VSync firing 255 times more frequently than normal during "display off" sleep on Windows
Categories
(Core :: Graphics, defect, P3)
Tracking
()
Performance Impact | low |
People
(Reporter: colormatch, Unassigned)
References
(Blocks 1 open bug)
Details
(Keywords: perf:resource-use)
User Agent: Mozilla/5.0 (Windows NT 6.2; Win64; x64; rv:69.0) Gecko/20100101 Firefox/69.0
Steps to reproduce:
Set laptop's display to turn off, but not to sleep.
Open Firefox (I have a few tabs open), and wait for the display to turn off.
A few minutes after display gets turned off, the fans start and gradually go to max, which indicates CPUs' throttling. As soon as you "wake up" the system, the fans gradually slow down and turn off as the CPUs cools down as the usage is back to "normal".
OS: Windows, FF 69 - stable.
Here is a link from Firefox Profiler, which I stopped just after I moved the mouse and the system "woke up": https://perfht.ml/35cEBKZ
Comment 1•5 years ago
•
|
||
Cannot make heads or tails from the perf report :(. Mike, bothering you again to take a look at one perf. profile.
Comment 2•5 years ago
|
||
Hi colormatch,
Do you need to have a particular website open in order for you to see the spin-up occur when the matchine goes to sleep? If you just have about:home loaded, or about:blank, in a single tab, do you notice the same issue?
I ask, because what I do see is an animation running inside one of the pages running in content process 7156, and I'm wondering if that's related.
Reporter | ||
Comment 3•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #2)
Hi colormatch,
Do you need to have a particular website open in order for you to see the spin-up occur when the matchine goes to sleep? If you just have about:home loaded, or about:blank, in a single tab, do you notice the same issue?
I ask, because what I do see is an animation running inside one of the pages running in content process 7156, and I'm wondering if that's related.
the "active tabs" (there are a few which were not closed from previous runs, but not "loaded"/made visible either), are of the type of facebook, google pages, etc... nothing specific with "animation", games, or similar.
again, note that while you are working, even if firefox is in the background, there is no CPU increase, and it only happens when you leave the laptop to idle (it took me while to figure that firefox is the responsible).
I have "Restore previous session" ON, will:
- opening a new tab
- closing firefox
- starting firefox (on the empty tab) and not clicking on any other tabs
- then idling and checking, do it for you?
Comment 4•5 years ago
|
||
(In reply to colormatch from comment #3)
I have "Restore previous session" ON, will:
- opening a new tab
- closing firefox
- starting firefox (on the empty tab) and not clicking on any other tabs
- then idling and checking, do it for you?
Yes, this should give me the information I need, thanks!
Reporter | ||
Comment 5•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #4)
(In reply to colormatch from comment #3)
I have "Restore previous session" ON, will:
- opening a new tab
- closing firefox
- starting firefox (on the empty tab) and not clicking on any other tabs
- then idling and checking, do it for you?
Yes, this should give me the information I need, thanks!
ran one test as described above, and one with additional "private browsing", and both were ok (no CPU throttling)
Comment 6•5 years ago
|
||
Okay. The next step is to figure out which site is the one doing the work in the background when the machine goes to sleep. Are you able to choose, say, half of the tabs in your tabstrip, and restore them, and see if the problem recurs?
If so, we can "bisect" by restarting the browser, and then choosing half of the tabs that still cause the issue, and repeating, until we're down to a minimal set.
Once we have that, if you're willing to share the site addresses, we can start trying to figure out what these sites are doing to cause Firefox to spin up like this.
Reporter | ||
Comment 7•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #6)
Okay. The next step is to figure out which site is the one doing the work in the background when the machine goes to sleep. Are you able to choose, say, half of the tabs in your tabstrip, and restore them, and see if the problem recurs?
We can say that there is at least one for sure - facebook.
I took a chance and tried with facebook only (no other tabs are active).
