Core Shutdown - it isn't temperature related

[yoavcs]

OK, so I've started seeing the dreaded core shutdown issue on my 1.6 Air.

And I don't think it is temperature related. This is a bug in the GMA X3100 graphics driver!

Happens when I play WoW, on even low graphics settings. Game starts out beautifully with high frame rates but after 5 or so minutes the system drops to an unplayable crawl.

When this happens Activity Monitor shows that one core is off, and the other is pegged at 100% activity by the kernel_task process (PID 0), which is taking up 50-60% of the CPU.

I am now convinced this is a bug in the GMA X3100 driver and **not** temperature related, for a few reasons:

I tried placing the Air on a cold A/C vent when this happens, dropping the temperature to real low levels and it still didn't help.

On the other hand, quitting WoW immediately brought the other core back to life and kernel_task returned to its usual 1% of CPU use. If this had been temperature related why would everything return to normal *seconds* after quitting the offending program?

Also, I tried opening a few Terminals and running "yes > /dev/null", which is a good way to exercise the CPU. Both cores shot up to 100% usage and the temperature sky-rocketed beyond anything I had seen before, but no core shutdown and no system slow-downs.

So... the fact this happens in WoW and for many others while playing longer Flash Player videos, while others never see it (presumably never tasking teh graphics part of their system) points to it being graphics related... either a hardware bug, or more probably a driver bug.

Since drivers run in kernel space, I believe there is a bug in the GMA X3100 driver which is causing this, as evidenced by the kernel space process (kernel_task) eating up all the CPU time of the single functioning core while this goes on.

Anyone else see similar things?

 

[Thetics]

interesting. maybe more people can try to see if they can replicate this. ie: watching long youtube videos and such while keeping the core temp low to see if it still causes a problem?

 

[ayeying]

I had 2 core shut downs so far, and I've noticed if I were to block the vents then watch something on youtube, the core will shut down. The temps are still around 65 deg C with fans at max but the core still shuts down. If I place the system anywhere else, where the vents aren't blocked, the core doesn't shut down.

I personally don't think the core shut down is related to the cpu temp since I have pushed this system well above 100 deg C and no core shutdowns.

There might be some truth with buggy X3100 drivers.

I play second life and I'll see if I can shut the core down while having the system on a AC vent... guess I'll have to use my car's AC but it should work. lol

 

[ayeying]

Alright. After playing 15 minutes of Second Life, I experienced NO core shut downs. CPU temp rose to 80 deg w/ fans at max but no core shut downs. The system is placed on a flat kitchen table.

 

[ericctc]

I tried to watch Diggnation from Revision3 (Flash player) on my MBA 1.6 with 80GB. I got one core shut down 12 minutes into the video. I rebooted MBA system into Windows XP (BootCamp). I watched entire video (59 minutes) without any core shutdown. My CPU temperature was between 73 to 74 degrees in the XP. My CPU load was between 55% to 65% during the video. This is getting interesting.

 

[ayeying]

Okay, Im currently running the following programs:

iChat w/ 2 Conversations
iCal (Running in the background)
Preview (Running in the background)
iTunes, playing 1 movie, MP4 file.
Quicksilver
Mozilla Thunderbird and Firefox (9 tabs opened, none of the tabs have flash running)
VMWare Fusion running Ubuntu 8.04 @ 256MB
Transmission w/ 4 torrents
Activity Monitor

Temperature is:

61-71 deg C for CPU
55-60 deg C for Heatsink
36 deg C Enclosure Base 1
36 deg C Enclosure Base 2
28 deg C Enclosure Base 3
45-47 deg C for Airport Card
60-63 deg C for Power supply 1
33 deg C for HDD

Fan speed of 6200rpm stable.

My set up is MacBook Air with internal screen at 1280x800 and external screen connected via DVI at 1680x1050. Movie is playing on the external, VMWare is on the internal screen.

I'm getting intermittent(?) core shut downs. I don't know why, the temps are low and stable, fans are at max. I noticed this started happening after installing 10.5.3... but I can't be entirely sure. Currently, Im inclined to agree with bad X3100 drivers thats causing this. I am playing a video in the background so its possible thats whats causing the shut downs. I can run 2x yes > /dev/null/ without any core shutdown, but im running a simple video with nothing but an idle VM and it shuts down?

 

[yoavcs]

@ayeing:

Can you get Second Life to core shutdown in anyway? Has it done so in the past? It could be Second Life isn't graphics-intensive enough to cause the bug to appear.


