the A1 I use 160gb Sony card in Slot 1 with compressed RAW and 80gb in slot 2 for JPEG as back up which gives about 2200 images for each slot
...Show more →

I just read his article and decided to give the ProGrade Gold 480gb cards a go. It's not cheap, but when you start looking at the smaller capacity cards, you're paying way more per gb.

Pergear has a 520gb card that has--according to them--sustained write speeds up to 900mb/s, but the min write speed is stated at 400mb/s, which seems like a big range. But they're pretty cheap. I may try them for the hell of it.

However, to your point, on my A1, I used 160gb and 260gb cards and never ran out of space on a day of shooting. Granted, I'm usually not actually shooting all day (I wish I could) and I'm pretty selective on what I'm shooting in bursts.

I'm guessing the A9iii is a totally different ball game when the key feature is all about speed. I'll curios to know if my shooting style/habits change with the A9iii, which I'm waiting delivery on.
Jul 22, 2024 at 09:07 PM
tctmp
Offline
Details
p.5 #5Sony A9III CFE-a Memory Card Performance Data
dclark wrote:
This data raises some interesting questions about how the sensor/camera/memory electronics works. The most obvious question is the one raised by @randomguy@, why does the buffer not fill and slow the frame rate to 5 FPS when running at 30 FPS or slower? In order to make this more clear I have added a couple of lines to the frames vs time chart. It seems that the buffer that can hold ~236 frames empties at a rate of 5 FPS when data is coming in at a rate of 120 FPS or 60 FPS, but empties at a rate of
...Show more →

I still can't wrap my head around why a setting of 120fps will result in a 5fps steady state while a setting of 30fps will result in a 13fps steady state. If the guess is that 120fps setting means the sensor is scanned and maybe even processed at that rate no matter how full the buffer is, and only 5fps is saved to the card, (which means 115fps are eventually discarded), that seems to be serious firmware bug. Seems to me the correct way of doing things should be to not scan/process the sensor data (other than for EVF purpose) until there is room to move the data off the sensor.
Jul 22, 2024 at 11:29 PM
aCuria
Offline
Details
p.5 #6Sony A9III CFE-a Memory Card Performance Data

dclark wrote:
Stereodude wrote
dclark wrote@Daran@@@@@ has provided a partial answer to your comment.

I did not compute the time to clear the buffer or the number of frames in the buffer when clearing is started, I measured it. I record the video from the EVF and then observe the number of frames displayed to determine how many frames are in the buffer when it starts to clear (it was always 236) and then measure the time it takes to clear the buffer by noting the time and frame number when the clearing starts and then moving forward to the frame when the buffer has completed clearing and note the time and frame number (the video analysis is done using Davinci Resolve). Consequently time difference can be measured accurate to 1/30 sec (the frame rate of the video). That time was 11.1 sec for all the cards.

The graph of frame vs time can be used to determine the buffer size by projecting the line back to t=0, as is shown on the chart. The number of frames that have already been written to the card is the difference in frames at the time the buffer fills and the projection back to time zero.
Thank you for the explanation. There must be something more complicated going on than a simple write buffer for the memory card at play if the shooting rate drops to 5fps after hitting 236 frames since it can write to the card faster than 125MB/sec. Like they're mitigating heat buildup in one of the components. As I mentioned in my other post the A7IV can shoot indefinitely at 6fps writing ~422MB/sec (6*70.3MB) to a sufficiently fast CFe-A card. It seems hard to believe that A9III can only manage 125MB/sec especially since the write speed while shooting changes based on the burst shooting frame rate prior.
There is only one processor. If it's busy responding to interrupts from data coming in, it does not have much time to transfer frames from the buffer to the card. It seems that when the buffer is full the sensor electronics is still trying to send files at the full rate of 120/60fps and consequently occupying the processor with frames that get dropped. That may also be part of the explanation why the frame intervals are so non-uniform.


For desktop processors, since 1998 we have had DMA which allows data transfer to happen without disturbing the CPU.

Idk what’s used on the camera, but it’s pretty lame not having DMA in 2024

https://en.m.wikipedia.org/wiki/Direct_memory_access
Nov 08, 2024 at 01:18 AM
Tony Ross
Offline
Details
p.5 #7Sony A9III CFE-a Memory Card Performance Data
snapsy wrote:
dclark wrote
snapsy wrote
dclark wrote
Daran wrote
snapsy wroteFully agree, which is why I questioned how @Daran@@@@@@ arrived at his conclusion of "A1 is pretty much done processing after pushing the image into the buffer."
Which processing do you think is not done right away when shooting lossy compressed?

