R5 recording limits temp/sensor correlations

Started Aug 13, 2020 | Discussions
Horshack Veteran Member • Posts: 8,739
R5 recording limits temp/sensor correlations
30

I've been participating on the FM thread (link) where the experiment was done which proves the R5's video thermal shutdown is not an indiscriminate cripple timer that has been described on some sites.

Right now the only temperature indicator we have from the R5 is what's reported in the EXIF data of images. However, if the R5 follows previous Canon bodies that temperature is sampled on a separate part of the PCB that is removed from the image sensor and DIGIC (see my post here - I'm "snapsy" on FM). This explains why a recent test shows the EXIF topping out well before the R5 reaches thermal shutdown.

There is however an alternate, indirect method to induce the relative temperature of the sensor. When CMOS image sensors heat up they produce more detectable image noise. I have a collection of R5 blackframes I've been using for the R5 raw noise-reduction investigation. A few have higher EXIF-reported temps than others. If I compare the noise between the lowest and warmest EXIF-reported temps I see a 4% difference in the standard deviation of noise:

EXIF temp 33C:
R: mean=511.84, stdev=1.44
G1: mean=511.90, stdev=1.30
G2: mean=511.91, stdev=1.31
B: mean=511.81, stdev=1.39

EXIF temp 46C:
R: mean=511.88, stdev=1.50
G1: mean=511.95, stdev=1.38
G2: mean=511.96, stdev=1.37
B: mean=511.89, stdev=1.45

The above is data for two blackframes, one where the EXIF-reported temp is 33C and the other 46C. In each of the 4 color channels there is approximately a 4% difference in the standard deviation of noise.

Even though the EXIF-reported temp seems to top out at around 46C I'm guessing the image sensor and DIGIC continue to get hotter as video recording occurs, and it's those rising temps which trigger the eventual thermal shutdown. I'd like to prove/disprove this theory by performing an experiment where raw images are taken before, during, and after thermal shutdown, plus additional raw frames during the recovery period, then comparing the image noise of the raw frames to see if there's a relative correlation between the noise (and thus sensor temperature) and the shutdown/recovery period. I don't own an R5 so I'm looking for volunteers who can perform the experiment then send me the resulting raw files

Experiment

  1. Starting from a completely cooled-down body, power on the camera and take the initial raw blackframe.
  2. Start recording 8K 30fps raw video. At 3 minute intervals, stop the raw recording and take another blackframe. Keep track of what the reported "available video time left" indicator reports on the EVF/LCD for each blackframe taken. Also take a blackframe the instant the camera starts showing a thermal warning. Continue this process until the camera reaches thermal shutdown.
  3. Power the camera back on after the thermal shutdown and take a blackframe. Power the camera back off
  4. Over the next 2 hours, power the camera on every 15 minutes to take a blackframe. Keep track of what the reported "available video time left" indicator shows for each blackframe.

Blackframe instructions

  • Exposure set to ISO 100 1/4000 f/8 with lens cap on. Shutter type set to EFCS (Electronic First Curtain Shutter)

Thanks!

Canon EOS R5
If you believe there are incorrect tags, please send us this post using our feedback form.
OP Horshack Veteran Member • Posts: 8,739
EXIF Temp vs Sensor Noise vs Indicated 8K Recording Time Avail
17

With the generous donation of Hoka Key's time I was able to obtain blackframes for the initial experiment. Below are the results, contained in three charts (direct link to charts).

The top chart shows the correlation between the EXIF-reported temperature and the increase in measured sensor noise at that temperature. This establishes a rough correlation between the EXIF temp and the relative temperature at the image sensor, as induced by the increase in sensor noise.

The bottom-left chart shows the camera-indicated available 8K recording time (reported to the user on the LCD/EVF) at each EXIF-reported temperature.

The bottom-right chart shows the camera-indicated available 8K recording time (reported to the user on the LCD/EVF) at each relative sensor noise measurement.

The data was obtained by shooting individual 3-minute 8K videos (ALL-I compression), in between which a blackframe was taken to provide a snapshot of the EXIF-reported temp and image data from which sensor noise can be measured. Between each clip the indicated available video recording time was noted, along with how much video was actual shot (3 minutes for all video except the last, where the camera wouldn't allow any more video to be shot even though it indicated 3 minutes available).