Here is the result: https://perfht.ml/33zSxNe
(In reply to colormatch from comment #7)
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #6)
Okay. The next step is to figure out which site is the one doing the work in the background when the machine goes to sleep. Are you able to choose, say, half of the tabs in your tabstrip, and restore them, and see if the problem recurs?
We can say that there is at least one for sure - facebook.
I took a chance and tried with facebook only (no other tabs are active).
Here is the result: https://perfht.ml/33zSxNe
so colormatch. I have had this problem for at least 2 yrs now and decided to check into it. recently I thought its website / link based just like you, but it is not. Reason being that on a fresh window with nothing on it except firefox, this problem does NOT show up. this is caused by something installed on our computer, combined together with the website triggering this, im not sure what it is but its doing writes onto your computer. The I/O and other stuff combined causing high CPU loads or the reverse.
the I/O is most likely not caused by cookies or session restore because it would not explain why new window with firefox of the same website have no such issue, it is something else.
this is my original video recorded https://vimeo.com/367007343
this is new screenshot showing when CPU goes up, I/O writes goes up at sametime (my cache is on ram/ramdisk, which might be the cause as loads on ram means loads on CPU, quite heavily too). https://imgur.com/a/DsDiZzP
I even tried to uninstall every netframework, c++ VC etc, no luck. gonna try process of elimination and hope to hit the jackpot eventually.
Comment 10•5 years ago
|
||
to add to my previous post. this is the link among some i have issue with, firefox's support page. https://support.mozilla.org/en-US/questions/1248277?page=2
also, this problem will NOT come up if:
- firefox is minimized to system taskbar
- firefox is not minimized but has a tab that doesnt cause this issue
- fresh windows with firefox and these links won't have this issue
what changes does not affect outcome of this issue:
- changing multi threaded process to as low as 1 does not reduce cpu load
- remove hardware acceleration changes nothing
- fresh profile
- reinstall firefox
- waterfox
- developer firefox
- nightly
- private window
- no extensions
Comment 11•5 years ago
|
||
Mike did the profile in comment 7 shed any light?
colormatch, are you still hitting this issue?
Comment 13•5 years ago
|
||
Looking at the profile from comment 7, it looks like an animation is occurring in the parent process, which is causing the compositor to composite:
Unfortunately, since this is a release instance of Firefox, I'm not able to get much more information besides that.
colormatch, do you recall seeing any part of the browser animating when this occurred?
Reporter | ||
Comment 14•5 years ago
|
||
The issue is still present.
version 70.0.1
Honestly, it seems like Firefox is mining crypto almost as soon as the screen goes dark.
@Mike, again, this happens after the display is turned off.
As soon as I move the mouse and the system turns on the monitor the CPU load drops, and nothing "out of the ordinary facebook" is there to see.
Just for kicks, give it a try.
- Set your screen to turn off after 5min of inactivity (don't put it to sleep, just turn off the monitor).
- Go to facebook (since I can confirm it happened there), turn on Gecko Profiler
- wait 10~15 minutes and check the profiler
If this doesn't happen, let me know what to try to give you more data so you can find the problem.
It's a pain to have forgotten Firefox open, just to come after a few ours and find your laptop have been overheated all that time because you forgot to turn off Firefox.
Reporter | ||
Comment 15•5 years ago
|
||
- of course, that means to wait 5~10min after the screen turns off
Comment 16•5 years ago
|
||
I'm afraid I don't have a Facebook account to help confirm / reproduce this. Going to redirect to our QA team to see if they can reproduce and provide a profile from a Nightly.
Hi brindusat, would you (or anyone on your team) have a moment to help try to reproduce this on Nightly, and provide a Gecko Profile if possible?
Updated•5 years ago
|
Reporter | ||
Comment 17•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #16)
I'm afraid I don't have a Facebook account to help confirm / reproduce this. Going to redirect to our QA team to see if they can reproduce and provide a profile from a Nightly.
