Audio Decoding KILLS MacPro (2009) green factor

[Max(IT)]

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?

I know it is annoying, and I'll try to get it fixed, but I just said that it is not something to be overly worried about.
Speaking about the car, in your example, if it is the car I use to do my job, I'll use it with the stereo switched off, while waiting for the manufacturer to fix the problem. I'm not denying the issue (we have something like 12 Mac Pro's, different models), I just said that in his specific case (49 degrees C cpu' s temperature) it is not an issue to be overly worried about.
The loss of performance seems to be the real problem here.
Cheers

 

[smacman]

Speaking about the car, in your example, if it is the car I use to do my job, I'll use it with the stereo switched off, while waiting for the manufacturer to fix the problem.

That's great if the manufacturer actually acknowledges the problem, or at least has a track record of fixing issues. Not the case with Apple.

I'm not denying the issue (we have something like 12 Mac Pro's, different models), I just said that in his specific case (49 degrees C cpu' s temperature) it is not an issue to be overly worried about.
The loss of performance seems to be the real problem here.
Cheers

Agreed... The temperature is only a symptom of the real problem. The excessive power consumption and loss of performance are worrying however..

 

[VirtualRain]

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.

That's a good theory behind why the CPU utilization is not reported, but it doesn't explain the 30-deg increase in core temps or the 70W increase in power draw. All four Tjunction temps are increasing on my rig... and I doubt a relatively small portion of non-core logic like the QPI can affect core temps or power draw to that level through convection, conduction or anything else. Having said that, I'd be happy as hell to be wrong if they fix it!

 

[dryjoy]

I know it is annoying, and I'll try to get it fixed, but I just said that it is not something to be overly worried about.
Speaking about the car, in your example, if it is the car I use to do my job, I'll use it with the stereo switched off, while waiting for the manufacturer to fix the problem.

Just spare a thought for those of us who use our Mac Pros exclusively for audio work, though!

 

[nanofrog]

That's a good theory behind why the CPU utilization is not reported, but it doesn't explain the 30-deg increase in core temps or the 70W increase in power draw. All four Tjunction temps are increasing on my rig... and I doubt a relatively small portion of non-core logic like the QPI can affect core temps or power draw to that level through convection, conduction or anything else. Having said that, I'd be happy as hell to be wrong if they fix it!

Actually, it could, assuming we're on the right track IMO. I just didn't get further into it.

I'm seeing it as a cascade event. Bad code producing unterminated loops, which keeps the QPI highly active (if not continually for the actual bandwidth needed if executed properly) while the program is running (i.e. audio, FW or USB devices plugged in,...). That in turn keeps the controller loaded with instructions waiting for data that won't arrive, in which it responds by keeping the cores powered up, but not functioning (burning clock cycles with no data flow). All this stalled activity keeps the controller from shutting the cores down (the level of activity prevents the power saving features from ever kicking in while the code is active). All these sections of the CPU are burning power, and generating heat. Yet not actually doing any work.

Hope this makes sense.

 

[Max(IT)]

That's great if the manufacturer actually acknowledges the problem, or at least has a track record of fixing issues. Not the case with Apple.

I'm quite confident about Apple knowing this issue pretty well, trying to fix the problem but unable to find a solution so far.
They take a look at this and other forums on the web, including their own

Agreed... The temperature is only a symptom of the real problem. The excessive power consumption and loss of performance are worrying however..

the loss of performance is the major issue here, considering the typical Mac Pro's user

 

[nanofrog]

I'm quite confident about Apple knowing this issue pretty well, trying to fix the problem but unable to find a solution so far.
They take a look at this and other forums on the web, including their own

The solution may in fact be the necessary re-writting of the compiler used to properly accomodate Nehalem architecture in OS X. That's a substantial bit of work.

the loss of performance is the major issue here, considering the typical Mac Pro's user

For some, yes. Others may not notice, depending on what they're doing at any given moment. (Not sure what the actual split would be % wise).

But either way, it needs to be fixed.

 

[frimple]

Not to be a negative nancy, but are you guys chasing your tails with XBench? The last release was in 2006! It's not being actively developed anymore and it's been shown to have issues in and by itself. For example, you can run the test, get a beer come back and run it again and get a different CPU RESULT!

I like the thermal evidence you've gotten, but IMO you need to throw your opinions about CPU utilization based on XBench out the window. How do you know that XBench doesn't discount a core entirely unless it can use it (and by use it I mean if iTunes has it locked it can't use it)?? As others have mentioned that would be the 25% drop for quad users!

 

[smacman]

Not to be a negative nancy, but are you guys chasing your tails with XBench? The last release was in 2006! It's not being actively developed anymore and it's been shown to have issues in and by itself. For example, you can run the test, get a beer come back and run it again and get a different CPU RESULT!

I like the thermal evidence you've gotten, but IMO you need to throw your opinions about CPU utilization based on XBench out the window. How do you know that XBench doesn't discount a core entirely unless it can use it (and by use it I mean if iTunes has it locked it can't use it)?? As others have mentioned that would be the 25% drop for quad users!

