Audio Decoding KILLS MacPro (2009) green factor

[cl-user]

...
I have been given the idea of sending a recorded letter to Apple HQ ...

I imagine that Apple read this and their own discussions forums where this issue has been loud and clear for a while.

AppleCare and Apple Geniuses try to handle what they are trained to handle (which does not include issues like this).

QED Apple know about this and chose to be silent.

The question: why the silence, is the problem beyond their technical ability (ability multiplied by motivation)?

Is the solution some hardware mod which would require the recall of every 2009 Mac Pro dual.

Are Apple accountants hoping we'll just shutup and the upcoming problems be seen as aging and attributable to user abuse.

For those thinking to jump ship to the upcoming 6 core systems, what reason is there to expect Apple to have sorted out these technical issues, or to have changed their attitude towards the user community when the next problem crops up.

All in Apple hasn't smelled so weak for some time. Perhaps it is just a matter of Mac Pros being sidelined by bigger more profitable ventures like iPhone, tablet, kids pretty colored MP3 players, ... The company evolves and desktop systems are a shrinking proportion of their interests, and presumably the developers of our systems have moved on to new projects.

Is the issue one that falls between hardware, OS and applications developer disciplines, hence it is everyone else's problem.

I suspect that we users could be on our own with this bug, and find the only Apple response on offer (if and when they ever get around to it) is platitudes and evasion.

This issue is blatantly obvious. Silence from Apple seems to infer that they have a problem with this. The length of time that they have been silent argues for their problem being a big one.

So perhaps the sane action for users is to begin looking into laying the foundations for our class action (presumably it may come to this eventually), in parallel with trying to find causes and cures, and trying to get through to Apple in the short term.

Some things seem certain, Apple is very unlikely to change their attitude. And their heads must be in the sand with this bug.

 

[cl-user]

If I was Apple, and if there was some reason why I was unable to address this issue (eg by issuing information and software patches), and if I was unconcerned with loss of brand image in this niche (Mac Pro business).

Then I think I'd plan on trying to isolate users, stall, and obfuscate. But that's not my style, I'm sure it would catch up with me eventually (it ought to).

Like asbestos insurance claims, there's one way out of a hopeless situation: the claimants will die off eventually or maybe give up hitting a punchbag.

I feel like Br'er Rabbit and the tar baby http://en.wikipedia.org/wiki/Tar_baby

... but do please, Brer Apple, don't fling me in dat brier-patch, don't swap my 09MP deep fryer for an efficient and debugged Gulftowm replacement.

 

[smacman]

Nice insights... If you have access to Windows still, can you run a stress test on Windows to see what the core temps hit there?

Ok I did it. If this thing cooks itself to death, it's Apples fault. It should be thermally stable out of the box.

Results:

Temps are exactly the same as what I am seeing in OSX under load and at idle. Audio playback had no effect whatsoever on temps and unlike OSX, the system remained stable throughout. The highest reported temp was 99C on Core1, but that was for a split second. For most of the 10 minute stress test, temps stayed in the 88-94C range. Now if I was in OSX and I started to play audio at this point, I would be greeted by a kernel panic within minutes.

Conclusion:

So long story short, if you want a thermally stable 09 Mac Pro, at the moment, the only true solution is to ditch OSX and run Win7. Alternative is to use OSX and never play audio or use firewire while doing cpu intensive work.

 

[Inconsequential]

@smacman, couldn't get me a jump seat flight could you?

My Mac Pro *hasn't* crashed, and i've had audio + handbrake + safari/mail/MSN/Photoshop/Lightroom/VMware and god knows what else running at the same time, maybe I'm just lucky?

@cl-user

The recorded letter suggests I would proceed with legal action if the problem is not resolved. I think, that, should kick apple into doing something.

My other idea would be to pack the MP up, take it to an Applestore and demand a upgrade to the new 2010MP if they don't have the issue.

If either 10.6.3/2010MP fixes the issue then i'll try for that, otherwise I think we need to do something more drastic.

 

[nanofrog]

cl-user: It's not in the hardware, but in OS X. That's been proven for some time, and smacman has just provided the latest confirmation.

 

[VirtualRain]

Ok I did it. If this thing cooks itself to death, it's Apples fault. It should be thermally stable out of the box.