I just tried it on Nightly - you don't need an account, just visit a page like " https://www.facebook.com/ZeroHedge/ " scrolled down a bit, had the screen turn off in 1 min, turned on the profiler.
a few seconds just after the screen turned off the fans started working, but I can't give you the profile as after I moved the mouse, Firefox crashed.
Reporter | ||
Comment 18•5 years ago
|
||
Managed to repeat it on reddit, on which happened to have a video preview on screen.
(It could be strictly animation/video related, as repeated scrolling and testing of the FB page didn't reproduce the problem every time.)
Screen turns off at about 60s.
here's the captured profile: https://perfht.ml/2OtUIfC
Comment 19•5 years ago
|
||
I can't see any declarative animations on that page (other than a very brief one when entering/leaving the top of the page) or the linked profiles so this seems to be unlikely to be "DOM: Animation".
Updated•5 years ago
|
Comment 20•5 years ago
|
||
I've tried to reproduce the reported cpu high usage throttle on a i3 laptop Win10 Pro x64 with Firefox Nightly 72.0a1 2019-11-20, but no luck at this point - I didn't notice any cpu throttle or obvious spikes in the perf. profile. It's either something windows 8 specific or hardware related or might not reproduce with Fx72.
I'm gonna follow-up tomorrow with a test using a Fx 70 release on the w10 laptop, then a desktop windows 8 with Fx72 or Fx70 (we don't have any laptops around with w8) and move forward from that.
Reporter | ||
Comment 21•5 years ago
|
||
(In reply to Adrian Florinescu [:adrian_sv] from comment #20)
I've tried to reproduce the reported cpu high usage throttle on a i3 laptop Win10 Pro x64 with Firefox Nightly 72.0a1 2019-11-20, but no luck at this point - I didn't notice any cpu throttle or obvious spikes in the perf. profile. It's either something windows 8 specific or hardware related or might not reproduce with Fx72.
The last profile was taken with nightly version 72.0a1 (2019-11-20) (64-bit)
Comment 22•5 years ago
|
||
Thanks for the profile, colormatch. Looking at it, it looks like the compositor starts animating something in this section, right towards the tail end of the profile (about 26 seconds worth):
Reporter | ||
Comment 23•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #22)
Thanks for the profile, colormatch. Looking at it, it looks like the compositor starts animating something in this section, right towards the tail end of the profile (about 26 seconds worth):
What you are seeing in the profile - after me scrolling reddit I let it without interaction so the monitor to turn off in 60 seconds.
So the "animating" you see happens just after the monitor turns off, and ends as soon as I move the mouse, and the monitor lights up, and I turned off the profiler.
After turning the monitor on, nothing on the reddit page was changed - it was the same with the video-loop preview showing in one of the posts, so there was no "real animation", which happened that can account for what you see, and for the CPUs throttling, and everything turned back to what you see in the first 60 second of the profiler.
Comment 24•5 years ago
|
||
Follow-up attempts on a desktop machine still didn't yield any results on my side. Mike, should I still try to reproduce this issue, or did you get enough info from comment 21/comment 22 ? Please re-set the need-info if necessary.
Comment 25•5 years ago
|
||
I don't have enough information, no. The profile that was gathered by colormatch is missing some pretty critical information - for example, markers, and labels.
How are you gathering this profile, colormatch? Are you using the Gecko Profile add-on or the built-in Profiler Toolbar Icon (activate-able from the Web Developer menu), or some other mechanism? Are you using the default settings, and default publishing settings?
Comment 26•5 years ago
|
||
Ok, let me put this back in my to-do list, then, although first order of business to be of any help would be to identify an environment in which I can reproduce the CPU throttle.
Reporter | ||
Comment 27•5 years ago
|
||
(In reply to Mike Conley (:mconley) (:⚙️) (Wayyyy behind on needinfos) from comment #25)
I don't have enough information, no. The profile that was gathered by colormatch is missing some pretty critical information - for example, markers, and labels.