Click "original size" to view.

This dataset establishes the EXIF and sensor-noise values associated the decreasing available 8K video recording time left, starting with the full available time of 15 minutes and counting down to when thermal shutdown would occur (or no more video allowed to be taken).

The next experiment to be performed is to shoot stills-only, until the available video recording time indicated drops to zero again. At that point a blackframe will be taken and I'll be comparing the EXIF-reported temperature and sensor noise measurements to the data from this first experiment. The purpose is to see if the R5's EXIF and sensor noise measurements for stills-only match the values at thermal shutdown that occurs from actually shooting video. This will hopefully provide more insight into the contours of the R5's thermal management algorithm/cutoffs.

koenkooi Contributing Member • Posts: 502
Re: EXIF Temp vs Sensor Noise vs Indicated 8K Recording Time Avail

Is that the data for just the heating up part or for both the heating up and cooldown/recovery part? If it's the latter, could you plot those values with a different colour? That would make it easier to see potential differences between heatup and recovery.

 koenkooi's gear list:koenkooi's gear list
Canon EOS 7D Canon EOS M Canon EOS M6 II Canon EOS R5 Canon EF 85mm F1.8 USM +20 more
OP Horshack Veteran Member • Posts: 8,739
Additional volunteers needed

koenkooi wrote:

Is that the data for just the heating up part or for both the heating up and cooldown/recovery part? If it's the latter, could you plot those values with a different colour? That would make it easier to see potential differences between heatup and recovery.

Just the heating up. The experiment also called for measuring the cooldown/recovery period but that's time consuming and hasn't been collected yet. I could use additional volunteers so I don't have to burden Hoka with all my experiment requests.

jonpais
jonpais Veteran Member • Posts: 3,079
Re: EXIF Temp vs Sensor Noise vs Indicated 8K Recording Time Avail

Is it your contention then that it requires two hours for the sensor to cool down? Or perhaps I’m misunderstanding...

-- hide signature --

an achievement unparalleled in human history
https://daejeonchronicles.com

 jonpais's gear list:jonpais's gear list
Sony a7S III Sony FE 55mm F1.8 Sony FE 135mm F1.8 GM Sony FE 20mm F1.8G Sigma 85mm F1.4 DG GN +2 more
koenkooi Contributing Member • Posts: 502
Re: Additional volunteers needed

Horshack wrote:

koenkooi wrote:

Is that the data for just the heating up part or for both the heating up and cooldown/recovery part? If it's the latter, could you plot those values with a different colour? That would make it easier to see potential differences between heatup and recovery.

Just the heating up. The experiment also called for measuring the cooldown/recovery period but that's time consuming and hasn't been collected yet. I could use additional volunteers so I don't have to burden Hoka with all my experiment requests.

Understood. I can help out after my R5 gets delivered, which will take a while I fear.

 koenkooi's gear list:koenkooi's gear list
Canon EOS 7D Canon EOS M Canon EOS M6 II Canon EOS R5 Canon EF 85mm F1.8 USM +20 more
OP Horshack Veteran Member • Posts: 8,739
How about 69 minutes of 8K record time?
19

Hoka has come through again with his extremely generous donation of time with an amazing new experiment. After last night's experiment he put his R5 in the meat drawer of his fridge for 11 hours - the surface temp inside the drawer was 0.4C (ie, just above freezing). After taking it out of the freezer he then repeated yesterday's experiment. Instead of getting only 15 minutes of 8K record time he got 69 minutes!

There is one other significant difference that might prove more interesting than the meat drawer. Yesterday's experiment was with a Sony CFE card. Today's experiment was with a ProGrade card. In today's experiment he actually ran out of space after 60 minutes of recording on the ProGrade, so he then switched to the Sony card. Just before the switch the camera was reporting near 0:00 available video recording time left (thermal indicator, not card space). After he switched from the ProGrade to the Sony the camera immediately jumped up to 15:00, and provided 15 minutes of recording before reaching thermal shutdown.

