FYI, the first part of the experiment I described here has been completed. A more detailed description of the experiment is here, and the results are posted here.
TeamSpeed wrote:
He obviously doesn't understand how the HDMI output works. Many cameras will output full sensor output to HDMI because it doesn't require alot of horsepower to do that. The horsepower needed is when you snap off frame after frame of a video feed and construct movie files. Debayering the data, applying picture styles, etc whoa all such hard things to do (just numeric processing on the data). [..]
When using HDMI output the camera still needs to debayer, apply picture styles, lens correction, etc. It also needs to convert it to YUV and scale some of the plane, since it's 4:2:2, not 4:4:4. Only if your camera supports RAW output over HDMI (e.g. Panasonic S1R, Nikon Z6) it can skip those steps.
TeamSpeed wrote:
Also, the Chinese desconstructor aimed an IR gun at the processor to register what temp?
Using an IR gun to measure temperature requires a lot of knowledge about what you are doing and what kind of material you are pointing the gun at. It is not in any way a reilable way to measure temperature in itself, especially not if you modify the surface with heat paste etc. between measurements, The IR emissivity of a surface (or fraction of emitted IR radiation relative to a "black" surface) can vary between 0.0 and 1.0. It is also highly wavelength dependent (surfaces have "color" also in the IR domain).
I am working with IR based surface defect detection in the steel industry. Customers have in some cases used these IR temp guns to prove the temperature of the blank grinded or peeled steel surface, when we say that it is too hot to pass in our inspection machine. We measured the temperature to be 70 degrees C with contact measurement, while their gun said it was 40. In those cases I have a simple experiment at hand. Take a bottle of cold or tempered water, say 20 degrees. Hold the gun pointing at the steel. Watch the temperature. As we pour the 20 degrees cold water over the hot steel, you should have seen the faces of the (chinese) customers. You would expect the temperture reading to drop, not? The temperature reading in the probe increases from 40 to 60-70 degrees as the cold water runs over the surface. This is because water is IR black, with emissivity close to 1. The hot steel quickly heats the thin water film, and while giving the IR gun the surface it needs to give a correct temperature reading.
The same thing (or the opposite) could be happening when applying heat paste over a component.
koenkooi wrote:
When using HDMI output the camera still needs to debayer, apply picture styles, lens correction, etc. It also needs to convert it to YUV and scale some of the plane, since it's 4:2:2, not 4:4:4. Only if your camera supports RAW output over HDMI (e.g. Panasonic S1R, Nikon Z6) it can skip those steps.
Yes and that might get the internals up a bit but nothing like having to create the frames, apply further processing and stitch into a large video file, coupled with the I/O needed. Even something like the M6II can do a good clean HDMI output with it's 33Mpx sensor, and doesn't require the same horsepower.
Speaking of that, I wonder something.
- The M6II is a much smaller package, does 4K30 video and is 33Mpx, which is within spitting distance of the R5.
- Both the R5 and R6 overheat, yet the M6II has no issues, despite it being a smaller body.
- The M6II doesn't have Eye AF and Deep Learning configurations to dig through, both the R5 and R6 do.
Has anyone tried video and stills with single point AF and turn off all the Eye AF functions/iTR type settings to see if there is any impact to temps and times? The big thing people are excited by and are using with these 2 cameras is eye AF, and this would be active whether stills or video. What if one could turn off all the deep learning algorithms and stick with traditional phase/contrast detect AF?
Snapsy, thanks for being a voice of reason on this topic, it is pity that there are so many hot heads some of whom would even go so far as to delete comments by knowledgeable posters to maintain their narrative.
I can understand thermal shutdown at 62 C, it seems perfectly reasonable, and I also understand why recovery in an insulated housing takes so long. But I wonder what is going on when the camera doesn't even start high-quality video recording after a relatively short (hour) capture of a small number of still photos. Could there be some component that is getting hot in that kind of use or maybe there is a firmware bug or oversight that makes the camera overly protective in this hybrid shooting scenario.
What would be interesting to do is to study this problem by using your approach in the following way:
1. Start with a cool camera at room temperature.
2. Record a still image every 5 min (interval timer) and monitor the temperature reported by the camera and based on the noise in the black frame. Maintain EVF or LCD viewing between frames (no standby timer) to simulate hunting for subjects and viewing scenes.
3. After 1-2 hours, test whether the camera will allow 4K HQ or 8K recording to start. (Some users report that after 1-2 hours of still shooting, the camera won't record in the high-quality video modes at all. Not sure whether elevated external temperature is affecting this scenario.)
ilkka_nissila wrote:
Snapsy, thanks for being a voice of reason on this topic, it is pity that there are so many hot heads some of whom would even go so far as to delete comments by knowledgeable posters to maintain their narrative.
