[W530] Freezes and Random Reboots. CPU or Motherboard?

[bit_twiddler]

I just thought of something else...

It could be your power supply, especially (but not only) if it crashes under high
load conditions. You are using the 170W supply, right? Perhaps someone
could lend you another power supply to test.

[bit_twiddler]

And, here is something else to try: taking the battery out.
Sometimes the battery can be defective and cause the
machine to shut down, as in:

http://www.sevenforums.com/general-disc ... power.html

[RasmusP]

And, here is something else to try: taking the battery out.

I tried this already. I don't seem to observe a consistent pattern wrt. to the battery.

It could be your power supply, especially (but not only) if it crashes under high
load conditions. You are using the 170W supply, right?

I use the big brick one, yeah. The thing is, when I stress test the cpu nothing happens. When I use the NVIDIA card nothing happens. These should be more stressful situations, no? Anyway, I'll see if one of my colleagues have got a big enough charger with the right plug. Or maybe run the test with battery only.

[RasmusP]

I came across this thread on the Intel board. The issue OP experienced are similar to what I experienced: namely it seems to increase with no. of attached memory blocks.
The OP seems to solve the issue by lowering the the voltage of the memory. Is that possible wiht a Thinkpad?

Also, I found the Lenovo part# of my memory is 0A65724 (edited).

Lenovo has asked me install win7 and confirm that the problem also happens there. There goes my weekend.

[bit_twiddler]

AFAIK, lowering memory voltages requires a special motherboard.
However, there are ways to undervolt a CPU, although I don't know
if Ivy/Sandy Bridge CPUs support this.

On Windows there are programs to undervolt the CPU, but
they are generally used by overclockers and other
practitioners of Voodoo - certainly nothing that I would
advise for a working system.

You can control Ivy Bridge w/ P-states on later kernels,
but I believe that you posted that you are not experiencing
any out of spec temperatures since you replaced the fan.

Please post your results after installing Win7, at least that may
help us rule out the OS.

You might compare your results to this guy's:
http://www.thinkscopes.com/blog/2013/04 ... st-result/

While there are no temperature sensors for the memory,
at least you could see if your other readings are close to his.

[bit_twiddler]

Just out of curiosity, do you happen to know if your RAM is
DDR3 or DDR3L? DDR3L would generate less heat
because it runs at a lower voltage (1.35V vs 1.5V).

Your W530 should run with either, so it may be
an option at some point.

[RasmusP]

AFAIK, lowering memory voltages requires a special motherboard.

OK. So probably no point in testing that.

You can control Ivy Bridge w/ P-states on later kernels,
but I believe that you posted that you are not experiencing
any out of spec temperatures since you replaced the fan.

I don't know that there is a point in undervolting the CPU... lm_sensors report a high temperature of 95 or so, and I haven't seen it above 90.

Just out of curiosity, do you happen to know if your RAM is
DDR3 or DDR3L? DDR3L would generate less heat
because it runs at a lower voltage (1.35V vs 1.5V).

Sorry, I posted the wrong part number. The RAM is this. "1600MHZ DDR3 204-Pin SODIMM Notebook Memory". So I guess DDR3.

[bit_twiddler]

That does look like DDR3. So, trying DDR3L is an option, if
you can't track down the problem to something else.

[RasmusP]

So my plans of installing windows got messed up 'cause I didn't immediately had a single partition with enough free space.

However, I did run more systematic RAM tests. Initial results suggested that module 1 was causing trouble whenever it was paired with another module... But I stumbled upon a configuration in the end that seems stable (I've run Furmark for hours and but also Furmark and mprime for a while without a crash). Of course the original configuration of all four modules as very unstable. I don't fully understand what is going on.

I hope this is the end of woes.