Your probably right about Xbench. It is a bit aged to be trustworthy.

I have also noted a 10-15% loss in Geekbench, which is benchmarking the processor and memory, so something is definitely happening. Interestingly, it's the memory performance that takes the biggest hit in Geekbench. The only app I could find that pushes all cores hard enough to notice this hit is handbrake which also showed a 20% loss.

 

[Inconsequential]

Not to be a negative nancy, but are you guys chasing your tails with XBench? The last release was in 2006! It's not being actively developed anymore and it's been shown to have issues in and by itself. For example, you can run the test, get a beer come back and run it again and get a different CPU RESULT!

I like the thermal evidence you've gotten, but IMO you need to throw your opinions about CPU utilization based on XBench out the window. How do you know that XBench doesn't discount a core entirely unless it can use it (and by use it I mean if iTunes has it locked it can't use it)?? As others have mentioned that would be the 25% drop for quad users!

True, but I haven't had time to do other tests and the machine is in its box with my parents atm!

I shall do some more thorough tests with the apps I use (Lightroom, VMWare, Quicktime encoding, etc) to see what happens.

Although don't doubt XBenches results that much, mine and the in-store MP got within 1% the same scores over 4 runs so it's not *that* bad...

 

[smacman]

True, but I haven't had time to do other tests and the machine is in its box with my parents atm!

I shall do some more thorough tests with the apps I use (Lightroom, VMWare, Quicktime encoding, etc) to see what happens.

Although don't doubt XBenches results that much, mine and the in-store MP got within 1% the same scores over 4 runs so it's not *that* bad...

It may also be worth noting that playing audio on other multi-core Macs does not affect Xbench scores. Regardless of testing method, I think it will be hard to conclusively measure the true performance hit, especially given the varying specs of Mac Pros (Quads, Octads, varying clock speeds), but it's fair to say something is getting tied up.

 

[Kallel00]

Can everybody with an 09 Nehalem reproduce this issue, or it's limited cases?



Just asking because on forums we tend to hear only from the 2-3% people having problems while the 97% who aren't, well, they aren't posting on forums

 

[eyeruh]

Well then thats a start, im convinced *every* Mac Pro out there is affected.

I reckon it IS 25% for Octos as well, as LR2 doesn't max out the CPU.

Do you mind downloading and running XBench for me please?

Here are my results from running XBench along with the cpu temps at the end of each run (sorry, I didn't think to record it until the 3rd run):

No iTunes
Run 1: 183.12
Run 2: 182.98
Run 3: 182 - 115/95

iTunes playing
Run 4: 179.01 - 124/106
Run 5: 154.70 - 135/111
Run 6: 145.44 - 138/117
Run 7: 153.24 - 138/115
Run 8: 153.38 - 140/118
Run 9: 151.93 - 144/118
Run 10: 153.74 - 142/117


Run 6 seems like an anomaly -- there may have been a background process like spotlight running. I think it's interesting (but not necessarily surprising) that the slowdown doesn't occur as soon as you start playing audio--it takes a few minutes for the impact to appear.

Hope this helps. Also, my machine is an early 2009 2.26 Octo MP w/12 gigs of RAM. I'm running Snow Leopard but I saw this occurring on Leopard too (as have others).

 

[nanofrog]

Can everybody with an 09 Nehalem reproduce this issue, or it's limited cases?

From the information that's been discovered (individuals who came across the thread have been able to reproduce it), it's the entire line, not a rarity.

 

[Cynicalone]

Can everybody with an 09 Nehalem reproduce this issue, or it's limited cases?

I think it is the entire 09 line.

...

I waited a long time to upgrade my equipment. It makes me kinda sad to know I might have a fatally flawed machine. If it must fail I hope it fails while under AppleCare.

The part that makes me mad is that if this was on the damn iMac/MacBook Pro line it would be fixed already.

 

[nanofrog]

It makes me kinda sad to know I might have a fatally flawed machine. If it must fail I hope it fails while under AppleCare.

Sad? Why not peeved?

Hopefully, it will get fixed, and no damage will occur in the mean time. But if it does, I do hope it happens under Apple Care, as it would mean new CPU's. And they're not cheap.

The part that makes me mad is that if this was on the damn iMac/MacBook Pro line it would be fixed already.

Maybe not in this case, as I suspect it's the compiler they use that needs to be updated (optimizations) for Nehalem architecture, particularly the LGA1366 parts as they're the only ones so far with QPI. Once done, OS X would need to be recompiled, and tested to verify the issue is gone (fixed if not). The i5 & i7 iMacs are all DMI based, and have more in common in that respect to the previous generation MP's, and is why they've been immune with the integrated audio chip (though I've not seen any info one way or the other about using USB and FW devices, but I'll presume it's not an issue for now).

 

[Max(IT)]

Sad? Why not peeved?

Hopefully, it will get fixed, and no damage will occur in the mean time. But if it does, I do hope it happens under Apple Care, as it would mean new CPU's. And they're not cheap.

just relax a little bit
No way you can damage the cpu with this temperatures. You can be upset about the loss of performance, but don't have to be' worried about the cpu's health.