@ericctc:

As for the no core shutdown in XP despite high temps: I think this lends credence to my graphics driver theory. If it was hardware, I think we'd see shutdowns in XP too, especially at high temps.

Something is causing OS X to shutdown a core way too early, and is using up all the CPU cycles of the available core during these episodes. I think that something is a buggy X3100 driver

 

[ayeying]

@ayeing:

Can you get Second Life to core shutdown in anyway? Has it done so in the past? It could be Second Life isn't graphics-intensive enough to cause the bug to appear.

Nope, but I am able to get a core shutdown by watching videos. Its awkward. I won't get any shutdowns playing Call of Duty 4 but I get core shutdown playing an iTunes movie.

 

[yoavcs]

What if you watch a video of someone playing Call of Duty 4?


I still see this as graphics driver related: some things are triggering the bug, some aren't. It seems Flash and WoW can both get the bug to trigger.

The fact it doesn't happen in XP shows it isn't hardware, rather software-related.

And the fact it always happens with graphics usage points to it being graphics driver related.

 

[dudup]

My only rant with the MBA is its graphics performance.

I've already posted it before on other threads: its performance is way inferior when compared to a regular '07 MacBook when playing video, specially HD content

Do a test: play a full HD video on ANY MacBook and then on a MBA, and you'll see what I'm taking about.

It's not supposed to be this way, specially with a brand new X3100 graphics, which is much faster than a GMA950.

 

[nintendude]

in sims 2 under boot camp for xp, i core shut downs really quick, about one minute w/ all low settings

 

[ayeying]

My only rant with the MBA is its graphics performance.

I've already posted it before on other threads: its performance is way inferior when compared to a regular '07 MacBook when playing video, specially HD content

Do a test: play a full HD video on ANY MacBook and then on a MBA, and you'll see what I'm taking about.

It's not supposed to be this way, specially with a brand new X3100 graphics, which is much faster than a GMA950.

HD content, on integrated graphics, don't use integrated graphics, instead, they use the CPU. Remember, our integrated graphics piggyback on the CPU for rendering and such. Comparing a 1.6GHz vs a 2.0/2.16/2.4GHz isn't a good and even fair comparison

 

[dudup]

HD content, on integrated graphics, don't use integrated graphics, instead, they use the CPU. Remember, our integrated graphics piggyback on the CPU for rendering and such. Comparing a 1.6GHz vs a 2.0/2.16/2.4GHz isn't a good and even fair comparison

C'mon, even Core Duo 1.83 ghz play HD content fine without sweat.

Don't get on the cpu-envy bandwagon. way less than 2 years ago a santa rosa core 2 duo of 1.6ghz would be a hell of a cpu.

I don't have any windows to test right now, but I will do so, this got me interested.

 

[ayeying]

There is evidence where when the system is heated up, the core actually throttles back if the core does not shut down. Maybe thats whats happening with your system under extremely heavy load such as HD content. I've played HD content only upto 720i and I have yet to get any studdering on my system. The only downside is the system gets extremely hot but beyond that, it runs fine.

 

[exabytes18]

Could core shut downs be dependent upon the CPU usage? Say if the CPU usage is under 50%, OSX will shut down one core? That would be a bogus rule to go by since the OS cannot predict spikes in CPU activity leaving apps to make do with only one core. I'm just throwing ideas out there... official word from Apple would be great though.

 

[yoavcs]

in sims 2 under boot camp for xp, i core shut downs really quick, about one minute w/ all low settings

Hmm, this seems to throw a wrench in my theory. If XP in Boot Camp is seeing core shutdowns, it can't be just the OS X driver... Can more people verify seeing shutdowns in XP? I seem to recall people stating that XP worked flawlessly under Boot Camp, which I saw as strengthening my theory.

As for the discussion on graphics performance:
A 1.6 GHz Merom + GMA X3100 should be more than enough for HD content, or WoW at max settings (albeit low frame-rate).

Other laptops, with better cooling, on a 2 GHz Merom *breeze* through it, and I doubt that the 1.6 -> 2.0 GHz difference would make such a *drastic* difference.

There is either a software bug in OS X on the Air, or the cooling on this system is so bad it keeps it from running at what I would call even 50% full potential.

 

[dudup]

You know, it seems to me that they are always forgetting about the video on new projects.

Remember the graphics freezing on the aluminum iMac?

I will try to setup a Bootcamp Windows tomorrow and do some tests. Unfortunately, I don't have a superdrive, will have to use the machine from work.