Results:

Temps are exactly the same as what I am seeing in OSX under load and at idle. Audio playback had no effect whatsoever on temps and unlike OSX, the system remained stable throughout. The highest reported temp was 99C on Core1, but that was for a split second. For most of the 10 minute stress test, temps stayed in the 88-94C range. Now if I was in OSX and I started to play audio at this point, I would be greeted by a kernel panic within minutes.

Conclusion:

So long story short, if you want a thermally stable 09 Mac Pro, at the moment, the only true solution is to ditch OSX and run Win7. Alternative is to use OSX and never play audio or use firewire while doing cpu intensive work.

Thanks! It's great to know that the temps being reported in OSX are accurate... which concerns me even more... the core temps are really out of control on these Mac Pro's whether playing audio or doing other demanding tasks... it's frightening how inadequate the cooling in these rigs are.

In my stress testing in OSX, the fans were not ramping up until core temps hit 95-deg C!!! And then they only increased about 100RPM! The cooling in these is a joke!

 

[daniep01]

Merry Christmas!

Un-boxed my brand new 8-core this morning, standard configuration.

After playing MP3 music in iTunes the CPU A temp rose quickly from 26 to 49, but then stopped and has been hovering between 46 and 49 for about 45 minutes.

I agree it's a bit strange, given CPU usage is 0, but I'm not overly concerned at the moment.

Temp measured in iStat Nano dashboard widget.

Hope this is useful information.

 

[Inconsequential]

Un-boxed my brand new 8-core this morning, standard configuration.

After playing MP3 music in iTunes the CPU A temp rose quickly from 26 to 49, but then stopped and has been hovering between 46 and 49 for about 45 minutes.

I agree it's a bit strange, given CPU usage is 0, but I'm not overly concerned at the moment.

Temp measured in iStat Nano dashboard widget.

Hope this is useful information.

Can you download XBench (or run something CPU intensive like 1080p encoding?) and see what performance drop is like? Im thinking either 12.5% or 25%.

Because Im thinking its using one core to decode audio...

 

[VirtualRain]

Can you download XBench (or run something CPU intensive like 1080p encoding?) and see what performance drop is like? Im thinking either 12.5% or 25%.

Because Im thinking its using one core to decode audio...

In my testing, the temp increases and AC power draw increase from idle when decoding audio were 60% of what a four core stress test would equal, leading me to believe that this problem is equivalent to fully loading 2 cores and then some!

 

[nanofrog]

In my stress testing in OSX, the fans were not ramping up until core temps hit 95-deg C!!! And then they only increased about 100RPM! The cooling in these is a joke!

A firmware update to adjust the SMC to an accurate profile (meets Intel's specs or better) would work here though, but it does mean Apple has to address it to begin with.

That's not the same as truly addressing the actual issue though, as OS X needs fixed. But it may help in the mean time. Ideally, both need to happen, as both seem to be borked.

In my testing, the temp increases and AC power draw increase from idle when decoding audio were 60% of what a four core stress test would equal, leading me to believe that this problem is equivalent to fully loading 2 cores and then some!

4x cores at 100% /= TDP though. There's the controller, IMC, and QPI as well, and if one of those sections are running flat out, it will burn power, yet not show up in the CPU % utilization reported by software apps like iSTAT.

 

[Inconsequential]

In my testing, the temp increases and AC power draw increase from idle when decoding audio were 60% of what a four core stress test would equal, leading me to believe that this problem is equivalent to fully loading 2 cores and then some!

True.

Seeing as we have turbo boost, maybe thats kicking in sending one core (in my instance) to 2.93Ghz, which could very well be 60% of a 2.8ghz 4 core load?

We are speculating tho. *shrug* Engineering does know about this issue now so now we wait for 10.6.3!

 

[smacman]

Un-boxed my brand new 8-core this morning, standard configuration.

After playing MP3 music in iTunes the CPU A temp rose quickly from 26 to 49, but then stopped and has been hovering between 46 and 49 for about 45 minutes.

I agree it's a bit strange, given CPU usage is 0, but I'm not overly concerned at the moment.

Temp measured in iStat Nano dashboard widget.

Hope this is useful information.

You should be concerned. That iStat measured CPU temp is from the heatsink sensor. Your CPU is hotter, and the cores even hotter yet.

I know it's a new toy, but when the euphoria wears off, I suspect you may be annoyed that you have a fundementally flawed machine.