Code: Select all

A test is at least 10 min of fullscreen Furmark.  
Reboot between tests.
✓: No crash; ✗: Crash
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Result        Top         Bottom        
          Lower  Upper  Left  Right 
────────────────────────────────────
 ✓          3                       
 ✓ ✓        4                       
 ✓ ✓        2                       
 ✓ ✓        1                       
 ✓          2      4                
 ✗ ✓        2      3                
 ✗ ✗ ✓ ✗    1      2                
 ✗ ✗ ✗      1      3                
 ✓ ✗ ✗ ✗    1      4                
 ✓ ✗ ✗      1            3      4   
 ✓ ✓        1      4     3          
 ✗          1      2     3          
 ✓ ✓        1      2     3      4   
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Thanks for all the help!
Rasmus

[bit_twiddler]

Hopefully, you've nailed it. At least you've narrowed it down to something
flaky in your memory sticks. Personally, I would pick up another
memory stick or 2 and replace any modules that appear to
be marginal, just to make everything more solid.

[RasmusP]

You are probably right. It's may be a ticking bomb.

Maybe I can ask Kingston to honour their "life time warranty" (I've never had to deal with them so I don't know how easy or hard this is).

Thanks for all the help again!

[RasmusP]

Hmm, in fact the problem has not gone away. At the moment I'm running with only two memory blocks and I'm still experiencing a problem. This is highly frustrating.

I found a cheap replacement battery and will see if that solves my issue, but I don't understand how that could result in a freezing. Otherwise I guess I will pick up a cheap disk to run windows on for Lenovo to do further support...

[bit_twiddler]

Some real gremlin is at work on your W530 - have your tried an exorcist?
(Just kidding, in my experience most exorcists are charlatans...)

Which ram slots are you populating?

Also, and this is a shot in the dark, are you always using the machine
at the same location (work, home, coffeeshop, etc?) when this occurs?
You could be suffering from hash on the AC line or an undervoltage
condition. Just to be clear, is the machine:

A. Freezing, you can see your X11 screen but you don't get any response to
the keyboard/trackpoint/trackpad?

B. Rebooting, as if you reset it.

C. Actually powering off (although this might look a lot like situation B.)

[RasmusP]

Some real gremlin is at work on your W530 - have your tried an exorcist?

Yeah, I already talked to the Lenovo guys. . .

Which ram slots are you populating?

Currently the ones under the keyboard, but with the RAMs that used to be in the lower compartment.

Also, and this is a shot in the dark, are you always using the machine
at the same location (work, home, coffeeshop, etc?) when this occurs?

Nope, it has happened in at least five different countries and with three types of converters. I have switched the plug from the base-charger to the outlet, but the charger is the original one.

The typical scenario is: use W530 at home to program or whatever at night. Suspend it too late and get too little sleep. Grab my X200s in the morning and leave. Come home. Turn on W530 immediately, before taking off my coat and shoes. W530 freezes. On fare cases I have seen it freeze while I work, but typically it's after a 'long' (typical more than 12h) suspend .

Just to be clear, is the machine:

A. Freezing, you can see your X11 screen but you don't get any response to
the keyboard/trackpoint/trackpad?

B. Rebooting, as if you reset it.

C. Actually powering off (although this might look a lot like situation B.)

When I turn it on and it freezes it's like A. I can't do magic SysReg reboot either. The mouse will work for a second and then the computer will freeze.

When I do extensive load of the GPU it's like B.

I have not seen C.

Thanks,
Rasmus

[bit_twiddler]

You could have more than one problem, which is making the diagnosis difficult.

I don't remember if we discussed this before, but how do you have your
graphics adapter selection set in the BIOS. Is it:

A. Discrete (Nvidia?) only?
B. Integrated (Intel) only?
C. Optimus (automatically switching) ?

Also, is there an option in the BIOS to have the operating system automatically detect which graphics adapter is running? I just went through this with a T400, for Linux you want to turn that off.

[RasmusP]

You could have more than one problem, which is making the diagnosis difficult.

Indeed. I have assumed this possibility away 'cause it's just too complex...

u

I don't remember if we discussed this before, but how do you have your
graphics adapter selection set in the BIOS. Is it:

A. Discrete (Nvidia?) only?
B. Integrated (Intel) only?
C. Optimus (automatically switching) ?

Also, is there an option in the BIOS to have the operating system automatically detect which graphics adapter is running? I just went through this with a T400, for Linux you want to turn that off.