Another interesting CFE observation is the indicated available video time on the ProGrade is displayed with granularity to the second - for example at the start of the test it showed 24:23 (24 minutes, 23 seconds), whereas the Sony always shows the seconds field of zero (for example, 15:00, 10:00, 05:00). It's very interesting that the precision of the available time left (presumably based exclusively on thermals) would be different for one card vs another. This implies the thermal calculation may be based in part on the CFE interface and specific card, either due to differences in some field the card returns in a CFE/NVMe protocol or even more interesting, based on a thermal element of the card itself. More experiments will need to be performed to investigate this aspect.

In the meantime, here is a new graph (direct link) - this one shows EXIF-reported temp over each sequential 3-minute 8K clip. Yesterdays' graph showed the indicated available recording time left instead. The graph on the left shows yesterday's experiment, at normal camera/ambient temperatures (15 minutes before thermal cutoff). The graph on the right shows today's experiment, performed after removing the camera from freezer overnight (69 minutes before thermal cutoff, and using a different CFE card). Notice the camera started at a much lower temperature base after being in the freezer...and took much longer to heat up.

Click "original size" to view

More analysis of this data set to come...

(unknown member) Regular Member • Posts: 376
Very interesting, and thanks to both of you!

Horshack wrote:

Hoka has come through again with his extremely generous donation of time with an amazing new experiment. After last night's experiment he put his R5 in the meat drawer of his fridge for 11 hours - the surface temp inside the drawer was 0.4C (ie, just above freezing). After taking it out of the freezer he then repeated yesterday's experiment. Instead of getting only 15 minutes of 8K record time he got 69 minutes!

There is one other significant difference that might prove more interesting than the meat drawer. Yesterday's experiment was with a Sony CFE card. Today's experiment was with a ProGrade card. In today's experiment he actually ran out of space after 60 minutes of recording on the ProGrade, so he then switched to the Sony card. Just before the switch the camera was reporting near 0:00 available video recording time left (thermal recording, not card space). After he switched from the ProGrade to the Sony the camera immediately jumped up to 15:00, and provided 15 minutes of recording before reaching thermal shutdown.

Another interesting CFE observation is the indicated available video time on the ProGrade is displayed with granularity to the second - for example at the start of the test it showed 24:23 (24 minutes, 23 seconds), whereas the Sony always shows the seconds field of zero (for example, 15:00, 10:00, 05:00). It's very interesting that the precision of the available time left (presumably based exclusively on thermals) would be different for one card vs another. This implies the thermal calculation may be based in part on the CFE interface and specific card, either due to differences in some field the card returns in a CFE/NVMe protocol or even more interesting, based on a thermal element of the card itself. More experiments will need to be performed to investigate this aspect.

In the meantime, here is a new graph (direct link) - this one shows EXIF-reported temp over each sequential 3-minute 8K clip. Yesterdays' graph showed the indicated available recording time left instead. The graph on the left shows yesterday's experiment, at normal camera/ambient temperatures (15 minutes before thermal cutoff). The graph on the right shows today's experiment, performed after removing the camera from freezer overnight (69 minutes before thermal cutoff, and using a different CFE card). Notice the camera started at a much lower temperature base after being in the freezer...and took much longer to heat up.

Click "original size" to view

More analysis of this data set to come...

Both of you are investing a lot of time and effort, and it is greatly appreciated.

I do not have 2 CFx cards, but if I get a second one, the thought is I could swap them out before the camera shuts down and gain maybe up to 15 minutes more record time?

Do we think one of the cards lasts longer than the other?  I'd be glad to buy another CFx card and test.

OP Horshack Veteran Member • Posts: 8,739
EXIF-reported temp vs image sensor noise
6

Here's a graph showing the relationship between the EXIF-reported temp (extracted from the still blackframe images shot between each 3-minute 8K clip) and the measured sensor noise (calculated from the image data in the blackframe). As CMOS sensors heat up their noise levels rise, so the relative temperature of the image sensor can be induced from the relative increase in noise.

As described in the OP, the location within the camera from where the EXIF-reported temp is measured is unknown at this point - in previous Canon bodies it was taken from a chip called the EFIC, which was responsible for some low-level lens interface and Speedlite logic and was on separate part of the PCB from the image sensor / DIGIC. It's not clear if this is true for the R5 as well.