I'll note that:
- the buffer size in terms of fitting image count is specified to be smaller when shooting compressed RAW+JPEG then when just shooting compressed RAW, so the JPEG data is not simply generated from the RAW at the output writing stage
- even compressed RAW image capture requires a JPEG version (be that the embedded preview or a separate JPEG)
- most JPEG affecting processing is also required by live view
- live view processing may be done on a lower resolution, but it certainly can not be delayed by buffering
- storing uncompressed data has identical computational requirements, except the lossy encoding, which we know to happen in hardware when feeding into the buffer
- storing uncompressed data from the buffer happens at about half the speed of lossy compressed data, which is a decent fit to the relative file sizes

Hence my guess that for lossy RAWs very little happens to the data while the buffer is cleared.

PS: One opposing data point is that shooting JPEG only does not result in an effectively endless buffer.
---------------------------------------------

snapsy wrote
Daran wroteWhich processing do you think is not done right away when shooting lossy compressed?

I'll note that:
- the buffer size in terms of fitting image count is specified to be smaller when shooting compressed RAW+JPEG then when just shooting compressed RAW, so the JPEG data is not simply generated from the RAW at the output writing stage
- even compressed RAW image capture requires a JPEG version (be that the embedded preview or a separate JPEG)
- most JPEG affecting processing is also required by live view
- live view processing may be done on a lower resolution, but it certainly can not be delayed by buffering
- storing uncompressed data has identical computational requirements, except the lossy encoding, which we know to happen in hardware when feeding into the buffer
- storing uncompressed data from the buffer happens at about half the speed of lossy compressed data, which is a decent fit to the relative file sizes

Hence my guess that for lossy RAWs very little happens to the data while the buffer is cleared.

PS: One opposing data point is that shooting JPEG only does not result in an effectively endless buffer.
I don't quite follow what position you're taking on this. Do you believe all the processing is done the instant the data is deposited into a buffer from the sensor? Or are you arguing the opposite?
---------------------------------------------

j4nu wroteYes, it wouldn't make sense IMHO to compress the data into the buffer, then decompress it to do some processing and compress it again onto the card...

Daran wroteWhich processing do you think is not done right away when shooting lossy compressed?

I'll note that:
- the buffer size in terms of fitting image count is specified to be smaller when shooting compressed RAW+JPEG then when just shooting compressed RAW, so the JPEG data is not simply generated from the RAW at the output writing stage
- even compressed RAW image capture requires a JPEG version (be that the embedded preview or a separate JPEG)
- most JPEG affecting processing is also required by live view
- live view processing may be done on a lower resolution, but it certainly can not be delayed by buffering
- storing uncompressed data has identical computational requirements, except the lossy encoding, which we know to happen in hardware when feeding into the buffer
- storing uncompressed data from the buffer happens at about half the speed of lossy compressed data, which is a decent fit to the relative file sizes

Hence my guess that for lossy RAWs very little happens to the data while the buffer is cleared.

PS: One opposing data point is that shooting JPEG only does not result in an effectively endless buffer.
Daran is quite clear that he believes that most of the processing has been completed when the frame is put into the buffer.

I added some data to the thread about the A1, A1 Performance, that is meant to provide a bit more information about the processing and JPEGs. It is not clear to me whether it is useful.

One point that I think is ignored is that some of the curves have two break points, which would be characteristic of a simple two sequential buffer model. It is also clear that sometimes the second buffer causes a break and sometimes it does not, even though in a simple model it should. The cameras seem to change modes of operation.
How can the processing be complete at the time it's read off the sensor and put into the buffer? That implies the long pipeline of image processing operations performed on the data are completed in real-time.
It is clear there are a series of processing steps that need to be completed at some point along the path from the sensor to the buffer(s) to the memory card. The question is what is done before the buffer(s) and what is done after the buffers. All the processing and data movement is "real time".
The data is moved off the sensor and into a buffer. The image processing steps are then performed on that buffer via the imaging ASIC. When those steps are complete the buffer is written to media and then recycled to be used for another frame.


Different buffers.

Image is read from the sensor into a very fast buffer where (some) processing is done. This buffer might hold multiple frames with a pipeline moving the image along. When finished it goes into a much bigger buffer that holds frames waiting to be written to the card. The much bigger buffer could be slower RAM than the one/s used for image processing.
Nov 08, 2024 at 03:17 AM
snapsy
Offline
Details
p.5 #8Sony A9III CFE-a Memory Card Performance Data
Tony Ross wrote:
snapsy wrote
dclark wrote
snapsy wrote
dclark wrote
Daran wroteWhich processing do you think is not done right away when shooting lossy compressed?