I think it's Optimus. But I always start programs that need discrete graphics with optirun or primusrun. I was under the impression that if I choose Integrated (Intel) only I would not be able to use optirun. But I honestly did not test this. Is my understanding wrong?

[bit_twiddler]

If you most use your W530 attached to AC, then I would try setting your graphics
in the BIOS to use either the integrated or discrete chipset, rather than optimus.

I know there were some resume bugs related to Bumblebee a few years ago,
perhaps they have not been completely ironed out.

If you mostly use your W530 on battery, then I would just set the BIOS to
use the integrated GPU.

BTW, you're running out of hardware to test, since you've replaced almost
everything. You quite possibly have the most extensively Q/A'd W530
hardware on the planet at this point.

P.S.: Optimus enables you to run GPU-intensive programs using your discrete GPU,
while using your integrated GPU the rest of the time. But, if you are using
your machine connected to AC most of the time, you might as well just set
the BIOS to use the discrete GPU all of the time. You won't lose any performance
and you'll be able to eliminate Bumblebee as a possible source of problems.

[RasmusP]

If you most use your W530 attached to AC, then I would try setting your graphics
in the BIOS to use either the integrated or discrete chipset, rather than optimus.

I know there were some resume bugs related to Bumblebee a few years ago,
perhaps they have not been completely ironed out.

Yeah, that could be, though I think a lot of Linux users use Thinkpads.

If you mostly use your W530 on battery, then I would just set the BIOS to
use the integrated GPU.

I run it on both, actually, but I almost never use the discrete GPU when not plugged in.

BTW, you're running out of hardware to test, since you've replaced almost
everything. You quite possibly have the most extensively Q/A'd W530
hardware on the planet at this point.

Yeah, it's sad. I will get the new battery on Monday... Let's see if that works. I still wonder if it's the CPU, somehow...

P.S.: Optimus enables you to run GPU-intensive programs using your discrete GPU,
while using your integrated GPU the rest of the time. But, if you are using
your machine connected to AC most of the time, you might as well just set
the BIOS to use the discrete GPU all of the time. You won't lose any performance
and you'll be able to eliminate Bumblebee as a possible source of problems.

Yeah, I will have to try it to iron it out. But I guess, I would have to run Intel-only and see if it still has issues. At least with the restart-bug it only happens with the Intel GPU.

[bit_twiddler]

Yeah, it's sad. I will get the new battery on Monday... Let's see if that works. I still wonder if it's the CPU, somehow...

Perhaps you could check with Lenovo to see if the new motherboard came with a new CPU.
It seems like it would require a lot of effort on their part to replace a motherboard without
swapping the CPU, when one considers the necessity of cleaning up the heat sink
compound and re-applying it.

Yeah, I will have to try it to iron it out. But I guess, I would have to run Intel-only and see if it still has issues. At least with the restart-bug it only happens with the Intel GPU.

It's got to be something in either the Intel GPU driver or Bumblebee. It could be that your
keyboard is defective - have you tried plugging in a usb mouse when this happens?

If that doesn't work, then I would advise trying another Linux distribution, such as Kubuntu 14.04,
and running with discrete graphics only for a while. Intel graphics are still not completely
solid, even after all these years. I'd like to say that this is why they have a new
CEO, but it probably has nothing to do with it.

If you want to try that, I can advise you on how to do it so that you can painlessly move between different
Linux distributions.

[RasmusP]

Perhaps you could check with Lenovo to see if the new motherboard came with a new CPU.
It seems like it would require a lot of effort on their part to replace a motherboard without
swapping the CPU, when one considers the necessity of cleaning up the heat sink
compound and re-applying it.

I'm not certain, but I'm pretty sure it did not. Wages are low here, but hardware is probably the same. The guy only came with one box. And they refused to change the CPU until I debugged on Win7. So I [emph]think[/emph] the CPU is the same. Another crazy thing was that they insisted that the iGPU was installed on the mainboard which makes no sense to me...

It's got to be something in either the Intel GPU driver or Bumblebee. It could be that your
keyboard is defective - have you tried plugging in a usb mouse when this happens?

