241775 Views Previous 1 2 3 4 5 6 … Next 209 Replies Latest reply: Oct 4, 2010 12:53 PM by JOMStar Go to original post
Unfortunately, there's a bunch of important-looking functions in the kernel that dtrace won't probe, (and probably for good reason), but there's a kernel function that can be overridden by a driver called 'pmSafeMode' that apparently puts the processor into a driver-defined "safe mode" if the CPU gets too hot. I think that that's what the power management driver is doing -- when the CPU gets too hot, the kernel notifies the driver that it should put the processor(s) into safe mode, which probably consists of something like running the processors in an idle loop.
On the other hand, it looks like CoolBook is somehow completely avoiding this safe mode. Using CoolBook with the default settings, without undervolting, I was unable to get kernel_task to crowd the processors like it does when CoolBook isn't enabled.
A safe mode to cool down the machine seems like a good idea, but not when it gets hit during a 10-minute video on youtube or hulu. I'll probably take my machine in to get it checked out next time I swing by an Apple store, since the last time I was in one the display Macbook Airs seemed to be running some pretty CPU-intensive tasks more or less flawlessly.
What is unbelievable is that even very low CPU tasks makes the kernel_task go crazy. Case in point: Microsoft Remote Desktop. So, because the Air cannot handle any CPU intensive tasks, I bought a fast PC to handle that, accessing it through remote desktop. However, the air cannot even handle remote desktop without overheating! So the air is not even useful as a terminal! That's absurd. Absurd. The air cannot handle CPU intensive task, and it cannot handle accessing a fast computer to do those task. I'm stuck.
My MacBook Air becomes unusable after 10 minutes of playing video due to this kernel_task process consuming all my CPU. I'm extremely disappointed for this situation, this computer was supposed to be a "marvel of engineering" but it became a piece of junk with a beautiful aluminium case that fits in a manila envelope. Mac OS X 10.5.5 did not fix the problem. This is a USD$ 3K notebook, I have owned several Apple notebooks/desktops before and this is by far the most disappointing and annoying computer, I've started to feel the same frustration that a Windows user feels.
Does anybody think this problem can be fixed in software? I think Apple should consider recalling this piece of junk and give us a real notebook. Is anybody at Apple reading these forums?
Hi I'd to add my 2c.
I've had the thermal issues discussed in this thread from the start. At first the symptom was the oft-mentioned "core shutdown" plus the kernel_task showing excessive CPU usage. The result made the machine unusable until it had cooled down.
After a SW update from Apple (http://support.apple.com/kb/HT2883), the second core no longer shuts down under thermal load. But the fundamental physical thermal issues are not solved. So now, under thermal load, both cores continue processing but the kernel_task thread still ties up the CPU like before. Same result: machine grinds to a halt. The patch seems to have been motivated by a corporate mentality "So, people are complaining that a core shuts down? OK, we'll just stop it from shutting down". It does not address the reason WHY the core shuts down in the first place (ie, thermal overload).
Thermal overload is caused by CPU-intensive or GPU-intensive tasks. You also need to take into account the ambient temperatures the machine operates in (I've measured my ambient room temps in summer and they can often exceed 30C. The MBA is hardly usable at all in that environment). I'm also assuming the machine is operated in such a way that the cooling vents are completely unhindered. Here are some of the things I do that quickly induce thermal overload:
o voice or video Skype calls (CPU intensive)
o Organizing large photo libraries in iPhoto (CPU load)
o Manipulating photos in PhotoShop (CPU load)
o Playing DVDs with DVD Player (CPU & GPU intensive)
o Using Google Maps or Google Earth (GPU load)
o Running machines in VMWare Fusion (CPU load)
o Watching clips on YouTube (CPU & GPU intensive)
o Hooking up an external 1920x1200 monitor (GPU load)
o Using Entourage on all my email accounts (CPU intensive)
Often I am doing several of these at the same time (or attempt to). All of them are CPU &/or GPU intensive tasks. However, most are absolutely commonplace tasks on contemporary computers. I run the exact same tasks on a variety of other HW and OS's without trouble. I observe system temperatures rise as expected in response to the CPU & GPU load, however, adequate cooling is provided which effectively extracts the thermal waste and thus maintains stable machine operation. Fan noise may become quite loud under load but I can deal with that. Having a machine grind to a halt when used under load is a show stopper.