 

[Max(IT)]

You should be concerned. That iStat measured CPU temp is from the heatsink sensor. Your CPU is hotter, and the cores even hotter yet.

I know it's a new toy, but when the euphoria wears off, I suspect you may be annoyed that you have a fundementally flawed machine.

45' C is not something to worry about, flawed or not

 

[Max(IT)]

Thanks! It's great to know that the temps being reported in OSX are accurate... which concerns me even more... the core temps are really out of control on these Mac Pro's whether playing audio or doing other demanding tasks... it's frightening how inadequate the cooling in these rigs are.

In my stress testing in OSX, the fans were not ramping up until core temps hit 95-deg C!!! And then they only increased about 100RPM! The cooling in these is a joke!

This seem to be a problem with ALL THE MACs : I've seen MBPs and Minis with same behavior. Apple seems to be much more focused on noise than cooling

 

[smacman]

45' C is not something to worry about, flawed or not

Your missing the whole point. The heat is a symptom of the real problem, which is grossly high power draw (near 100% on my quad core Mac Pro) whenever audio is played, AND a performance loss of 25%. Not worth worrying about? Do you even have a Mac Pro?

 

[Max(IT)]

Your missing the whole point. The heat is a symptom of the real problem, which is grossly high power draw (near 100% on my quad core Mac Pro) whenever audio is played, AND a performance loss of 25%. Not worth worrying about? Do you even have a Mac Pro?

I missed nothing. He said he's not overly worried and he's right: nothing to' be so worried for a 49'C cpu's temperature.
There is an issue, but in his case it's not so bad

 

[Inconsequential]

I missed nothing. He said he's not overly worried and he's right: nothing to' be so worried for a 49'C cpu's temperature.
There is an issue, but in his case it's not so bad

49C is fine, agreed, but I think thats because he has a octo not a quad, the quad suffers more for some reason.

The performance loss AND power draw increase IS the major issue.

Works like this:

1. Playing Audio/FW device/FW or USB Audio device =
2. Undetected CPU loading of one, maybe two cores
3. Usage of these cores reduces system performance (as a whole) by 25% for quads and (untested) 25% for octos.
4. This increased core usage = higher power draw
5. Increased power draw = higher temperatures
6. Increased temps towards design limits shortens the MPs life in some cases.


So from what I can tell, the quads heat up more as their CPU is being hammered harder than the octos as maybe the octos have one core on each CPU being used up?

Im willing to bet that if you disabled the second CPU in the octos you would get the same behaviour.



So to summarise:

It is NOT the temperatures that you should really worrying about, its the LOSS OF PERFORMANCE OF A £££££ MACHINE.

In my case im loosing £500 of the £2k I paid for the thing! And if its 25% for the octos then thats £625 and up of your machine your loosing performance wise!

 

[Transporteur]

So from what I can tell, the quads heat up more as their CPU is being hammered harder than the octos as maybe the octos have one core on each CPU being used up?

From what I remember from my tests, only one CPU heats up, the other stays on the same level.

That said, I don't expect the performance lost of the system to be 25%. 25% equals one core of the system that is being used to 100%. If this issue affects only one core, the octo looses 'only' 12.5%.
But that's just my expectation.

 

[Inconsequential]

From what I remember from my tests, only one CPU heats up, the other stays on the same level.

That said, I don't expect the performance lost of the system to be 25%. 25% equals one core of the system that is being used to 100%. If this issue affects only one core, the octo looses 'only' 12.5%.
But that's just my expectation.

I personally agree its one core being used up but performance is 24.6% if you want to be accurate on my system. I will when I get back to machine do more serious testing. 1080p video encoding, VMWare W7 benchmarks, lightroom etc.

We need someone with an octo to test their performance loss, im expecting 12.5% ish personally, but 25% would be an interesting find aswell.

Either way, ANY 2009 Mac Pro owner should contact apple about this even if they don't care their machine has issues, do it for the rest of the Mac Pro community.

 

[smacman]

I missed nothing. He said he's not overly worried and he's right: nothing to' be so worried for a 49'C cpu's temperature.
There is an issue, but in his case it's not so bad

It's not about the temp! He is lucky enough to have a dual processor machine so although it heats up, not to the same extent as a quad core machine. He is however still sucking way more energy out of the wall and losing substantial performance by playing audio. If you bought a brand new car, and whenever you turned the stereo on, it consumed 25% more fuel, ran hotter, and went slower, would you not be annoyed?