I can understand thermal shutdown at 62 C, it seems perfectly reasonable, and I also understand why recovery in an insulated housing takes so long. But I wonder what is going on when the camera doesn't even start high-quality video recording after a relatively short (hour) capture of a small number of still photos. Could there be some component that is getting hot in that kind of use or maybe there is a firmware bug or oversight that makes the camera overly protective in this hybrid shooting scenario.
What would be interesting to do is to study this problem by using your approach in the following way:
1. Start with a cool camera at room temperature.
2. Record a still image every 5 min (interval timer) and monitor the temperature reported by the camera and based on the noise in the black frame. Maintain EVF or LCD viewing between frames (no standby timer) to simulate hunting for subjects and viewing scenes.
3. After 1-2 hours, test whether the camera will allow 4K HQ or 8K recording to start. (Some users report that after 1-2 hours of still shooting, the camera won't record in the high-quality video modes at all. Not sure whether elevated external temperature is affecting this scenario.)
Thanks. The next experiment on deck is to measure the EXIF/sensor noise for a stills-only session that leads to zero-video recording time available, which is what EOSHD reported. I'm having to rely on volunteers to perform the experiments so the results will come in fits and starts. The other part of the first experiment was to measure the cooldown/recovery times but that hasn't been performed yet (it's a big ask for someone to spend two hours on it).
ilkka_nissila wrote:
Snapsy, thanks for being a voice of reason on this topic, it is pity that there are so many hot heads some of whom would even go so far as to delete comments by knowledgeable posters to maintain their narrative.
I can understand thermal shutdown at 62 C, it seems perfectly reasonable, and I also understand why recovery in an insulated housing takes so long. But I wonder what is going on when the camera doesn't even start high-quality video recording after a relatively short (hour) capture of a small number of still photos. Could there be some component that is getting hot in that kind of use or maybe there is a firmware bug or oversight that makes the camera overly protective in this hybrid shooting scenario.
What would be interesting to do is to study this problem by using your approach in the following way:
1. Start with a cool camera at room temperature.
2. Record a still image every 5 min (interval timer) and monitor the temperature reported by the camera and based on the noise in the black frame. Maintain EVF or LCD viewing between frames (no standby timer) to simulate hunting for subjects and viewing scenes.
3. After 1-2 hours, test whether the camera will allow 4K HQ or 8K recording to start. (Some users report that after 1-2 hours of still shooting, the camera won't record in the high-quality video modes at all. Not sure whether elevated external temperature is affecting this scenario.)
Thus my question about the fact that the R5/R6 both get hot where something like the M6II doesn't get hot... I wonder how much of the heat is the processing for eye AF and deep learning matrices that it has to go through for decision making. Has testing been performed where Eye AF is turned off and just a single point AF is used? That is the common component that spans both stills and video, and the R5/R6 and doesn't involve the M6II (which has 33Mpx and 4K video in a smaller package). It is also the largest exciting thing about the cameras, so everyone is using it...
TeamSpeed wrote:
Yes and that might get the internals up a bit but nothing like having to create the frames, apply further processing and stitch into a large video file, coupled with the I/O needed. Even something like the M6II can do a good clean HDMI output with it's 33Mpx sensor, and doesn't require the same horsepower.
Speaking of that, I wonder something.
- The M6II is a much smaller package, does 4K30 video and is 33Mpx, which is within spitting distance of the R5.
- Both the R5 and R6 overheat, yet the M6II has no issues, despite it being a smaller body.
- The M6II doesn't have Eye AF and Deep Learning configurations to dig through, both the R5 and R6 do.
Has anyone tried video and stills with single point AF and turn off all the Eye AF functions/iTR type settings to see if there is any impact to temps and times? The big thing people are excited by and are using with these 2 cameras is eye AF, and this would be active whether stills or video. What if one could turn off all the deep learning algorithms and stick with traditional phase/contrast detect AF?...Show more →
The 4k on the M6II is done through line/column skipping and then upsampled from 3.something k to 4k. The 90D with the same sensor and digic does downsample 4k. So you get much nicer 4k from the 90D than the M6II. There are more things off about the M6II, like the lack of EFCS, so I'm hesitant to use it as an example.
koenkooi wrote:
The 4k on the M6II is done through line/column skipping and then upsampled from 3.something k to 4k. The 90D with the same sensor and digic does downsample 4k. So you get much nicer 4k from the 90D than the M6II. There are more things off about the M6II, like the lack of EFCS, so I'm hesitant to use it as an example.