How are you gathering this profile, colormatch? Are you using the Gecko Profile add-on or the built-in Profiler Toolbar Icon (activate-able from the Web Developer menu), or some other mechanism? Are you using the default settings, and default publishing settings?
I was using the add-on, and basically checked all possible boxes (minus DNS Resolver, but had checked the privacy box) to collect maximum info, but uploaded the profile without the screenshots.
What settings should I use so the profile can be most useful?
Reporter | ||
Comment 28•5 years ago
|
||
would this one do? https://perfht.ml/35o7oLJ
Reporter | ||
Comment 29•5 years ago
|
||
here's one profile with screenshots (and privacy turned off): https://perfht.ml/2rleMIR
Comment 30•5 years ago
|
||
Yeah, this last one is great, thank you colormatch!
This is the profile I'm looking at: https://perfht.ml/2rleMIR
Hey mstange... do you know why the compositor might start to get really active when the display goes to sleep? I see a video animating in the screenshots, but it's animating even before the compositor starts to get really busy (around the 82 second mark in the profile).
Updated•5 years ago
|
Comment 31•5 years ago
|
||
It looks like this is a problem with our vsync notification mechanism. We seem to be firing vsync notifications at a crazy high rate!
Here's a view of the profile that shows a 16ms time span: https://perfht.ml/2KIcHxC - 225 vsyncs!
Rather than getting one vsync every 16.6ms, we seem to get one vsync every 0.0655ms on average. That's 255 times as many as normally.
(I used the following code on the web console in the profiler in order to compute the marker frequency: (filteredMarkers[filteredMarkers.length - 1].start - filteredMarkers[0].start) / (filteredMarkers.length - 1)
)
Comment 32•5 years ago
|
||
[Markus, please adjust my qf assessment if you think it's off-base]
Updated•5 years ago
|
Comment 33•5 years ago
|
||
so will this be fixed on the next version of firefox? or will developer simply force user to upgrade to windows 10 again like the other bug with built in media decoder?
Reporter | ||
Updated•5 years ago
|
Reporter | ||
Comment 34•5 years ago
|
||
I can confirm that is happening on Windows 8 too.
Comment 35•5 years ago
•
|
||
Removing my NI, since there seems all things came together already.
Comment 36•4 years ago
•
|
||
Removing my NI again :), seems it failed first time.
Comment 37•4 years ago
|
||
any update on this? this gonna get fixed?
Comment 38•4 years ago
|
||
The priority flag is not set for this bug.
:jbonisteel, could you have a look please?
For more information, please visit auto_nag documentation.
Comment 39•4 years ago
|
||
Tania, would it be possible to get QAs help to see if this can be reproduced?
Updated•4 years ago
|
Comment 40•4 years ago
•
|
||
Hi Rares, as communicated yesterday, please assign a team member to investigate this.
Comment 41•4 years ago
|
||
irrelevant |
Windows 10 - 1809, turn off screen - 60s;
https://perfht.ml/36GuFt9
(filteredMarkers[filteredMarkers.length - 1].start - filteredMarkers[0].start) / (filteredMarkers.length - 1)
11.662230372807116
windows 10 1903, turn off screen - 60s;
(filteredMarkers[filteredMarkers.length - 1].start - filteredMarkers[0].start) / (filteredMarkers.length - 1)
4.3515750929368116
windows 8 version 6.2 build 9200, turn off screen - 60s;
(filteredMarkers[filteredMarkers.length - 1].start - filteredMarkers[0].start) / (filteredMarkers.length - 1)
28.03776498607597
I've assumed that reproducing the issue would be signaled by an increase in the vsync fire rate as described in comment #31, but that doesn't seem the case on the 3 machines I've tested upon. Markus, can you confirm that I'm reading the results correctly?
Reporter | ||
Comment 42•4 years ago
|
||
(In reply to Adrian Florinescu [:adrian_sv] from comment #41)
I've assumed that reproducing the issue would be signaled by an increase in the vsync fire rate as described in comment #31, but that doesn't seem the case on the 3 machines I've tested upon. Markus, can you confirm that I'm reading the results correctly?