The purpose of correlating the EXIF temp to the sensor noise is to attempt to establish how representative/correlated the EXIF temp is of the image sensor's temp. The purpose of that is to help determine which temps might be used in the R5's thermal shutoff algorithm.

The absolute temp of the image sensor is unknown but we can compare the relative temp of the sensor (derived from noise) vs the EXIF temp by looking at how the two correlate as the EXIF temp increases.

Here's the graph, including the raw data (direct link):

Click "original size"

(unknown member) Regular Member • Posts: 376
Actual sensor temp.

I did a test about a week ago with an IR thermometer. I measured the sensor temp at 112 F degrees after my R5 shut down from about 20 minutes of shooting 8K (I think in that test it was to the SD card and not RAW)

I could try to do a test where I measure the sensor with the thermometer ever 3 minutes if you like. I don't know if I can video with the lens off. I may have to remove it every 3 minutes like he did in your test. I wish I could better measure the CFx port inside the CFx slot.

Pealut Nutter Regular Member • Posts: 103
Re: R5 recording limits temp/sensor correlations

Wow, what a nice experiment Horshack. Man, do I love your posts!

OP Horshack Veteran Member • Posts: 8,739
Graph of Remaining recording time vs EXIF temp
9

Here's a graph and raw data (direct link) showing the indicated available time remaining for 8K recording vs the EXIF-reported temp. Each point on the graph represents 3 minutes in time, since the methodology is to shoot a 3-minute 8K clip, followed by a single still image raw blackframe (to get the EIXF temp/sensor noise).

I've added a new column to the raw data - "Delta Available Time Indicated vs Last Sample". This calculates the difference in indicated available record time between two consecutive 3-minute clips. If the algorithm behind that calculation were accurate the value in this column should decrease by 3 minutes for each 3-minute clip. However you'll notice the value actually decreases by about half that over the entire experiment, meaning it is underestimating how much 8K video the camera ultimately allows Hoka to record in this experiment.

This has interesting implications. First, it means the camera is obviously calculating the remaining time incorrectly in this scenario, perhaps because the camera is heating up much more slowly than is typical due to it being previously frozen (ie, the firmware is projecting a thermal ramp based on assumed, normal ambient conditions rather than the actual observed rate the camera is heating up). Second, it means the logic that triggers the thermal shutdown is independent of the logic that is projecting when that shutdown would occur (ie, one piece of logic determines the actual shutdown criteria whereas the other is for display-purposes only).

The big knee you see in the graph at 69C is when Hoka had to swap out the full ProGrade CFE for the Sony CFE. As mentioned previously, this reset the available recording time back to 15:00, which is extremely interesting and warrants further investigation.

Click "original size"

traderjay
traderjay Regular Member • Posts: 466
Re: Graph of Remaining recording time vs EXIF temp

Horshack wrote:

Here's a graph and raw data (direct link) showing the indicated available time remaining for 8K recording vs the EXIF-reported temp. Each point on the graph represents 3 minutes in time, since the methodology is to shoot a 3-minute 8K clip, followed by a single still image raw blackframe (to get the EIXF temp/sensor noise).

I've added a new column to the raw data - "Delta Available Time Indicated vs Last Sample". This calculates the difference in indicated available record time between two consecutive 3-minute clips. If the algorithm behind that calculation were accurate the value in this column should decrease by 3 minutes for each 3-minute clip. However you'll notice the value actually decreases by about half that over the entire experiment, meaning it is underestimating how much 8K video the camera ultimately allows Hoka to record in this experiment.

This has interesting implications. First, it means the camera is obviously calculating the remaining time incorrectly in this scenario, perhaps because the camera is heating up much more slowly than is typical due to it being previously frozen (ie, the firmware is projecting a thermal ramp based on assumed, normal ambient conditions rather than the actual observed rate the camera is heating up). Second, it means the logic that triggers the thermal shutdown is independent of the logic that is projecting when that shutdown would occur (ie, one piece of logic determines the actual shutdown criteria whereas the other is for display-purposes only).

The big knee you see in the graph at 69C is when Hoka had to swap out the full ProGrade CFE for the Sony CFE. As mentioned previously, this reset the available recording time back to 15:00, which is extremely interesting and warrants further investigation.