I wonder if the smaller battery in the M6II is a factor here related to how the M6II features were designed.
Of course, it could just be that Canon is placing products into their designated categories (marketing).
TeamSpeed wrote:
Yes and that might get the internals up a bit but nothing like having to create the frames, apply further processing and stitch into a large video file, coupled with the I/O needed. Even something like the M6II can do a good clean HDMI output with it's 33Mpx sensor, and doesn't require the same horsepower.
Speaking of that, I wonder something.
- The M6II is a much smaller package, does 4K30 video and is 33Mpx, which is within spitting distance of the R5.
- Both the R5 and R6 overheat, yet the M6II has no issues, despite it being a smaller body.
- The M6II doesn't have Eye AF and Deep Learning configurations to dig through, both the R5 and R6 do.
Has anyone tried video and stills with single point AF and turn off all the Eye AF functions/iTR type settings to see if there is any impact to temps and times? The big thing people are excited by and are using with these 2 cameras is eye AF, and this would be active whether stills or video. What if one could turn off all the deep learning algorithms and stick with traditional phase/contrast detect AF?...Show more →
You still have to complete the frames. The difference between HDMI and internal basically boils down to whether the data undergoes compression (either intra or inter frame depending on all-i or ipb), and where the data gets stored.
For the HDMI case the frames are fully completed when sent over HDMI, they just bypass the file compression stage.
The fact that IPB vs ALL-I makes no difference to record times is (not conclusive) evidence that compression does not play a large role in heat generation as IPB is significantly more compute hungry as it has to analyze deltas from previous frames (and future frames if wishing to generate B frames) than ALL-I.
The heat difference between transferring over HDMI vs transferring to an internal card should not be very large. Keep in mind you can store 4kHQ to an sdcard and the camera still overheats, indicating it is not a CFExpress exclusive issue.
4k60 4:2:2 over HDMI requires 17.2Gbps of transferred data, or in the realm of 2.15 GB/s. That should also be generating a large amount of heat if transferring internally is also generating lots of heat.
Jesse Evans wrote:
You still have to complete the frames. The difference between HDMI and internal basically boils down to whether the data undergoes compression (either intra or inter frame depending on all-i or ipb), and where the data gets stored.
For the HDMI case the frames are fully completed when sent over HDMI, they just bypass the file compression stage.
The fact that IPB vs ALL-I makes no difference to record times is (not conclusive) evidence that compression does not play a large role in heat generation as IPB is significantly more compute hungry as it has to analyze deltas from previous frames (and future frames if wishing to generate B frames) than ALL-I.
The heat difference between transferring over HDMI vs transferring to an internal card should not be very large. Keep in mind you can store 4kHQ to an sdcard and the camera still overheats, indicating it is not a CFExpress exclusive issue.
4k60 4:2:2 over HDMI requires 17.2Gbps of transferred data, or in the realm of 2.15 GB/s. That should also be generating a large amount of heat if transferring internally is also generating lots of heat. ...Show more →
One post I made on EOSHD was explaining how the thermal budget for DIGIC could be based on how many of its internal cores are actively under load. It's akin to an AMD/Intel processor that can run at higher clocks when only one or two cores are active but then has to throttle down when all cores are running. So it may not be just a function of how compute-intensive one role is vs another but instead the total compute/heat generated over the entire chip for all roles running simultaneously.
I would still love to know whether the eye af/deep learning libraries that have to be referenced are part of the equation here. Eye AF could be very CPU intensive and I/O intensive as well, and would affect both stills and movie.
Has anyone tried taking movies with a fixed AF point, or AF turned off to see if there is any difference at all in overheating times?
Canon has other cameras that can do 4K30/60, etc and none have this overheating issue. It's not the CF express that is the issue, it has to be something that is common to both the R6 and R5, and the only thing common to these are they share the same eye AF, IBIS and digic processor. (Different cards, different sensor resolutions, etc....)
I am not even sure they share the same main board with 4 ram chips? Nobody has done a tear down on the R6 yet.
The 1DX3 has basically the same type of AF for live view, correct, and sensor as the R6? However it has both a DigicX and Digic8 processor, with one dedicated to AF functionality, so processing will occur differently on the main board for that camera.
The only commonality I can come up after looking at all the factors has to be the Eye AF/Deep Learning matrices. This could explain why Canon's next firmware will raise the ceiling on recording time just a bit, and then also offer the light raw recording mode (allow the heat to build up just a bit more, and reduce file I/O to the card).