Adrian, those profiles are less than 30 seconds long, and your monitor is set to turn off (go to sleep) in 60s.
Shouldn't your profiles be at least 90s long, since the issue shows up after the monitor goes dark?
Comment 43•4 years ago
•
|
||
(In reply to colormatch from comment #42)
Right, seems I was going at it backwards. Not entirely sure I'm approaching it correctly even now. Redid the test with properly sized profiles and I'm also seeing a vsync fire rate increase after 60s of sleep time on win 8: https://perfht.ml/2PXfrc6 , which I don't see on the Win 10 baseline.
Windows 8 version 6.2 build 9200, turn off screen - 60s; sleep time ~180s - https://perfht.ml/2PXfrc6
Windows 10 - 1809, 73.0a1 20191216214733, turn off screen - 60s; sleep time ~60s - https://perfht.ml/38OO2C4
** Edit after discussing the results with :mconley and the two above profiles still do not reproduce the problem.
Reporter | ||
Comment 44•4 years ago
|
||
** Edit after discussing the results with :mconley and the two above profiles still do not reproduce the problem.
Make sure you have some video element (like the video preview seen in this profile https://perfht.ml/2rleMIR ).
The CPUs throttle like there's no tomorrow. You can't miss it.
Updated•4 years ago
|
Comment 45•4 years ago
|
||
(In reply to Adrian Florinescu [:adrian_sv] from comment #43)
(In reply to colormatch from comment #42)
Right, seems I was going at it backwards. Not entirely sure I'm approaching it correctly even now. Redid the test with properly sized profiles
and I'm also seeing a vsync fire rate increase after 60s of sleep time on win 8: https://perfht.ml/2PXfrc6 , which I don't see on the Win 10 baseline.Windows 8 version 6.2 build 9200, turn off screen - 60s; sleep time ~180s - https://perfht.ml/2PXfrc6
Windows 10 - 1809, 73.0a1 20191216214733, turn off screen - 60s; sleep time ~60s - https://perfht.ml/38OO2C4
** Edit after discussing the results with :mconley and the two above profiles still do not reproduce the problem.
I agree with mconley's assessment. The problem was not happening in these profiles. The first profile has vsync firing at the regular 16.66ms rate, and the second profile doesn't have vsync firing for most of the time at all, which is even better.
We have yet to find a way to reproduce colormatch's problem.
Comment 46•4 years ago
|
||
(In reply to Markus Stange [:mstange] from comment #45)
(In reply to Adrian Florinescu [:adrian_sv] from comment #43)
(In reply to colormatch from comment #42)
Right, seems I was going at it backwards. Not entirely sure I'm approaching it correctly even now. Redid the test with properly sized profiles
and I'm also seeing a vsync fire rate increase after 60s of sleep time on win 8: https://perfht.ml/2PXfrc6 , which I don't see on the Win 10 baseline.Windows 8 version 6.2 build 9200, turn off screen - 60s; sleep time ~180s - https://perfht.ml/2PXfrc6
Windows 10 - 1809, 73.0a1 20191216214733, turn off screen - 60s; sleep time ~60s - https://perfht.ml/38OO2C4
** Edit after discussing the results with :mconley and the two above profiles still do not reproduce the problem.
I agree with mconley's assessment. The problem was not happening in these profiles. The first profile has vsync firing at the regular 16.66ms rate, and the second profile doesn't have vsync firing for most of the time at all, which is even better.
We have yet to find a way to reproduce colormatch's problem.
when using nvidia GPU getting this problem with firefox if it is to do with vsync. i have the same problem as him so there is definitely an issue here.
Reporter | ||
Comment 47•4 years ago
|
||
Markus, it's not "my problem", it's Firefox's problem ;)
I don't question that your tests don't reproduce the problem, yet this is a real problem, which has gone under the radar (and let's hope someone hasn't lost a graphics card or cpu because of it), so let's figure out what is different in your setup.