Click "original size"

Once a new "COLD" CFE card is inserted, did hoka get the full 15 min record? If so I think the camera's own algorithm is protecting the card from overheating/damage?

-- hide signature --

Workstations - EPYC 7702 + 3950X || NVIDIA RTX 2080ti

 traderjay's gear list:traderjay's gear list
Canon EOS R5 Canon RF 28-70mm F2L USM Canon RF 70-200mm F2.8L IS USM DxO Optics Pro Elite +1 more
OP Horshack Veteran Member • Posts: 8,739
Re: Graph of Remaining recording time vs EXIF temp
3

traderjay wrote:

Horshack wrote:

Here's a graph and raw data (direct link) showing the indicated available time remaining for 8K recording vs the EXIF-reported temp. Each point on the graph represents 3 minutes in time, since the methodology is to shoot a 3-minute 8K clip, followed by a single still image raw blackframe (to get the EIXF temp/sensor noise).

I've added a new column to the raw data - "Delta Available Time Indicated vs Last Sample". This calculates the difference in indicated available record time between two consecutive 3-minute clips. If the algorithm behind that calculation were accurate the value in this column should decrease by 3 minutes for each 3-minute clip. However you'll notice the value actually decreases by about half that over the entire experiment, meaning it is underestimating how much 8K video the camera ultimately allows Hoka to record in this experiment.

This has interesting implications. First, it means the camera is obviously calculating the remaining time incorrectly in this scenario, perhaps because the camera is heating up much more slowly than is typical due to it being previously frozen (ie, the firmware is projecting a thermal ramp based on assumed, normal ambient conditions rather than the actual observed rate the camera is heating up). Second, it means the logic that triggers the thermal shutdown is independent of the logic that is projecting when that shutdown would occur (ie, one piece of logic determines the actual shutdown criteria whereas the other is for display-purposes only).

The big knee you see in the graph at 69C is when Hoka had to swap out the full ProGrade CFE for the Sony CFE. As mentioned previously, this reset the available recording time back to 15:00, which is extremely interesting and warrants further investigation.

Click "original size"

Once a new "COLD" CFE card is inserted, did hoka get the full 15 min record? If so I think the camera's own algorithm is protecting the card from overheating/damage?

Yes he did, and that is obviously one of the prime areas we'll be focused on during the next experiments. It's not clear yet if that was related specifically to inserting a "cool" card or if it was a side effect of the thermal logic run afoul due to the irregular temperature ramp of the frozen camera. Whichever the case it seems to indicate the CFE plays some role in the camera's thermal projection calculations.

TnSkiDude Regular Member • Posts: 178
Re: How about 69 minutes of 8K record time?
3

Thanks to everyone involved in this great experiment.  It's really insightful that different brand cards report different time resolution.  I'm curious was the card "frozen" with the camera?  Perhaps we don't have to cool the camera but just the cards.

OP Horshack Veteran Member • Posts: 8,739
Re: How about 69 minutes of 8K record time?
1

TnSkiDude wrote:

Thanks to everyone involved in this great experiment. It's really insightful that different brand cards report different time resolution. I'm curious was the card "frozen" with the camera? Perhaps we don't have to cool the camera but just the cards.

Both the cards and battery were removed prior to freezing.

Wing2 Regular Member • Posts: 480
Re: How about 69 minutes of 8K record time?

Horshack wrote:

TnSkiDude wrote:

Thanks to everyone involved in this great experiment. It's really insightful that different brand cards report different time resolution. I'm curious was the card "frozen" with the camera? Perhaps we don't have to cool the camera but just the cards.

Both the cards and battery were removed prior to freezing.

How about keep swapping two cards every 10-20 minutes ? Please test swapping before and swapping after overheat warning came on.

Would it reset the timer to 15 minutes every time a card is swapped?

OP Horshack Veteran Member • Posts: 8,739
Re: How about 69 minutes of 8K record time?

Wing2 wrote:

Horshack wrote:

TnSkiDude wrote:

Thanks to everyone involved in this great experiment. It's really insightful that different brand cards report different time resolution. I'm curious was the card "frozen" with the camera? Perhaps we don't have to cool the camera but just the cards.