I bet they have no way to cool this off any further without downgrading eye AF, or redesign it to be more efficient with system resources. Just a guess though at this point. I wonder if they have the software folks digging through all the code to find ways to reduce CPU drag. Been there, done that...
TeamSpeed wrote:
Yes and that might get the internals up a bit but nothing like having to create the frames, apply further processing and stitch into a large video file, coupled with the I/O needed. Even something like the M6II can do a good clean HDMI output with it's 33Mpx sensor, and doesn't require the same horsepower.
Speaking of that, I wonder something.
- The M6II is a much smaller package, does 4K30 video and is 33Mpx, which is within spitting distance of the R5.
- Both the R5 and R6 overheat, yet the M6II has no issues, despite it being a smaller body.
- The M6II doesn't have Eye AF and Deep Learning configurations to dig through, both the R5 and R6 do.
Has anyone tried video and stills with single point AF and turn off all the Eye AF functions/iTR type settings to see if there is any impact to temps and times? The big thing people are excited by and are using with these 2 cameras is eye AF, and this would be active whether stills or video. What if one could turn off all the deep learning algorithms and stick with traditional phase/contrast detect AF?...Show more →
How about the 1DX3? But as we know, it has dual processors, one dedicated to AF. The R5/R6 do not. so the DigicX has to do all the work that dual processors do on the 1DX3.
In any case, I just wonder if there is a way to test without using any AF at all in the camera, turn off AF, lens and body, etc and see if there is any difference at all in heat cycling times.
you should have some of the astro guys to run your experiment in an app like APT. APT has the ability to run a scripted session that would record the data and automate the process for both stills and video recording. APT can also set the camera parameters allowing one to record long and short exposures effectively removing the human in the loop.
How about the 1DX3? But as we know, it has dual processors, one dedicated to AF. The R5/R6 do not. so the DigicX has to do all the work that dual processors do on the 1DX3.
In any case, I just wonder if there is a way to test without using any AF at all in the camera, turn off AF, lens and body, etc and see if there is any difference at all in heat cycling times.
bigger DSLR bodies, different cards, bit rates, it could be various things. Just addressing that the M6ii to R5 comparison isnt valid.
TeamSpeed wrote:
I would still love to know whether the eye af/deep learning libraries that have to be referenced are part of the equation here. Eye AF could be very CPU intensive and I/O intensive as well, and would affect both stills and movie.
Has anyone tried taking movies with a fixed AF point, or AF turned off to see if there is any difference at all in overheating times?
Canon has other cameras that can do 4K30/60, etc and none have this overheating issue. It's not the CF express that is the issue, it has to be something that is common to both the R6 and R5, and the only thing common to these are they share the same eye AF, IBIS and digic processor. (Different cards, different sensor resolutions, etc....)
I am not even sure they share the same main board with 4 ram chips? Nobody has done a tear down on the R6 yet.
The 1DX3 has basically the same type of AF for live view, correct, and sensor as the R6? However it has both a DigicX and Digic8 processor, with one dedicated to AF functionality, so processing will occur differently on the main board for that camera.
The only commonality I can come up after looking at all the factors has to be the Eye AF/Deep Learning matrices. This could explain why Canon's next firmware will raise the ceiling on recording time just a bit, and then also offer the light raw recording mode (allow the heat to build up just a bit more, and reduce file I/O to the card).
I bet they have no way to cool this off any further without downgrading eye AF, or redesign it to be more efficient with system resources. Just a guess though at this point. I wonder if they have the software folks digging through all the code to find ways to reduce CPU drag. Been there, done that...
So I just tried shooting 4k/120 using paused servo (single shot focus) and I shot quickly for 22 minutes, yielding 68 clips of 10-15 second durations, before the camera stopped recording. The card was hot when I pulled it out.
Now, this is not my style of capturing footage. But if I had conducted a capture session and came back with 68 clips of 4K/120, I would be more than satisfied.
So Canon says 15 minutes, you got 7 more minutes... Tough to tell anything exactly about this, unless you can get the camera cooled down to what it was before, and do it again, but have AI Servo one, eye AF, etc and see if you get the same 22 minutes.
If only Canon had added a 2nd Digic (even an older one) and offset functions like IBIS, metering, etc, perhaps we wouldn't be having these discussions. I think that DigicX processor is just being asked to do too much.
I really hope I am wrong and there are just some stupid limits in the firmware preventing things, but I highly doubt it.
The R5 owner who did the previous experiment on my behalf just did a new experiment that resulted in a total 8K recording time of 69 minutes! It involved deep-cooling the camera before starting shooting but there are other differences as well that are looking very interesting. Should have the data fully analyzed and understood later today.