I'll note that:
- the buffer size in terms of fitting image count is specified to be smaller when shooting compressed RAW+JPEG then when just shooting compressed RAW, so the JPEG data is not simply generated from the RAW at the output writing stage
- even compressed RAW image capture requires a JPEG version (be that the embedded preview or a separate JPEG)
- most JPEG affecting processing is also required by live view
- live view processing may be done on a lower resolution, but it certainly can not be delayed by buffering
- storing uncompressed data has identical computational requirements, except the lossy encoding, which we know to happen in hardware when feeding into the buffer
- storing uncompressed data from the buffer happens at about half the speed of lossy compressed data, which is a decent fit to the relative file sizes

Hence my guess that for lossy RAWs very little happens to the data while the buffer is cleared.

PS: One opposing data point is that shooting JPEG only does not result in an effectively endless buffer.
---------------------------------------------

snapsy wroteI don't quite follow what position you're taking on this. Do you believe all the processing is done the instant the data is deposited into a buffer from the sensor? Or are you arguing the opposite?
---------------------------------------------

j4nu wroteYes, it wouldn't make sense IMHO to compress the data into the buffer, then decompress it to do some processing and compress it again onto the card...
Daran is quite clear that he believes that most of the processing has been completed when the frame is put into the buffer.

I added some data to the thread about the A1, A1 Performance, that is meant to provide a bit more information about the processing and JPEGs. It is not clear to me whether it is useful.

One point that I think is ignored is that some of the curves have two break points, which would be characteristic of a simple two sequential buffer model. It is also clear that sometimes the second buffer causes a break and sometimes it does not, even though in a simple model it should. The cameras seem to change modes of operation.
How can the processing be complete at the time it's read off the sensor and put into the buffer? That implies the long pipeline of image processing operations performed on the data are completed in real-time.
It is clear there are a series of processing steps that need to be completed at some point along the path from the sensor to the buffer(s) to the memory card. The question is what is done before the buffer(s) and what is done after the buffers. All the processing and data movement is "real time".
The data is moved off the sensor and into a buffer. The image processing steps are then performed on that buffer via the imaging ASIC. When those steps are complete the buffer is written to media and then recycled to be used for another frame.
Different buffers.

Image is read from the sensor into a very fast buffer where (some) processing is done. This buffer might hold multiple frames with a pipeline moving the image along. When finished it goes into a much bigger buffer that holds frames waiting to be written to the card. The much bigger buffer could be slower RAM than the one/s used for image processing.


Covered here:

https://www.fredmiranda.com/forum/topic/1856860/4#16549795
Nov 08, 2024 at 04:17 AM
aCuria
Offline
Details
p.5 #9Sony A9III CFE-a Memory Card Performance Data
jhapeman wrote:
arbitrage wrote
dclark wroteSince the Z8 buffer size is the same, 21 Frames, for all RAW types, that seems to indicate the Nikon stores unprocessed RAW data in the buffer. Sony seems to compress the data before storing in the buffer for the lossy compressed and stores unprocessed RAW data for lossless compressed and uncompressed. Nikon Z8 lossless compressed is about the same file size Sony A1 uncompressed and they both hit the buffer at ~80 frames. Sony lossless compressed is smaller and hits the buffer at ~105 frames.

It is not documented by Nikon how fast it captures frames once the buffer is filled, so one would need to make some assumptions to get an estimate. Also there is no data I see for how fast the buffer is cleared once capture stops. It's a small buffer so I assume it should be very quick.

If I had a Z8 I would make the measurements (it's easy), although I think it's buffer limits are probably not operationally important to the way I operate a camera. At 20fps the same is true for the A1 or the A9III. At 30fps, which the Z8 cannot do, the Sony A1 is somewhat limiting and the A9III limit is beyond anything I require. I believe the main operational lesson of this data for the A9III is that I would normally use 120fps or 60fps only in boost. There may be a few situations where operating at 120 or 60 fps in conjunction with the pre-capture may be useful. Others may have different methods and will reach other conclusions, but I believe all the information needed for such decisions is presented.

Anyone using any of these cameras does not have much of a basis to blame the camera if you're not getting the shot. IMO, the weak link is the human attached to the finger that is pushing the shutter button.
I found using 120FPS on the A9III to be severely limiting. You need to have a very specific goal and strictly follow it. My example is I wanted to try 120FPS with precapture on an osprey entering the water. Just to try and get that perfect moment as the talons just touch the water without even breaking the surface. But old habits die hard and on my first couple attempts I fired during the dive and the buffer was done and when the osprey hit, precapture was probably giving me 5FPS as you show in the data.