I sometimes have an external mouse connected. I don't recall clearly if it has been connected when the problem appeared, but I guess so. Still, you could be right. But the mouse works for a couple of seconds before the system freezes...

With Nvidia Optimus and 16GB RAM installed, I can still crash my computer with Furmark. . . . I really need to create a USB-Win7 and test Furmark on that system....

However, I have not had the resume-bug while having Optimus turned off. So it may just be two different problems... Darn.

[bit_twiddler]

Glad to see that you haven't given up.

You're right, the iGPU is on the CPU.

Bumblebee/Optimus on Linux sounds buggy to me although I haven't actually
run it. I did sit next to someone on a plane ride a few years ago who
was playing with it. I'm not sure if he was just a user or a contributor,
but he did seem to know a lot about the Debian kernel.

He mentioned that the last remaining problems with Bumblebee were
related to resuming from a suspend.

Does the crash on Furmark still happen if you turn Optimus off in
the BIOS and just set it to use the discrete graphics?

Also, you haven't mentioned if you are using Nvidia proprietary
graphics or Noveau, although, I dimly seem to recall that Bumblebee
doesn't work with the proprietary drivers. (It's been a while since I
looked at it.)

Bringing up W7 sounds like a good idea if you are comfortable with it.
It's a great operating system, and will help you to isolate Linux-related
issues.

But, I would advise removing all Linux boot drives from your system while
you do the install, because W7 can clobber your MBR during installation
or during an update. I actually install and boot off operating systems on a drive
in the ultrabay so that they can't clobber each other's MBR.

[RasmusP]

Glad to see that you haven't given up.

Just busy with other deadlines.

You're right, the iGPU is on the CPU.

What I thought. But the Lenovo phone engineer was of a different opinion.

Bumblebee/Optimus on Linux sounds buggy to me although I haven't actually
run it. I did sit next to someone on a plane ride a few years ago who
was playing with it. I'm not sure if he was just a user or a contributor,
but he did seem to know a lot about the Debian kernel.

Bumblebee works very well. Basically, I prefix graphically intensive jobs with primusrun or optirun and it will be executed using the Nvidia card. All other tasks are carried out via the iGPU. So, do test it I'd do

Code: Select all

primusrun glxspheres64 
# or 
primusrun start_furmark_windowed_1024x640.sh 

He mentioned that the last remaining problems with Bumblebee were
related to resuming from a suspend.

Possibly. There's reported problems when using the vgaswitcheroo. I've installed a fix for such a problem. I don't know if bumblebee uses this. Or if it has it's own set of problems. However, the discrete card should be turned off except when I start a command with primusrun or optirun.

Does the crash on Furmark still happen if you turn Optimus off in
the BIOS and just set it to use the discrete graphics?

The crash only happens with the iGPU/Intel graphics. Discrete/Nvidia is rock-solid. But I haven't tested if it crashes when using Nvidia.

Also, you haven't mentioned if you are using Nvidia proprietary
graphics or Noveau, although, I dimly seem to recall that Bumblebee
doesn't work with the proprietary drivers. (It's been a while since I
looked at it.)

I use Nvidia drivers, as they offer better performance ATM. AFAIK, only Nvidia drivers work with bumblebee.

Bringing up W7 sounds like a good idea if you are comfortable with it.
It's a great operating system, and will help you to isolate Linux-related
issues.

<rant>
I've had to use it and I find it very awkward. Yesterday, I had to create a ssh key with putty. The key was put into some secret folder, which I could only see from the open dialog in Putty, nowhere else. I could not even "cd" into it.
It seems like it a system that is not meant for productivity---at least with the kind of tools I use.
</rant>

But, I would advise removing all Linux boot drives from your system while
you do the install, because W7 can clobber your MBR during installation
or during an update. I actually install and boot off operating systems on a drive
in the ultrabay so that they can't clobber each other's MBR.

Oh, I won't let it touch my main drive. I'll have to either install it on a usb or a separate drive.

I will report back when I find a spare drive.

[bit_twiddler]

Oh, I won't let it touch my main drive. I'll have to either install it on a usb or a separate drive.