Is your testing methodology correct:
-
is anything "moving" on the page Firefox has loaded (you can see a video, canvas is blending images, etc)
here is a page with "something moving on it" : https://www.dailymotion.com/us -
is you monitor turning OFF after 60s (the problem happens when the computer turns off the display)
If you have done this properly and you still don't see any issues - can it be hardware related?
Is hardware acceleration used?
I have an nVidia, driver version 425.31
Comment 48•4 years ago
•
|
||
(In reply to colormatch from comment #47)
Is your testing methodology correct:
is anything "moving" on the page Firefox has loaded (you can see a video, canvas is blending images, etc)
here is a page with "something moving on it" : https://www.dailymotion.com/usis you monitor turning OFF after 60s (the problem happens when the computer turns off the display)
If you have done this properly and you still don't see any issues - can it be hardware related?
Is hardware acceleration used?
I have an nVidia, driver version 425.31
I think it is:
- Yes, for the test below used https://www.dailymotion.com/us with some random muted video playing. (on the tests before, used facebook)
- Yes, the screen sleep is set to 1 min.
- New profiles used, HW acceleration on by default in Nightly.
- It's quite possible to be hardware related, even narrower: might be specific to a certain combination of HW / OS / drivers
colormatch, tallyo, could you please help further by listing the hardware (processor/video card and if it's a desktop/laptop) and exact version of OS both of you can reproduce this problem on. I'll try to find and use something closely matching your HW. Also, something is intriguing me from comment 10 :
also, this problem will NOT come up if:
- fresh windows with firefox and these links won't have this issue
so, could not hurt to double check if it is equally easy for you to reproduce the issue on a new clean profile.
I've attempted to reproduce the issue with a nvdia GPU on win 10 machine, and the profile seems to yeld nothing special http://www.perfht.ml/38SIM0g (cannot upload the profile with the screenshots to prove animation, since it's too big for upload - might've been already blacklisted from perfht.ml for trying).
Windows 10 Pro x64
GeForce 1050
Driver version: 441.41
Nightly 73 - 2019.12.17
Reporter | ||
Comment 49•4 years ago
|
||
The posted profile was taken on laptop, page reddit, showing some random video clip in the feed:
Windows 8 x64 / Intel i7
GeForce GT 740M
Driver version: 425.31
Nightly 72.0a1 (2019-11-20) (64-bit)
Reporter | ||
Comment 50•4 years ago
|
||
- just to make clear - the 60s screen sleep doesn't matter, it happens if it is 15min, it's all the same.
(testing with 60s is just faster and saves the wait)
Reporter | ||
Comment 51•4 years ago
|
||
Based on the discussion so far - instead of looking to find the exact driver/hardware so you can replicate the issue, you should just check your code.
Is there a sanity-check for setting the values for V-sync?
- If "yes" - why does it fail?
- If "no" - implement a sanity-check.
Since V-sync should be dependent on device's refresh rate, and you never know if some driver from the past or future will report 0Hz when the monitor goes to sleep, this "fix" (sanity-check) should not be hardware/driver dependent.
Ho Ho Ho
Happy Holidays :)
Comment 52•4 years ago
|
||
(In reply to colormatch from comment #51)
Based on the discussion so far - instead of looking to find the exact driver/hardware so you can replicate the issue, you should just check your code.
You're welcome to take a look, the code is here. My guess is that the call to DwmFlush()
takes a much shorter time than expected. I see some error correction code nearby but maybe it's not enough.
Reporter | ||
Comment 53•4 years ago
|
||
I think the code here does not perform sanity check on the received values for the refresh rate.
// We get the rate in hertz / time, but we want the rate in ms.
float rate = ((float)refreshRate.uiDenominator / (float)refreshRate.uiNumerator) *1000;
So you get something 0Hz (for turned-off device), maybe divide by 0, and things get funky.