Both the cards and battery were removed prior to freezing.

How about keep swapping two cards every 10-20 minutes ? Please test swapping before and swapping after overheat warning came on.

Would it reset the timer to 15 minutes every time a card is swapped?

Already on the list of experiments 😀

I would caution others to not read too much into the card-switching observations just yet. There were other factors, particularly the frozen start, which may be behind the differences observed.

RDM5546
RDM5546 Senior Member • Posts: 3,036
Re: R5 recording limits temp/sensor correlations
1

Pealut Nutter wrote:

Wow, what a nice experiment Horshack. Man, do I love your posts!

Hurray for Horshack!!!

Hurray for Hoka Key!!!!!!!!!

It really refreshing to see such clearly established facts here to replace the wild speculation.    The swapping of the CFEpresss type B cards was a a breakthrough momemnt.   15 minutes regained. Means as simple watchdog timer design using camera 8K write operation does not explain this fantastic chart.    The protection system genorous atypical give back of 15 minutes is one breakthrough but the fact this is a different time than if the same card was returned to the drive after stopping the recording for a brief moment means the cards have a unique signature that the DIGIC X includes in the thermal management strategy.

It maybe early to speculate but it may appears true that the exif temperature data location in the EXIF data is not related to the CF express card hot temperature.  It further seems that 73C is the the trigger but this may not be a real temperature but an algorithm score from time and temperature or either in combination with 8k video write cycles being operative.  It seems to be not a real temperature because it start at 5C which is very cold so there is reason to doubt this an accurate 5 degrees C temperature inside the camera some where.  I may be a combination score of multiple factors though.

More fresh stone cold CF express card would handy to do swap out after 8 minutes at which point the EXIF data score will be about 43C and then repeat with a fresh card to see if this gives back 15minutes for each 8 minute interval of time the operating card virgin cool card replace the warmed card.  Af the first 8 minutes the report temp goes up .  While I suggesting using multiple card and switching them frequently I am not think this is a practical time period for widespread use but merely a simplifying tool for better understanding of the thermal management method.

Unfortuneately the cards are very expensive so maybe two cards will have to do though three cards would be even better.   If my camera arrives next week or soon I will join the hunt this hunt for answers with my own camera. I bought two 128GB Sony CF Express type B cards on Amazon Prime that are supposed to arrive in 10 days which may be the soonest that my R5 arrives if I am in batch #2.

 RDM5546's gear list:RDM5546's gear list
Canon RF 70-200mm F2.8L IS USM Canon RF 100-500mm F4.5-7.1L IS USM Canon G5 X II Canon EOS 70D Canon EOS 7D Mark II +43 more
RDM5546
RDM5546 Senior Member • Posts: 3,036
Re: How about 69 minutes of 8K record time?

Wing2 wrote:

Horshack wrote:

TnSkiDude wrote:

Thanks to everyone involved in this great experiment. It's really insightful that different brand cards report different time resolution. I'm curious was the card "frozen" with the camera? Perhaps we don't have to cool the camera but just the cards.

Both the cards and battery were removed prior to freezing.

How about keep swapping two cards every 10-20 minutes ? Please test swapping before and swapping after overheat warning came on.

Would it reset the timer to 15 minutes every time a card is swapped?

I have heard on forums that swapping cards after the timeout is a two hour penalty!!. Timeout is common arround 20minutes but not consistent to the second so if you are going to swap cards do it when the is still four minutes left or more or the demon will get you for two hours.

Apparently if you put the camera is a ziploc bag in you freezer it will operate again after around 30 minutes in the freezer. I am not sure people tried freezing just he cards only though in anything that I have read.  Frome the Hoka data we only different card give more time. Does it just remember the last card ID or do you need several cards.

It is not clear that the freezing alone will do it when you consider the Hoka data because the Camera is remember the ID of at least one card back in time history.

 RDM5546's gear list:RDM5546's gear list
Canon RF 70-200mm F2.8L IS USM Canon RF 100-500mm F4.5-7.1L IS USM Canon G5 X II Canon EOS 70D Canon EOS 7D Mark II +43 more
Keyboard shortcuts:
FForum MMy threads