 

[ayeying]

Hmm, this seems to throw a wrench in my theory. If XP in Boot Camp is seeing core shutdowns, it can't be just the OS X driver... Can more people verify seeing shutdowns in XP? I seem to recall people stating that XP worked flawlessly under Boot Camp, which I saw as strengthening my theory.

As for the discussion on graphics performance:
A 1.6 GHz Merom + GMA X3100 should be more than enough for HD content, or WoW at max settings (albeit low frame-rate).

Other laptops, with better cooling, on a 2 GHz Merom *breeze* through it, and I doubt that the 1.6 -> 2.0 GHz difference would make such a *drastic* difference.

There is either a software bug in OS X on the Air, or the cooling on this system is so bad it keeps it from running at what I would call even 50% full potential.

I can't say if the core actually shut down but when the temp rose to a certain limit the cores both hit 100% in the process and the graphics slowed to a crawl in Sins of A Solar Empire. When I moved the system to a cooler location, the lag went away.

Here's an infamous pic of a core shutdown with running a virtual machine and other stuff.

 

[yoavcs]

I can't say if the core actually shut down but when the temp rose to a certain limit the cores both hit 100% in the process and the graphics slowed to a crawl in Sins of A Solar Empire. When I moved the system to a cooler location, the lag went away.

If both cores were pegged at 100% that isn't a core shutdown. That's just the system being stressed to max, possibly by Windows lowering the CPU speed to help alleviate heat build-up. You end up with a dual-core 600 MHz CPU (as an example) which isn't enough for what you are doing and therefore the system crawls.

Core shutdowns are what you depicted in the image you attached: one core completely turning off, 0% activity, leaving the other to fend for itself.

What I am seeing is one core shutting off, and the other being pegged at 100% by a supposedly errant kernel process (the graphics driver) - therefore leading to a system slowdown where none is warranted. This isn't simple heat build-up, causing an underclock. This is a bug.

 

[ayeying]

If both cores were pegged at 100% that isn't a core shutdown. That's just the system being stressed to max, possibly by Windows lowering the CPU speed to help alleviate heat build-up. You end up with a dual-core 600 MHz CPU (as an example) which isn't enough for what you are doing and therefore the system crawls.

Core shutdowns are what you depicted in the image you attached: one core completely turning off, 0% activity, leaving the other to fend for itself.

What I am seeing is one core shutting off, and the other being pegged at 100% by a supposedly errant kernel process (the graphics driver) - therefore leading to a system slowdown where none is warranted. This isn't simple heat build-up, causing an underclock. This is a bug.

But im not doing anything graphic intensive... except watching a nonhd itunes movie.

 

[nintendude]

the macbook also uses the same GMA X3100 so maybe if someone w/ a macbook tested this it may help.....

 

[yoavcs]

But im not doing anything graphic intensive... except watching a nonhd itunes movie.

Doesn't have to be graphics intensive, just activate the buggy code-path in the X3100 driver, if I am correct. Seems Flash and WoW are good at that, other graphics software not so much.

The suggestion re: testing on a plain MacBook is a very good one. Can any MacBook owner help out here?

 

[Phil A.]

I found a way to kill a core on my 1.8 MBA that definitely seems related to video and it's when I'm running the latest Beta of VMWare Fusion:
1) MBA connected to an external monitor
2) Setup Fusion to run on one of the spaces
3) Set Fusion to Full Screen
4) Within Fusion setup the screen to expand onto second monitor

Within a minute the second core will start shutting down. As soon as I disable full screen on fusion it comes back to life, and I can replicate this all day. It does it if the Fusion space is the active one or not. Interestingly, this only seems to have started since the update to 10.5.3, so it could be a new issue introduced with that update

This is a strange combination of OSX Drivers and the VMWare XP driver, but together they certainly don't play nicely!

 

[nintendude]

I found a way to kill a core on my 1.8 MBA that definitely seems related to video and it's when I'm running the latest Beta of VMWare Fusion:
1) MBA connected to an external monitor
2) Setup Fusion to run on one of the spaces
3) Set Fusion to Full Screen
4) Within Fusion setup the screen to expand onto second monitor

Within a minute the second core will start shutting down. As soon as I disable full screen on fusion it comes back to life, and I can replicate this all day. It does it if the Fusion space is the active one or not. Interestingly, this only seems to have started since the update to 10.5.3, so it could be a new issue introduced with that update

This is a strange combination of OSX Drivers and the VMWare XP driver, but together they certainly don't play nicely!

so this is more evidence that the problem is graphics related?

 

[nintendude]

sorry double post