Reporter | ||
Comment 54•4 years ago
|
||
I know MS assures you that this: QueryPerformanceFrequency(&frequency); will never fail (after XP) and return 0, but if you ask me, this is what happens, and then on Line 1798 :
int64_t usAdjust = (adjust * microseconds) / frequency.QuadPart;
Comment 55•4 years ago
|
||
(In reply to Adrian Florinescu [:aflorinescu] from comment #48)
(In reply to colormatch from comment #47)
Is your testing methodology correct:
is anything "moving" on the page Firefox has loaded (you can see a video, canvas is blending images, etc)
here is a page with "something moving on it" : https://www.dailymotion.com/usis you monitor turning OFF after 60s (the problem happens when the computer turns off the display)
If you have done this properly and you still don't see any issues - can it be hardware related?
Is hardware acceleration used?
I have an nVidia, driver version 425.31I think it is:
- Yes, for the test below used https://www.dailymotion.com/us with some random muted video playing. (on the tests before, used facebook)
- Yes, the screen sleep is set to 1 min.
- New profiles used, HW acceleration on by default in Nightly.
- It's quite possible to be hardware related, even narrower: might be specific to a certain combination of HW / OS / drivers
colormatch, tallyo, could you please help further by listing the hardware (processor/video card and if it's a desktop/laptop) and exact version of OS both of you can reproduce this problem on. I'll try to find and use something closely matching your HW. Also, something is intriguing me from comment 10 :
also, this problem will NOT come up if:
- fresh windows with firefox and these links won't have this issue
so, could not hurt to double check if it is equally easy for you to reproduce the issue on a new clean profile.
I've attempted to reproduce the issue with a nvdia GPU on win 10 machine, and the profile seems to yeld nothing special http://www.perfht.ml/38SIM0g (cannot upload the profile with the screenshots to prove animation, since it's too big for upload - might've been already blacklisted from perfht.ml for trying).
Windows 10 Pro x64
GeForce 1050
Driver version: 441.41
Nightly 73 - 2019.12.17
my comment on fresh window with fresh firefox profile is actually tested on windows 8.1, not windows 8, which am running as primary while 8.1 as duo boot.
I am same with color match.
specs:
windows 8
GeForce 1070
Driver version: 441.66 (even with older version 361 has problem too)
Firefox pre quantum, Firefox Quantum, Firefox v69, v70 etc all tested, all the same.
I think it is to do with windows 8
Comment 56•4 years ago
|
||
oh to add, I also tried Gsync enabled/disabled, problem still exist.
Reporter | ||
Comment 57•4 years ago
|
||
int64_t usAdjust = (adjust * microseconds) / frequency.QuadPart;
QueryPerformanceFrequency function
"Return value
If the installed hardware supports a high-resolution performance counter, the return value is nonzero.
If the function fails, the return value is zero."
Any word on fixing the division by zero?
Updated•4 years ago
|
Updated•4 years ago
|
Comment 58•4 years ago
|
||
its been over a year since this bug opened, and this issue present all from several years back.
when will this be fixed? any update on this?
Updated•2 years ago
|
Updated•2 years ago
|
Comment 59•1 year ago
|
||
Florian, is this something you'd be interested in looking into?
Comment 60•1 year ago
|
||
(In reply to Markus Stange [:mstange] from comment #59)
Florian, is this something you'd be interested in looking into?
If I could reproduce, sure. I tried on Windows 11 with a page with a large CSS animation. This is the power profile I got: https://share.firefox.dev/3m2LueN
We animate at 60Hz when the screen is on, and at 22Hz when the screen is off. That seems a lot more than what we need (should we stop animations entirely when the screen is off?), but I can't reproduce the crazy high refresh rate reported by the bug reporter.
In my profile, 90s after the screen turned off, there's significant CPU use. I think this isn't related to Firefox (Windows Update, malware detection service, etc...)
When the screen turns off, the parent process receives a WM_SYSCOMMAND message with wParam = SC_MONITORPOWER, and the value "2 (the display is being shut off)". Maybe we should use it?
Description
•