TL;DR - The P3 camera (pro in my testing) has a maximum write speed to the mSD card of 11.5MB/s - even if the card is much faster. This negatively affects multishot photos (AEB and rapid fire) because the camera locks up for 20+ seconds while the imges write to the card. Is the internal bus only capable of 11.5MB/s?

Discussing photography only, every class 10 card is capable of recording 4K/UHD at roughly 7.5MB/s. I searched through the forum and could not find any specific threads on this issue.

Why is this important? Because the camera locks up (blue spinning circle on screen) while writing out the buffer to the card after AEB or rapid fire multiple shots for 20-30 seconds (4-5 seconds per picture shooting J+R), which severely impacts the ability to get follow up pictures. I started testing faster cards (The Lexar 16GB 633X card that came with the P3P tested at 24MB/s sequential write up to he SanDisk ExtremePlus that tested at 64MB/s sequential write) and discovered that it didn't matter how fast the card was at writing, there is an internal limitation in the P3 camera that caps the data transfer at roughly 11.5MB/s (8.5MB/s to 11.5MB/s depending on how measured).

Is the camera not capable of using the UHS-I protocol (95MB/s)?

The good news - taking 5AEB photos of a digital stopwatch showed an average separation of .06 seconds (60miliseconds) between the capture of the photos, so the camera is getting the data into it's internal buffer quickly, so the limitation is in transfering the pictures through the image processor and writing to the card.

Is it known what is causing the bottleneck?

Hi-ho,  

Not sure what the bottle neck is, but I can confirm it's a real pain...   

I had thought it was downloading the images to the tablet, but after a bit of testing it does appear it's something onboard that's causing the problem.

Using a P3 Advanced here, not the pro.

Well that saves me a job as I was planning on trying a different card this weekend. Having not had a P2 Vision I was wondering if the speed writing to card was an issue there as I have previously only used a Go Pro.

Yep that is the same conclusions I arrived some time ago regarding the time lapse (lowest interval being 5 seconds), it really looks there is something wrong in the way the images are written, but the interesting part is that 4K video streams at 60Mbps which means that the card should be able to handle the pictures faster than it actually does...

isaache Posted at 2015-6-9 17:20
Yep that is the same conclusions I arrived some time ago regarding the time lapse (lowest interval b ...

TL;DR - There appears to be a bug in the camera firmware that restricts to the SD controller to the slowest speed (12.5MB/s) and does not enable UHS-1 (U1) transfer speeds (up to 104MB/s bus. This is bug that strongly affects photographers trying to capture rapid fire or AEB J+R photographs that generate 150-200MB of data in a few hundred milliseconds.

Based upon https://www.sdcard.org/consumers/speed/bus_speed/index.html - my best guess is that the CD controller on the camera is only capable of "normal speed" data transfer, which is limited to 12.5MB/s.

Because 12.5MB/s is sufficient for 4K/UHD video at 60Mbps (roughly 7.5MB/s) I speculate that the DJI engineers did not even try to enable "High Speed" (25MB/s) from SD card spec v2.0. Given that UHS-I (SD spec version 3.01) was introduced in 2009, I find it hard to believe that DJI was even able to source a SD controller that is not capable of UHS-I or even just High Speed, but without knowing exactly what chips are in the camera it's impossible to know. The SDXC support includes High Speed and the inclusion of U1 in the DJI camera specifications suggests that the controller is UHS capable of at least UHS-I SDR50 (50MB/s bus), if not SDR104 (104MB/s bus).

Combined these strongly suggest that the bus speed is being limited to 12.5MB/s by the firmware programming and likely ignored because the bus speed limitation only affects photographers, not video. Hopefully we can get the attention of the DJI tech support and have support for High Speed and UHS-I added as a feature request or bug report.

There is another bottleneck you may not have considered.  Typically digital cameras have a limited dynamic RAM buffer that holds the raw sensor data for a few images..  But the raw sensor data is not a JPEG or DNG file.  The sensor data has to be processed and re-formatted to DNG or JPEG before it can be saved onto the SD card.  It is likely that the delay you are experiencing is this processing delay.

So... The camera may be able to "burst" capture a few shots into the RAM buffer.  But the camera processor must then convert that data into DNG or JPEG before it can save out out to the SD card.  Until the data is converted it cannot empty the buffer for more pictures to be taken.

I am pretty sure the bottleneck is probably the camera processor.

chris.pugrud@gm Posted at 2015-6-9 23:36
TL;DR - There appears to be a bug in the camera firmware that restricts to the SD controller to th ...

Although it seems like Chris explanation is really solid, I'm wandering why DJI programmers would shoot  their own foot by limiting the software / firmware while the hardware would be theorically capable of higher speeds {:3_42:}

DJI representatives may we have a clarification here please?

Thanks

Isaac

Fulgerite Posted at 2015-6-9 23:55
There is another bottleneck you may not have considered.  Typically digital cameras have a limited d ...

That's interesting and it suggests that they used the same image processing chip for still photography as the p2v+. I've heard rumors that the bottleneck is the image processor, but have never been able to find confirmation from DJI. If the bottleneck is the image processor it could probably be helped by also giving us a RAW only option (J, J+R, R) which requires less processing than JPEG.