Not sure if I was clear here - you have to physically remove any
Linux drive that you would normally boot from - Windows
will go out of its way to clobber your MBR during
the installation or when it updates.

Just installing on a separate drive is not enough.

[RasmusP]

That was perfectly clear. I was not going to risk my working environment in any case.

I found a decent deal on the Crucial MX100 and installed Windows 7 on it tonight. And...: no crashes! I tried with both 16Gb and 32Gb installed. To be fair, when I swapped my main disk back, it also didn't crash in the Furmark test. This was with optimus disabled. Now that I have the W7 disk I can swap if/next time I have problems and see if they also manifest themselves there. I'm trying some other boot flags for the Linux Intel drivers now.

Also: I got like 2-3 more FPS with the Intel drivers on W7 than the Intel drivers on Linux.

[bit_twiddler]

That's some good news after all this effort on your part.

To hazard a guess, it's sounding like the problem is either in i915 drm or Bumblebee at this point.

2 questions if/when you turn Optimus/Bumblebee on again in Linux, assuming
that your problems do not recur with them turned off:

Do you have RC6 enabled in i915?

I'm not sure if you can do this on Arch Linux and whatever display manager
you are using, but do you have a way to switch virtual consoles when
X is running? In your resume from suspend bug, your keyboard may not actually be frozen,
just the display. If so, you could poke around a bit and take a look at the
output of dmesg and the X server log.

[RasmusP]

2 questions if/when you turn Optimus/Bumblebee on again in Linux, assuming
that your problems do not recur with them turned off:

In a longer post while sleeping the computer crashed from Linux...

Do you have RC6 enabled in i915?

at the moment I'm using sna acceleration, tear free and I have enabled i915.semaphores. I guess RC6 is on by default these days. I'll disable it and see if it makes a difference.

In your resume from suspend bug, your keyboard may not actually be frozen,
just the display. If so, you could poke around a bit and take a look at the
output of dmesg and the X server log.

I don't know the status of this bug. But I can't switch to the console when it freezes. Unfortunately.

Edit: Just verified that Furmark crashes the PC from Fedora 21 live usb (Gnome).
Edit2: So far the LXDE spin of Fedora 21 has been running for half an hour without a crash.
Edit3: On my regular setup, it's also pretty stable when running Furmark from i3 as opposed to Gnome. Though it crashed almost instantly when I ran Furmark from i3 after reenabling Optimus in the bios.
Edit4: I spoke too soon. It just took longer for the system to crash under i3 than under Gnome. I'll run a longer test on windows and see if it's completely stable there...

[bit_twiddler]

Just so I understand, have you had any crashes running Furmark
on Linux with Optimus disabled in the bios?

[RasmusP]

I believe I have had crashes with Optmus disabeled. I reenabled Optimus in the morning, so the test I ran during the night crashed the computer with only Intel graphics enabled.

W7 has been running Furmark for a couple of hours without issues. The temperature is similar (shy of 90C) under both OSes.

[bit_twiddler]

It sounds like you're getting close to the point where you
should send a core dump to somebody, but the question
is "to whom?"

If you got a crash with just the Intel graphics (i.e., discrete
and Optimus disabled in BIOS), then there is probably
a bug in the i915 driver that Intel should be interested
in looking at.

Since the graphics drivers have access to kernel space,
you might want to read up on how to configure kdump
with Arch.

Can you verify that you don't get a crash with the same
test case (integrated graphics only, Furmark, overnight)
with W7?

[RasmusP]

I will let W7 run overnight.

Since I see it with both Fedora 21+Gnome, Arch+i3 and Arch+Gnome (without optimus) I guess it's looking like a bug.

From the kdump Arch wiki I gather that it's a toolt for dumping an 'image' of the memory when Linux crashes. Is that an accurate ELI5 idea?

Is there some live distro that is (i) recent; and (ii) already have the tools for generated the appropriate backtraces and data?

Intel graphics bug reports request a scary amount of stuff (for someone who's not familiar with the Kernel other than as a user). E.g. how do I capture dmesg when the error results in the computer crashing? I wonder if I could tell systemd to record it continuously...