 

[nanofrog]

just relax a little bit
No way you can damage the cpu with this temperatures. You can be upset about the loss of performance, but don't have to be' worried about the cpu's health.

Tell that to those that have loaded it enough to hit 88C+ or so.

Seriously, I am calm. Those that aren't aware of this issue and have exceeded the 67C limit (long term = 24/7), or came to know about it late (they may not know how high it went, let alone for how long), may have reason to be though.

For those who's systems haven't been loaded enough to exceed 67C, there's no damage, as that's the max temp in Intel's Thermal Profile (long term). The short term can go higher, but it's just that: short term. So long as they've not had an excessive amount of time at high temps, it will be fine.

A loss of performance, though sucks, isn't going to damage the processors on it's own, unless the temps exceed Intel's Thermal Profiles (long and/or short term, which can be found in the CPU datasheet <vol. I>).

 

[lbodnar]

Tell that to those that have loaded it enough to hit 88C+ or so.

Are you talking about IHS temperature? SMC would not allow it to go anywhere over 65-67ºC. Otherwise hardware must be seriously faulty. Core temperatures should be left alone as they are not specified. They can reach 100+ºC without any problems.

 

[nanofrog]

Are you talking about IHS temperature? SMC would not allow it to go anywhere over 65-67ºC. Otherwise hardware must be seriously faulty. Core temperatures should be left alone as they are not specified. They can reach 100+ºC without any problems.

The SMC seems to be borked, as the fans aren't spinning up properly. Idle to 67+ C ends up spinning the fans by an additional 100 rpm, from the information posted. Not much at all, and partially why the temps are as high as they are. BTW, Tj max is ~105 - 108C (Intel won't specify an exact temp, but the material isn't going to be capable of higher than what I listed), so at 100C, you're dangerously close. That's electromigration territory, and the longer it occurs, the higher the potential for damage.

 

[smacman]

Are you talking about IHS temperature? SMC would not allow it to go anywhere over 65-67ºC. Otherwise hardware must be seriously faulty. Core temperatures should be left alone as they are not specified. They can reach 100+ºC without any problems.

My IHS temp reaches 65-70 within minutes when I play audio. The SMC cooling logic is definitely flawed.

It also seems like the Quad chips are the most critical with respect to temperature, which makes sense I guess.

 

[Max(IT)]

Tell that to those that have loaded it enough to hit 88C+ or so.

Seriously, I am calm. Those that aren't aware of this issue and have exceeded the 67C limit (long term = 24/7), or came to know about it late (they may not know how high it went, let alone for how long), may have reason to be though.

For those who's systems haven't been loaded enough to exceed 67C, there's no damage, as that's the max temp in Intel's Thermal Profile (long term). The short term can go higher, but it's just that: short term. So long as they've not had an excessive amount of time at high temps, it will be fine.

A loss of performance, though sucks, isn't going to damage the processors on it's own, unless the temps exceed Intel's Thermal Profiles (long and/or short term, which can be found in the CPU datasheet <vol. I>).

ok, you're right, but if the problem is related to playing audio, I don't think that people are going to listen iTunes 24/7, aren't they ?
And a cpu can work at 70°C 8-10 hours/day without any problem. As you said Thermal Profiles for long term speak about 24/7.

 

[nanofrog]

ok, you're right, but if the problem is related to playing audio, I don't think that people are going to listen iTunes 24/7, aren't they ?
And a cpu can work at 70°C 8-10 hours/day without any problem. As you said Thermal Profiles for long term speak about 24/7.

70C isn't a dangerous level though (that's really just the limit, and the +3C isn't that big a deal, as there's enough safety margin). It's those running over 81C (top temp in the short term profile IIRC) that have a right to be concerned though, particularly those that have seen 88C.

Also keep in mind, it's not just with audio, but FW devices such as external HDD's have caused similar behavior too (presumably USB would as well). So it's more wide spread than originally thought.

Long term use at high temps is actually possible, especially in the cases where the user has either an FW or USB audio device, and runs it for hours to get work done. It could also be possible with external drives, though for that much data (to run for hours and hours, say over 8hr/day in backups), I'd think an eSATA card + PM enclosure would make more sense, and avoid this issue by happenstance (PCIe isn't a causality in this mess ).

 

[smacman]

ok, you're right, but if the problem is related to playing audio, I don't think that people are going to listen iTunes 24/7, aren't they ?
And a cpu can work at 70°C 8-10 hours/day without any problem. As you said Thermal Profiles for long term speak about 24/7.

Just launching Logic Pro, and doing nothing else, results in core temps above 88C on my machine.

 

[DoFoT9]

ok, you're right, but if the problem is related to playing audio, I don't think that people are going to listen iTunes 24/7, aren't they ?
And a cpu can work at 70°C 8-10 hours/day without any problem. As you said Thermal Profiles for long term speak about 24/7.

wow. serious? i know people who never stop listening to music, me included! in the car, when im at the computer (90% of the time), when drumming, when listening to my ipod.. my computer hardly ever stops playing music..

ill admit the majority of people wouldnt listen 24/7, but a substantial amount would.