Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Audio Decoding KILLS MacPro (2009) green factor
- Thread starter handheldgames
- Start date
- Sort by reaction score
Mac Pro. The forgotten child.
It seems as if Apple does not care about it's so called 'Pro' line at all.
People complain about performance issues on the notebook line, shortly after that we've got an update.
People complain about serious hardware issues on the Nehalem Pro, no update after almost a year and it's very unlikely that we will ever see an update.
Well done Apple!
It seems as if Apple does not care about it's so called 'Pro' line at all.
People complain about performance issues on the notebook line, shortly after that we've got an update.
People complain about serious hardware issues on the Nehalem Pro, no update after almost a year and it's very unlikely that we will ever see an update.
Well done Apple!
More and more things come to light every day.
If it wasn't for the fact i've submitted a bug report to apple i'd email places like engadget to show the story of the issues.
CPU uses 50W more, 50W!!! When just playing audio.
Ridiculous.
If it wasn't for the fact i've submitted a bug report to apple i'd email places like engadget to show the story of the issues.
CPU uses 50W more, 50W!!! When just playing audio.
Ridiculous.
Just the '09 MP's.
50W is quite a bit, especially when the CPU utilization is so low (less than 1%!).
Well I don't know about you guys, but since I discovered that this is more than a thermal issue (performance hit as well when playing audio), now I am REALLY resolved to do something about this. If Apple doesn't want to support their machines, then I'll make it a customer service issue. It may just be time to stop playing nice, and try and get this issue the widespread coverage it needs.
Please find time to submit your feedback here: http://www.apple.com/feedback/macpro.html
If *all* Mac Pro Nehalem owners can submit their feedback, I certainly hope it will get Apple's attention. I already submitted mine. Cheers.
If *all* Mac Pro Nehalem owners can submit their feedback, I certainly hope it will get Apple's attention. I already submitted mine. Cheers.
Yup! I agree. I have already done that. I have even opened up two support tickets with Apple only to hear back that an 89C CPU core is within limits, or that this behavior is by design. I think it's time to step it up a notch.
I haven't really had this issue, so I'm wondering
right now my computer is running the kernal_task process at 100% and has been doing so for the past week. It doesn't really bother me since I have all this extra processing power, but is this the same as the audio issue or am I having another issue entirely?
It's a different issue it seems.
The audio portion is when the underlying Quicktime X code is used (iTunes, Flash,... but CPU utilization is under 1%), and the temps hit ~67C+.
What're the temps at BTW in your situation?
And have you done a cold boot to see if it changes?
I just read through this thread last night.
My idle temps are high (usually around ambiant + 25-30 deg. C) but playing music in iTunes has very little impact on my temps. I see maybe a 2-3 deg increase in temps.
My high idle temps could be one of three things:
1. Cooling system design (trading off effectiveness for silence)
2. Poor TIM application
3. Incorrect temperature readings
I recall near the beginning that Tesselator questioned the validity of the temperature readings... was this ever explored and dismissed as a possible issue?
If I understand how CPU temperature calculations are done, they need to account for the TjMax of the specific processor they are monitoring. You can see from the changelog of a popular PC temperature app, how often this is done... http://www.alcpu.com/CoreTemp/
Here's a post from a relevant thread on XS...
So there are several questions that need to be answered about the temperature reporting tools we are using such as what value are they assuming for TjMax and are these relevant for the Xeon parts in our machines?
If we don't know exactly how these tools are determining temperatures and what assumptions they are making, there's no way we can be sure that the fluctuations are accurate, to scale, or not being influenced by something other than the actual temperature of the silicon.
Also, can anyone explain how a 20-deg. increase in temps occurs with 1% or less CPU utilization? Nano, do you think poor compiler optimization could possibly load a processor but somehow escape being accounted for by activity monitor?
My idle temps are high (usually around ambiant + 25-30 deg. C) but playing music in iTunes has very little impact on my temps. I see maybe a 2-3 deg increase in temps.
My high idle temps could be one of three things:
1. Cooling system design (trading off effectiveness for silence)
2. Poor TIM application
3. Incorrect temperature readings
I recall near the beginning that Tesselator questioned the validity of the temperature readings... was this ever explored and dismissed as a possible issue?
If I understand how CPU temperature calculations are done, they need to account for the TjMax of the specific processor they are monitoring. You can see from the changelog of a popular PC temperature app, how often this is done... http://www.alcpu.com/CoreTemp/
Here's a post from a relevant thread on XS...