My suspicions point back to the SD bus because it lines up so closely with the peak transfer speed.  On the other hand the image processor helps explain some of the random variations observed in bench testing.

Either way there are more than a few people that would just like confirmation from DJI of this limitation for photographers.

Deeper digging by Disassembling the DJI Phantom 3 Camera reveals that the P3Pro camera uses the Ambarella A9 SOC which has some really impressive capabilites for a 2 year old 1st gen 4K/UHD image processor. The A9 provides a direct SDIO SDXC connection, so as long as DJI is using that connection there should be no bottleneck between the chip and the SDXC card.

(side note - the internal design of the P3 camera is very elegant and a work of art - congratulations DJI!)
The A9 appears to have some additional capabilities that DJI has not enabled (yet?). From the press release:

A9 Feature Summary

• 4K Ultra HD video recording at 30 frames per second for incredible detail.
• High definition video recording with 240 frames per second at 720p resolution.
• Burst mode support for still image capture of over 700 Megapixels per second.
• Multi-exposure High Dynamic Range (HDR) imaging support, including local tone mapping, preserves image details in high contrast scenes.
• Electronic Image Stabilization (EIS) with rolling shutter correction.
• 3D Motion-Compensated Temporal Filtering (MCTF) with de-ghosting for reduced artifacts and excellent low light performance.
• Support for applications including wireless preview, playback, and upload of stills and videos to a smartphone or directly to the Internet over WiFi™.
• A rich set of interfaces including DDR3L, USB 2.0 host, HDMI, and SDXC SD™ card.
  

If you want to get excited for the future, check out the next generation Ambarella H1, which includes 4k60 and H.265 encoding.

In summary it appears that the A9 chip and board have important capabilites that DJI is not taking advantage of and there is room for DJI to improve the camera performance and capability through firmware. Hopefully DJI is willing to bring in Ambarella to help optimize the camera firmware and increase our capabilities. One concern is that Ambarella is the exclusive supplier of image processing SOC for GoPro, but I'm sure that DJI is also a significant customer of Ambarella and they should be wanting to help DJI optimize their firmware and showcase the capabilities of the A9.

If you want to laugh a bit look at the specifications of the CCD sensor http://www.sony.net/Products/SC- ... ol_70/imx117cqt.pdf

Cappable of spiting 10 bit 12 Megapixel raw data at 40 frames per second {:3_57:} Is there any hardware cappable to process that? {:3_45:}

Isaac

Probably ... just not affordable for mass production

My Samsung NX1 sensor can spit 6K @ 120fps ....  that'd be lovely if they had the processor to match, haha.

Really interesting read this one although I have to say that I dont find it particularly limiting to me personally at present.

I reckon that the issue may possibly be improved upon with the next firmware update if DJI are wise.

Cheers,

Nick

apocnz@gmail.co Posted at 2015-6-13 10:46
Probably ... just not affordable for mass production  

My Samsung NX1 sensor can spit 6K @ 120fps ...

No your Samsung NX1 can NOT spit 6K video @ 120fps.  The Samsung NX1 shoots 4K.  (Not 6K)  The maximum frame rate on the NX1 is 30fps for 4K.
http://www.dpreview.com/products ... _nx1/specifications

Nick Martin Posted at 2015-6-13 13:18
Really interesting read this one although I have to say that I dont find it particularly limiting to ...

As best we can tell this "bug/feature" has been around for quite a while (p2v+?).

Because the bug/feature only affects photographers, and really only photographers shooting RAW (J+R, DNG), and is only particularly when shooting 5/7 rapid shot or 3/5AEB it does not affect the majority of users.

There may be a belief inside DJI firmware engineering that restricting the performance of SD cards will level the playing field between SD cards and minimize support incidens caused by users using slow cards.  The reality is that this punishes more knowledgeable users and really hinders the P3 as an aerial photography platform.

I'll test in depth with the new 1.2.6 firmware and see if that helps, but my first test just after the upgrade did not show a measurable improvement. I fear this is something that DJI is doing intentionally and it won't be fixed until they understand why it needs to be fixed.

Fulgerite Posted at 2015-6-16 00:34
No your Samsung NX1 can NOT spit 6K video @ 120fps.  The Samsung NX1 shoots 4K.  (Not 6K)  The maxi ...

The sensor ... not the camera     Read all the words

apocnz@gmail.co Posted at 2015-6-16 16:04
The sensor ... not the camera     Read all the words

You are still wrong.  While the sensor may be capable of shooting a 6K still and 4K video at 30fps... It does NOT have the buss speed and throughput to pipe 6K at 120fps.  The bandwidth of the chip buss is not sufficient.

High end sensors used on cameras like Arri Alexa and RED use dual buss architecture to allow the transfer bandwidth.  A consumer grade sensor does NOT have the buss bandwidth.

Hi,

any improvement on that side ?
I'm shooting timelapses, and now tried aerial timelapses with the phantom 3 Pro, and that 10sec interval in RAW is a real pain for me. Making timelapses from video is not good if you want the best quality (You can't blur any moving object due to the short exposure, and need to process mp4 files instead of RAW files). So you loose a lot of possibility just due to software apparently. I bought an Sandisk Extreme Pro MicroSD to get the maximum write speed but it's quite useless for now.

I really hope that DJI will unlock thoses restrictions for us photographers.