I found that 60FPS with precapture was useable as a general setting most of the time. My preferred mode was 30FPS with precapture on all the time. That was a good middle ground to not be running into buffer issues.
120fps even with a bigger buffer is a tough use case because of the sheer number of photos you have to slog through. I mostly use my A9III at 30fps with precapture. I used 120fps when I was trying to capture hummingbirds in flight or similar things where the incredibly fast action also calls for an incredibly fast frame rate but you most certainly have to train your brain and finger to work differently in these cases.

For something like your example of the osprey, hitting the shutter right at the moment it is hitting the water is probably ideal
...Show more →

The camera does let you lock the frames you want, then delete the rest of the burst.

Jun 25, 2025 at 11:16 PM
aCuria
Offline
Details
p.5 #10Sony A9III CFE-a Memory Card Performance Data
Hodie wrote:
UKBob wroteShuttermuse has an interesting article about the best card for an A9iii and he recommends one with the highest sustained write speed ie the ProDigital Gold Type A card with a sustained write speed of 600mb/secBest memory cards for the A9iii -I was wrong
Can I ask what capacity cards would be recommended for an exciting day of wildlife action and how do you configure slots 1 and 2
For the A1 I use 160gb Sony card in Slot 1 with compressed RAW and 80gb in slot 2 for JPEG as back up which gives about 2200 images for each slot However I sort through the images each day and have never run out of space on the 160gb card with my shooting style which is short bursts
I just read his article and decided to give the ProGrade Gold 480gb cards a go. It's not cheap, but when you start looking at the smaller capacity cards, you're paying way more per gb.

Pergear has a 520gb card that has--according to them--sustained write speeds up to 900mb/s, but the min write speed is stated at 400mb/s, which seems like a big range. But they're pretty cheap. I may try them for the hell of it.

However, to your point, on my A1, I used 160gb and 260gb cards and never ran out of space on a day of shooting. Granted,
...Show more →

CF-A gen 3 can theoretically write at 1000MB/s. We may benefit from cards capable of 1000MB/s sustained, which the gen4 cards can do.
Jun 26, 2025 at 04:43 AM
aCuria
Offline
Details
p.5 #11Sony A9III CFE-a Memory Card Performance Data
dclark wrote:
TimMunsey wroteWith the camera determining performance we can conclude buying a cfexpress type 4 card will offer no in camera performance boost?
In order for a CFE-a type 4 memory card (i.e. a card with a PCIe-4 data lane) to provide any improvement, the camera would need to have PCIe-4 electronics, not PCIE-3 electronics.
So far as we know there are no Sony cameras with a PCIe-4 data lane, but we can't be certain about the A9III.
Someone needs to put a type 4 card in an A9III and make the measurements.
I will be surprised if it is faster, but pleasantly surprised.


Perhaps, the NAND on a type 4 card could be faster?.

CF Express 2.0 is supposed to support 1GB/s transfers, but according to your tests we are not getting that.

https://compactflash.org/wp-content/uploads/2025/06/CFexpress-Form-Factors-Website.png
Jul 03, 2025 at 12:54 AM
dclark
Online
Details
p.5 #12Sony A9III CFE-a Memory Card Performance Data
aCuria wrote:
dclark wrote
TimMunsey wroteWith the camera determining performance we can conclude buying a cfexpress type 4 card will offer no in camera performance boost?
In order for a CFE-a type 4 memory card (i.e. a card with a PCIe-4 data lane) to provide any improvement, the camera would need to have PCIe-4 electronics, not PCIE-3 electronics.
So far as we know there are no Sony cameras with a PCIe-4 data lane, but we can't be certain about the A9III.
Someone needs to put a type 4 card in an A9III and make the measurements.
I will be surprised if it is faster, but pleasantly surprised.
Perhaps, the NAND on a type 4 card could be faster?.

CF Express 2.0 is supposed to support 1GB/s transfers, but according to your tests we are not getting that.



The in-camera performance is limited by the camera electronics, not the card electronics.

The reports you see that show in-camera differences are from people who know how to generate clicks but don't know how to make the measurement.

Card readers may show considerably higher speeds.
Jul 03, 2025 at 09:32 AM
FM ForumsSony ForumJoin Upload & Sell
  
 

You are not logged inLog inRegister