Link
So there are several questions that need to be answered about the temperature reporting tools we are using such as what value are they assuming for TjMax and are these relevant for the Xeon parts in our machines?
If we don't know exactly how these tools are determining temperatures and what assumptions they are making, there's no way we can be sure that the fluctuations are accurate, to scale, or not being influenced by something other than the actual temperature of the silicon.
Also, can anyone explain how a 20-deg. increase in temps occurs with 1% or less CPU utilization? Nano, do you think poor compiler optimization could possibly load a processor but somehow escape being accounted for by activity monitor?
That is a bit high for idle temps.
The cooler's contact surface may be rather rough as well, as faster machining lowers costs, but leaves a rougher surface. I've lapped a few coolers, and it's made a notable difference. Of course the TIM is re-applied, but I've even tested that (pulled the cooler, re-applied TIM, tested, pulled again, lapped, re-applied TIM and tested). This can still leave a little variance with TIM application, but not as much as I'm accustomed to seeing from parts off the line.
The combination gave the best results, but the TIM out of the factory may be good to start with, and the surface the issue.
BTW, what is the room's ambient temp typically?
As per the temp being accurately reported, it could be off, unless it's been calibrated (i.e. K type thermocouple set in a Cu base, mounted between the CPU and cooler + TIM on both sides).
But I'm going on the fact the rise in temperature would be accurate when comparing idle to working temps under testing conditions (same system, so the differential is accurate for this purpose). ~25 - 30C rise for less than 1% CPU utilization is way off (IIRC, idles have been reported ~40C, less in some cases).
Testing under Windows, the temps stay the same during testing. Short or long term duration (I've tried between a couple of minutes to a few hours). No significant change in temps reported (a degree or two is all).
This is the real question.
I've gone back and checked Intel's ark site, to see if there's a new stepping that might account for differences. There's not, as everything listed in both the 3500 and 5500 series are listed as D0.
Now it seems the Nehalem based iMacs aren't afflicted with this problem, and are also D0 parts. So the difference architecturally speaking, is the LGA1156 parts use DMI to communicate to the chipset, while the LGA1366 parts use QPI (DMI is still used to attach the ICH10R to the chipset). So it would seem that what ever the cause, would lie in the QPI communication from a hardware perspective (more specifically, I'm thinking in how it's attached to the controller).
So that leaves either the OS or the firmware, as the possible culprits. These issues aren't present under Windows on PC's or MP's from what I've seen.
Now if it were the firmware, it's in the EFI aspect, as Windows is loaded under BIOS emulation. Otherwise it's the software's lack of optimization for Nehalem (unterminated loops for example, and could explain the temp rise - in the controller, not the cores, which isn't reported in the CPU % utilization).
Beyond this, I'm not sure, as there's no other information I'm aware of.
i wonder if the format of the audio plays a role in temperature increases? has everybody so far only tested in mp3? i currently only have AL (Apple Lossless) on my iMac to test. temperatures jumped 3°C from 39°C to 41°C (with iTunes using ~5%/800% CPU).
I'm not sure, but the decoding isn't that strenuous on the CPU. The answer could give a clue though.
Worth checking at this point, as it's something we can try.
yup even AL isnt strenuous at all. max of 1411kb/s is nothing!
that's what i was thinking. anything could help!Worth checking at this point, as it's something we can try.
I agree, but as I said, there are several potential reasons for it. Factory applied TIM is notoriously poor in my experience and as you can see below, I don't know that we have any reason to trust the software we're using here at all to report temperatures. BTW, my Ambiant is around 20-deg. C usually.
Are the actual temps reported in Windows utilities, the same as reported in OSX utilities? That is... an idle temp of 40-deg. C in OSX is 40-deg. C in Windows? You've done this testing yourself... on a 2009? Did you report the details in this thread? Did I miss them?
Yeah, I don't know why this isn't getting more attention here. The first thing I would assume from this craziness is that the software being used to report temps is out to lunch... yet most people up in arms about this issue seem to think it's infallible! What evidence do we have that it's not seriously bugged or has an issue with Nehalem Xeon chips?! Could this not be a red herring? EDIT: It might also be the firmware... the firmware is not reporting/transferring the correct values? This might also explain why the iMac's don't appear to have this issue... they may have had this issue fixed in their firmware?
I understood, and realize the potential issues. I'd go in pull the CPU's, and re-apply the TIM myself to fix the "slopped on TIM", assuming that even happened (I've seen it done nicely, and where it looked like peanut butter melted out onto the board
).As per the temps being accurate, NO, they're not, unless the software's been calibrated properly. But the rise is accurate (assuming nothing else has been damaged, moved,...). That's what got my attention. (I'm not sure if Apple installed a separate CPU temp sensor, or uses the one on the die, as most software apps use).