Some of my friends have criticized my remarks, claiming I'm some kind of "power user" who is expecting "server performance" from a light laptop. My response is pretty simple. I'm not expecting server performance. Server performance is like a dual-quad core machine with 32 GB memory and fast disk IO subsystems. I have those machines too. I'm not expecting that from my MacBook Air. What I am expecting is simply what I've got: a 1.6GHz Core 2 Duo CPU and 2GB RAM. The problem with the MBA is as soon as you start using your CPU cores at 100% (or tax the GPU) the OS grinds to a halt. You are not getting the full performance of that CPU. There is nothing "wrong" with running your CPU at 100% full time. Apple just seem to have created a machine that can't handle it, even for short periods of time if ambient temps are high enough.
I personally find the situation extremely disappointing for a laptop which could have otherwise achieved a perfect score. If I get the opportunity, I will certainly try to return it or get it serviced. But from the continued activity on this thread it appears a real fix from Apple is still forthcoming.
Do you think this problem can be fixed in software? If not what would be our options? I think this is a problem that is affecting every single MacBook Air.
I'm extremely frustrated with this computer, like you said we can't do any CPU semi-intensive work without the machine slowing down to the point of becoming unusable. I remembered when Steve Jobs release this notebook he said they didn't want to sacrifice performance like other UMPC did, so Intel design a custom chip for the MacBook Air so we can have almost the same performance of a MacBook or MacBook Pro, and everybody was WOWed by the size of the CPU, but that hard work made by Intel became worthless when you can't use the full performance of the CPU.
I don't think the problems can be fixed via a software patch at all. The only thing a software patch can do is attempt to prevent heat build up (by throttling the cpu, again, robbing you of performance) or increase the capacity to dump heat waste (running the fans faster). We've seen that even with the fans running at 6200rpm it's still not sufficient to prevent thermal overload.
I've been personally building and assembling x86 computers from scratch for the last 10 years. Generally I tend to build high-performance machines and heat dissipation is a major issue that I have learned (the hard way) must be taken very seriously if you want the machine to be stable under continuous full load. In my quest to make my machines as silent as possible (eg via water cooling or even fanless techniques) I have run into countless issues with heat dissipation. It's a complex design topic taking into account expected ambient operational temps, adequate heat transfer methods for heat-emitting components and even case airflow characteristics.
Based on my personal experience building machines, my theories about the MBA problem are:
(a) hardware assembly defects concerning the thermal measurement and transfer components (heatsinks, fans, thermal grease, temperature diodes, etc) which seriously impair the theoretical maximum cooling ability of the machine
(b) ventilation flow limits imposed by that oh-so-beautiful case make it impossible to dissipate all the heat the CPU & GPU can generate under full load, even with fans running at full speed
(c) problems in the cooling feedback system (ie temperature sensors and/or fan controllers and related software in BIOS & kernel)
If it's (a), great, returning the machine to Apple should be able to solve it if you are one of the unlucky ones with that defect.
If it's (b), I'd personally take the machiine back as it' simply not suitable for what I (and I guess most people) do
If it's (c), there may be a chance that a software patch can fix things, unless it concerns the hardware temp sensors. Then you're back to (a).
You have stated things very well. This is my first Mac for work, and I am somewhat disappointed with Apple due to this problem. Fortunately, the MBA works for 95% of what I need to do for work, but it's annoying that I loose the last 5% due to the heating problem (and that fans run constantly when the ambient temperature is high).
I just want to keep you guys updated on this issue. I went today to the Apple Store Genius bar for another MacBook air issue and I also complained about the overheating issue and Kernel_task, and he told me 3 times that Apple is aware of this issue and they will soon come out with a solution for this. He couldn't give me a date but he told me that Apple is aware of it. Whatever that means.
Marc Van Olmen
Message was edited by: Marc Van Olmen1
Have to restate my previous statement. Right now I'm working with some big MS word documents, with an important deadline Friday. The MBA cannot handle it. The kernel_task shoots up to 150%. Yes, these are 100 page documents with lots of graphics, but I never had a computer that could not handle word documents, all the way back from my SE/30. Unacceptable.
Now I have used the fan "solution" for some time, and it appears to work. I have had no kernel_task going crazy events. I can do ordinary office work, and I even tried to tax the CPU very hard, and never any problems. It just appears that the temperature in my home office was too high. This makes me confident that if I go and do a presentation somewhere, I will not have problems, as the room surely will be air-conditioned.
This can also explain why some people never experience kernel_task going crazy (or previous core shutdown). If the climate is temperate, this problem may never occur.