820-3662 - Sudden power off

[smiba]

Hey everyone,

A Macbook 15" Late 2013 model here with no visible signs of water damage. The device suddenly powers off after using it for somewhere between 10 to 40 minutes.
Having a high load running on it won't charge the issues (Furmark + Prime95 for 20 minutes would not cause the device to shutdown) but just idle running a youtube video made it crash at one point.
The power off is not registered by the system as a kernel panic and does not show signs of the keyboard incorrectly pressing the power button. (The Shutdown window does not appear, as it would when the power button is being held down). The system does not reboot after the power off has occurred but would stay in S5

Is this model also affected by bad soldering? If so what chip would be the cause of this?
Any idea where to look?

Cheers!

 

[JohnB8812]

Duke helped me with one of these a while back. It seems the 3662 and 3787's have unexpected shutdown issues, and that there isn't a solution yet.

 

[dukefawks]

The usual 3662 issue. I have one here to right now. I really have no time to dig into this problem at the moment.

 

[smiba]

Hey duke,

Is there any information or speculation about this issue so far? I have some free time on my hands this week so I woudn't mind experimenting around, who knows what I might find.
(I tried to DM you about this but it seems your "Quota" has been reached and it won't let me send)

Thanks in advance!

 

[dukefawks]

It must be something is the power rails to the CPU. I have 0 time to deal with this at the moment as repairs are stacked up here.

First to see which rail causes this hook up your 2 channel scope to both CPUVR_PGOOD and ALL_SYS and trigger on the falling edge to see which one drops off first.

 

[smiba]

No Title

ALL_SYS_PWRGD drops before CPUVR_PGOOD drops.

I guess at this point its just a matter of going through a whole bunch of power signals to hopefully see which one goes first? CPUVR_PGOOD would be much easier

 

[MOberdick]

Try retinning all the pads on U7200. I am currently experimenting with a board like this and after checking the chip at all 4 angels and making sure each joint was solid it seems to have stopped happening.. I am still testing however just figured it was worth putting out there for someone else to try as well... If It continues to work for over 24 hours I will report back.

 

[dukefawks]

Don't heat anything up on the board as this may mask the problem again for weeks.

Now comes the shitty part in separating the ALL_SYS signals and see which one falls first. It seems like ALL_SYS is not a "hard" drop and it gradually drops to zero. What is your time base in that picture?

 

[smiba]

Let me know if it works for you. I'm currently keeping track of any power lines dropping just before ALL_SYS_PWRGD does before I solder anything

I'm sceptical about it being a contact issue as the problem gets more frequent once it keeps crashing and you keep rebooting it. But I'm hoping
820-3662 board as well?

 

[MOberdick]

Yep same board 820-3662. Also I didn't use hot air duke, just my micropencil to connect the joints with fresh solder.. shouldn't have a large impact. So far I've been able to run migration assistant, complete an external drive restore, and an internal in-place software upgrade without problem. I have this thread bookmarked however and if for some reason it does come back in the future for anything I will be sure to keep everyone in the loop.

 

[smiba]

What is your time base in that picture?

4μs/div by the way

The wait is the worst, at this rate its going to take a couple of days...I already confirmed PP1V35_S3RS0_CPUDRR not being at fault and I'm currently testing PP3V3_S0.

Are there any power lines you recommend taking a look at first?

(On the picture the voltages you see on the side are for Probe A and I didn't align B. Sorry for the unclarity. Probe B is seeing normal 1.35V)

 

[dukefawks]

U7200 is on my suspect list but I would have expected CPUVR_PGOOD to drop before ALL_SYS. Resoldering it with an iron can not hurt, just don't use hot air or it may disturb the actual problem.

Do remember that this issue only happens when the machine is "idle" or light load. As soon as you hammer the CPU it keeps running fine. So no use is stress testing with this problem.

 

[MOberdick]

In the situation with the machine I have, it seems to happen when firefox is opened (usually playing a youtube video) and then there is graphical activity on the dock from the magnification feature.. a few swipes back and fourth with the cursor and it'd crash out. Tried that for about an hour on a cold boot. doing the more extensive stuff now.. I'll probably let it idle overnight tonight on a video or something and retest everything out in the AM tomorrow.

 

[smiba]

Can confirm that low load is more likely to let it crash. In my case I often make it stream some music from my local library and thats been pretty effective at making it crash. Stuff like YT also works alright
Stress testing the device will not make it crash at all.

Trick seems to be not to blast the CPU but just keep it at a couple of cpu%. But this is purely speculation based on just a couple of samples so far though

EDIT: Its also not PP3V3_S0. Testing it with PP3V3_S5 now

 

[MOberdick]

Got it. I have it off and will let it cool off for the next 2 hours and retest cold before i leave the office.
\

 

[smiba]

PP3V3_S5 is looking alright as well.
I think I saw the scope catch multiple falling edge triggers though. So I'm running it again at 20ms/div to see if I can catch those fluctuations. But I might have been wrong

EDIT: The system seems to be very sensitive to my desk light? If I have it close to the keyboard and I disable the light the system will crash and freeze (But not like the random crashes would cause). I think this is what I saw with the multiple falling edge triggers as it happened while I stopped the lamp. I've never noticed this before but its probably unrelated as its just a shitty light. I'll keep on searching
EDIT2: Its probably noise its picking up on the wires that are sticking out off the motherboard.... Move along. Running the PP3V3_S5 test again as its been skewed by me turning off the light and thus causing a pulse on its power

 

[dukefawks]

When separating the PGOOD signals make sure there still is a pull up resistor. All the signals on the board only have one pull up resistor for all.

 

[smiba]

When separating the PGOOD signals make sure there still is a pull up resistor. All the signals on the board only have one pull up resistor for all.

Will do, I think the leakage of the transistor inside of most ICs will make a PGOOD signal float around ground level without pull-up as well.

After crashing the system because of a voltage spike on PP3V3_S5 caused by a pulse from my desk light I have not been able to get it to crash again....... Been running without crash for 2 hours so far, I've tried YT, Music and just idleing on the desktop.
I'll try again tomorrow

 

[dukefawks]

That internal pull up is very weak and may affect measurements as you saw with your light.

 

[MOberdick]

Update: Macbook ran overnight playing from iTunes library without issues. Came in and it was still powered on. Currently back to my youtube and dock magnification testing around.. probably gonna call it good and send it out though. If it comes back I will let yall know.