Again, it's in the rise, not the reported temp (working - idle = rise).
Yes, I've done the testing, and I've calibrated mine now. But it's a PC based system, not an MP. So there's going to be differences, such as coolers used, airflow, baffling,... So I'm looking mostly at the rise. That measurment is still accurate, and allows for the comparison of Windows vs. OS X.
And yes, there can be quite a difference in reported temps to real, prior/post calibration. Up to 15C, depending on the processor/s used. The differences have been increasing from C2D to Nehalem prior to calibration.
You have a point, that the temps themselves can't be relied upon as absolutely accurate. They can't be, unless it's been calibrated, which requires a temp sensor of some sort (thermocouple or IR temp reader).
But my point, is the rise is accurate.
Variance + real working temp = reported working temp
Variance + real idle temp = reported idle temp
Reported Working - Reported Idle = real differential (or temp rise)
The variance is eliminated, even though the temps themselves may be off by 15C.
It's applicable for the same system (comparing OS's to one another) or different systems (MP vs. i7 iMac for example). But only for comparing the rise, not reported idle or working temp.
This could be the case, and why I responded to DoFoT9's post that even testing the file types (mp3, lossless,...) might produce something useful.
Do we know that for sure? How? In my mind, a bug in the temp. reporting could just as easily affect scale as absolute values. Is a reported 20-deg rise, really only a 2-deg rise? 10-deg? Again, how can we trust the information being provided to us is correct?
Another big clue would be whether the cooling system kicks into high gear? I recall reading that this wasn't happening. Is that your recollection as well or are the fans going high speed with this issue?
If the cooling system is not kicking into high gear... why not? It could be either ignoring the high temps (very bad design) or getting different temp data than we are (which might explain this bizarre issue).
The calibration is just an offset (i.e. reported = x, real = y <measured>, offset = reported - real), which is set in the temp application.
What you're getting into, is the variance of the temp sensor itself. It's designed to be accurate within the range it will be exposed to. Now if you had a device that's only rated to say 0 - 75C, and it's exposed to say 108+C max, it's going to give inaccurate results (possibly be destroyed this way). This isn't the case though. Intel's not that stupid (I hope
).Now in the case of MP's, IIRC there's a thermocouple in the cooler that reports temps. So does the CPU. Now which the fans are set to, I can't be certain, but I'd think it's off the cooler located sensor. So the fans stay in check, given the SMC settings Apple created (+25C isn't that horrible, and the stock rpm values are likely adequate for that temp, as like you, I don't recall posts indicating the fans ramped). The baffle around the CPU section will help, as does i'ts physical location in the bottom of the case (i.e. no heating of intake air by the graphics card or any other PCIe device,... in the upper sections of the case).
This sensor is calibrated (to a good extent anyway), by Apple via test systems to gather sample data. The data during this testing would have been used to determine the settings in the SMC code. So it's variance is likely within 1 - 2C (takes case,... into consideration, as it's based on real data for the specific system).
I'm assuming the temp reporting may be coming off the CPU's sensor.
Good questions, no doubt, but the rise should be accurate for test comparisions (assuming nothing was seriously borked on the SMC/thermocouple selection, data or physical placement aspect). Not as likely as software (Quicktime X) or firmware anyway.
No... that's not at all what I'm suggesting. It could be a simple matter that there's a bug in the firmware or software used to report temps that's doing bad math and it could be doing any number of bizarre things to the final output.
The fact is, I don't believe anyone's established that the reported temps OR rise is actually real... have they?
Look at the related info:
- Temp's rise 20+ deg. C
- Less than 1% CPU utilization
- No ramp in fan speeds
It seems misguided to assume the temp. increase is actually happening. Until someone gets an IR thermometer on the heatsink to determine the actual temp. increase, the most logical explanation is that there is some bug in the firmware/software.
There could also be other strange things at work as well. For example, I once had a PC who's temps shot up just like this then when I plugged in a particular USB Bluetooth dongle. It could be that the BT dongle was somehow mixing up register values that the firmware was using because the CPU certainly wasn't getting hot. I'm sure you've seen equally strange things that turn out to be somewhat different than initial evidence might indicate.