 

[lbodnar]

Thermal probe (i.e. K type thermocouple in a copper block) mounted between the CPU and cooler.

Proper technique is described here starting from page 17: http://www.intel.com/Assets/PDF/designguide/313067.pdf

Copper block should not be used because it will add thermal inertia, heat averaging due to high conductivity and heat dissipation through edges.
Micro thermocouple attached directly to the IHS is the way to go.

 

[eyeruh]

I personally agree its one core being used up but performance is 24.6% if you want to be accurate on my system. I will when I get back to machine do more serious testing. 1080p video encoding, VMWare W7 benchmarks, lightroom etc.

We need someone with an octo to test their performance loss, im expecting 12.5% ish personally, but 25% would be an interesting find aswell.

I tested this on my Early 2009 Octo (2.26) a few days ago using Lightroom since that's the most taxing application I use on a daily basis. My test involved exporting 100 raw images from my Canon 40D (average of 12MB each) as full-sized JPEGs using Lightroom 2. Lightroom 3 pushes the multicores in my MP harder but it's still an early beta so I stuck with the latest release version (2.6). CPU usage fluctuated between 300% and 500%.

Bottom line is that the test results are consistently 19% slower when the MP is playing audio (187 seconds vs 158 seconds). I ran 3 runs both ways to ensure that all the images were loaded into RAM so disk-access times weren't a factor.

While I ran the first set of 3 runs (without audio), the CPU temps rose from 44C/34C to 52/41. I then quit Lightroom and let the CPU temps drop to their original level before playing music with iTunes. Over the next 20 minutes, CPU temps rose from 42/33 to 60/45 before stabilizing--just from playing audio. I then let the audio continue while running the 2nd set of 3 runs and temps increased from 60/45 to 66/49.

The CPU temps are worrisome but the fact that I'm losing 19% of my performance on a machine in this price-range just from playing mp3s is flat-out obscene. If this happened on a PC, you'd never hear the end of "PC's are terrible for music, get a Mac".

 

[Inconsequential]

*Removed*

 

[nanofrog]

The performance loss AND power draw increase IS the major issue.

Works like this:

1. Playing Audio/FW device/FW or USB Audio device =
2. Undetected CPU loading of one, maybe two cores
3. Usage of these cores reduces system performance (as a whole) by 25% for quads and (untested) 25% for octos.
4. This increased core usage = higher power draw
5. Increased power draw = higher temperatures
6. Increased temps towards design limits shortens the MPs life in some cases.

It appears it's the QPI, not the cores that's getting loaded (the USB, FW, and audio chip data use the PCI links in the chipset, then transfer via QPI to the CPU). QPI does NOT figure into the CPU % utilization figure reported, nor does the IMC or controller. But they still generate heat. If it gets hot enough, the controller will throttle the cores back (it's a protection feature designed into the chip).

So from what I can tell, the quads heat up more as their CPU is being hammered harder than the octos as maybe the octos have one core on each CPU being used up?

Im willing to bet that if you disabled the second CPU in the octos you would get the same behaviour.

The DP based systems (Octad's) have dual QPI's to the chipset. The SP systems only have one QPI. So the Quads can't benefit from load sharing across the 2x QPI's.

Regardless, it's a real problem and needs to be addressed. Some systems have hit the upper 80'sC (~88C IIRC), and is outside of Intel's Thermal Profile, no matter if it's the short term or long term model.

 

[Inconsequential]

It appears it's the QPI, not the cores that's getting loaded (the USB, FW, and audio chip data use the PCI links in the chipset, then transfer via QPI to the CPU). QPI does NOT figure into the CPU % utilization figure reported, nor does the IMC or controller. But they still generate heat. If it gets hot enough, the controller will throttle the cores back (it's a protection feature designed into the chip).



The DP based systems (Octad's) have dual QPI's to the chipset. The SP systems only have one QPI. So the Quads can't benefit from load sharing across the 2x QPI's.

Regardless, it's a real problem and needs to be addressed. Some systems have hit the upper 80'sC (~88C IIRC), and is outside of Intel's Thermal Profile, no matter if it's the short term or long term model.

Ah now that makes much more sense! Tar Dude.