VirtualRain,
I know what you're trying to say but the temp increase is real! Like nanofrog said, the actual temp values may be slightly inaccurate, but there is definitely a marked rise in temp when playing audio. In my case, playing audio on my Mac Pro has the same effect that putting a space heater under my desk would. Within minutes, I can feel hot air rising from under the desk. Other temperature sensors that are proximal to the CPU also backup this rise (ie. Memory Module Temps all increase by a few degrees when playing audio. Power Supply temps also increase due to the 50W CPU power draw).
Let's not forget as well that this is not simply a thermal issue. In addition to excessive unnecessary temps, this is also a performance issue, and an excessive power draw issue. Xbench tests done while playing audio are reflecting a CPU score that is 25% lower. 50W+ of power draw just to play a song in iTunes is also less than ideal.
Apple can claim that these temperature reading tools are unsupported, or that the temps are still within limits, but these last two issues are even more serious in my opinion. It is a reasonable expectation when buying an expensive workstation, to be able to playback audio without taxing the system.
I know what you're trying to say but the temp increase is real! Like nanofrog said, the actual temp values may be slightly inaccurate, but there is definitely a marked rise in temp when playing audio. In my case, playing audio on my Mac Pro has the same effect that putting a space heater under my desk would. Within minutes, I can feel hot air rising from under the desk. Other temperature sensors that are proximal to the CPU also backup this rise (ie. Memory Module Temps all increase by a few degrees when playing audio. Power Supply temps also increase due to the 50W CPU power draw).
Let's not forget as well that this is not simply a thermal issue. In addition to excessive unnecessary temps, this is also a performance issue, and an excessive power draw issue. Xbench tests done while playing audio are reflecting a CPU score that is 25% lower. 50W+ of power draw just to play a song in iTunes is also less than ideal.
Apple can claim that these temperature reading tools are unsupported, or that the temps are still within limits, but these last two issues are even more serious in my opinion. It is a reasonable expectation when buying an expensive workstation, to be able to playback audio without taxing the system.
Ok, perhaps I'm harping too strongly on the software being at fault. It's just that there's a lot of assumption making going on in my opinion and I don't think there's enough information to lay this problem at Apple's feet.
I personally don't have this problem. See my screens shots of iStat with both no music playing and music playing. The temps just aren't changing and my Xbench scores are the same whether iTunes is playing music or not. Therefore, this problem must in some way be environmental and related to your configuration.
Are your cooling fans ramping up when your computer becomes a space heater?
Could it be other hardware in the system that's causing a bizarre quirk? Is it the audio interface people are using? Is there something in common between people that have the problem besides just a Nehalem Xeon? A standard trouble-shooting technique is to pull every non-essential piece of hardware from the system to see if that helps... maybe that's worth a try?
I recall a problem where people with an ATI 4870 in their system were having problems with iTunes crashing when the computer resumed from sleep... that's a good example of the kinds of bizarre interactions that can occur in a computer. Perhaps that's even related to this issue?
I'm as stumped as anyone, and I realize I'm seriously late to this party, so I understand if there's no appetite to continue to trouble-shoot this, but I can't imagine that we don't collectively have what it takes to get to the bottom of this.
Attachments
I'm not making any assumptions. Just stating the facts. My system gets really hot when playing audio. It cools off substantially when I stop playing audio. When it is exhibiting this "hot" behavior, the power consumption rises dramatically, and the CPU scores substantially lower when benchmarked.
Your starting to sound like you work at Apple. So just because your machine may not be affected, the rest of us don't deserve a solution from Apple? Why people like you make statements like this boggles my mind. What do you have to lose if Apple investigates and fixes this issue for the rest of us? If you really feel this way, then kindly stay out of this thread as it does not concern you.
I personally do have this problem. So does almost everyone else that has chimed in.
Fair enough. You may not be observing the same issue, but it doesn't mean that the problem does not exist. Several people on the forum have observed this behavior, and I personally have reproduced it on every 09 Mac Pro that I've had the opportunity to test (nearly 10 different machines).
BTW - Your iStat CPU reading of 55C is measured on the heat sink. In other words, your heat sink is approximately 55C. Add 5-10C to that for your CPU temp. If you want a much better temperature reading, use TEMPERATURE MONITOR which reports much more extensively. It sounds to me like not only are you suffering from the same problem, in your case, it does not cool off even when you stop playing audio. That would explain why your benchmark stays the same too. The question is, what else do you have running that may be keeping your CPU in this strange state? Make sure all your browsers are closed, and anything else that has anything to do with audio / flash rendering etc. is closed. 55C on the heat sink is way too hot for idling. My heat sink temp at idle is 32C. People with these chips overclocked to 4.0Ghz+ on Windows machines report those type of heat sink temps under heavy load.