|Re: 2009 Mac Pro High CPU Core Temperatures w/OS X audio playback (or FW I/O)|
(Notes on Turbo mode, Temperature monitoring in OS X vs Windows)
Updated: Dec 3rd (more notes from Author of Temperature Monitor)
Updated: Dec 4th (notes on more Vista/HWinfo tests/core speeds during MP render tests)
Updated: Dec 5th (short notes/observations using iStat Menus)
Updated: Dec 29th for Quad 2.93GHz owner's test summary
Updated: Dec 31st for report on PCIe Audio cards
Updated: Feb 11th, 2010 - FYI on 'fix' for OS X 10.6.2
Updated: Feb 22nd, 2010 - FYI on 'audio update' causing noise w/FW audio devices
Updated: May 3rd, 2010 - FYI on Audio Update 1.1 (to fix FW audio noise)
(FYI: On Feb 11th, 2010 apple posted a fix for OS X 10.6.2 users - see
2009 Mac Pro Audio Update 1.0.
BTW - A reader said the "Audio Update" also fixed the high wattage/high temps seen during Firewire transfers, but some Firewire audio interface users are reporting the "audio update" has caused a High pitch squeal when using Firewire audio devices. (Apple thread no longer online.)
Update: On May 3rd, 2010 Apple posted Mac Pro Audio Update 1.1 that is said to fix the Firewire Audio noise issue seen w/v1.0 of the update. However still no updates to fix the core issue also seen in OS X 10.5.x.)
Background: The Nov. 27th, 2009 news page had a reader post regarding higher than expected CPU core temperatures w/2009 Mac Pros and Audio Apps/iTunes (copy below) where several owners were concerned over reported CPU core temperatures (in the 80's C or more in some cases) just from playing audio (for instance iTunes but reportedly any OS X Core Audio shows this - iTunes is just commonly used to illustrate the problem). (At least one owner believes that his cores had ran hot enough to cause shutdowns and an eventual failure.) I own an (early shipped) 2009 Mac Pro (dual 2.66GHz) but hadn't installed any thermal monitoring utilities in the past to monitor that. (It isn't the primary work machine but I had used it in 10.5.x, 10.6.x, Vista and Win 7 - OS X use primarily converting video and for tests used in reviews such as the GTX-285 card.) I'd never had a kernel panic, thermal shutdown or any problems to date (even after hours of video encoding in handbrake) but after getting that mail/reading the posts in the apple forum thread decided to install Temperature Monitor in OS X 10.5.8 to check reported core temps with iTunes audio playback. (I also wanted to check/compare reported temps in bootcamp/Windows after playing songs for appx the same duration. And to confirm that "turbo mode" was supported in Windows, as someone had questioned that. See below for my Windows tests on that (HWinfo32 v3.30 reports turbo speeds on all cores with typical use.)
This page is just a catch-all for previous posts in the last few days + some quick tests I ran this weekend. It's not anywhere near a real analysis of this topic - just the results of some quick tests I ran and my ramblings/thinking out loud... Marcel (the author of temperature monitor and hardware monitor) has repeatedly answered questions (and in some cases incorrect assumptions) about how the reported temperatures are derived (following intel and OS X guidelines). And BTW - you can't compare Temperature Monitor's core or CPU reporting to iStat Menus' CPU Temperature reporting (which as of v2.0 is the CPU HEATSINK temperature) - see first tests with iStat Menus below for more on that.)
FYI: Although this page is about CPU/Core temperatures (and turbo/core speeds), someone brought up total system power usage. I didn't originally test for that here, but I did record some data on that back in early June with my 2009 8-core 2.66GHz Mac Pro. See the bottom of the 1st page of my June 2009 GTX-285 review (vs ATI 4870 and GT120) for tests on total system wattage when Idle, during Motion 3 tests, 3D gaming, Benchmarks, etc. with the 3 different video cards. At that time I had only one optical drive and only 2 hard drives installed - now I have 2 optical drives and 4 hard drives inside. The point of those tests (of course) was to show the differences with the various graphics cards. But note even the OS X idle wattage (with no apps running) I saw then ranged from 118W w/GT120 card installed to 168W w/ATI 4870 installed. (In some 3D gaming tests I saw as high as 408W reported on the UPS.)
Update: I re-tested with my current 8-core Mac Pro config (Dual 2.66GHz CPUs, GTX-285 card, Four 7200rpm HDs, 2 Optical drives, a PCIe eSATA card, 12GB RAM, Logitech G15 (w/LCD) keyboard, Logitech Mouse RF wireless USB xcvr, Energy saver sleep disabled, Airport on, BT off, using external 2.0/AC Powered speakers). Only the Mac Pro was connected to the UPS which has an LCD for monitoring Wattage use real-time. As you can see from the results below, simplying playing audio in iTunes resulted in over 100W increase in OS X on this system vs literally nil increase when playing WMP audio in Windows. (I used WMP as I didn't have iTunes installed in windows, but iTunes playback in windows doesn't ramp up temperatures/wattage like OS X either per other 09 Mac Pro owner's tests.)
- OS X 10.6.2 idle (4 min settle): 168W (varies a few watts)
- OS X 10.6.2 w/iTunes audio playback (only) after 15 min: 282W (varies a few W)
- Vista Ultimate 64bit (Bootcamp) Idle: appx 220W (avg)
- Vista Ultimate 64bit (Bootcamp) WMP Audio (15min): appx 222-224W
(Win 7 HD has a lot less apps/addon software installed than Vista. Both set to max perf.)
- Windows 7 Pro 64bit (Bootcamp) Idle: appx 209-216W (varies)
- Windows 7 Pro 64bit (Bootcamp) WMP Audio 15min: appx 218-220W
BTW: A reader asked if I'd check for similar behavior during Firewire file copies. I also saw an appx 120W increase (about a 50% increase from idle wattage) during FW800 file copies. Of course CPU temperatures also ramped up, but I only copied a few GB of files. This was not seen during eSATA I/O - even after hours of constant SATA I/O (580GB clone of internal HD to eSATA HD).
Again however the tests I ran here (below) were just to check CPU Core Temperatures w/audio playback - to see if I could repeat what other owners had noted - 80C or more core temperatures seen after only a few minutes of just iTunes (or any core audio reportedly) audio playback. Plus I wanted to test with Windows audio playback for comparison, and to check CPU core clock speeds reported in Windows. (As no OS X utility that I know of currently will monitor/report the core's clock rates.)
Apparently some are upset at the author of Temperature Monitor, however if you read his explanations (several of them below), it's clear to me that his software is following the proper guidelines/methods. Any complaints over the Mac Pros thermal behavior should be directed to Apple. You can send them feedback on products at www.apple.com/feedback/. The Mac Pro specific form is at www.apple.com/feedback/macpro.html.
(To save going off this page, here's a copy of the posts/comments from the Nov. 27th news page.)
"iTunes problem in Nehalems
I'm surprised I haven't seen any press on this yet. This has been a major topic on the Apple Discussion forums.
Just playing a song in iTunes can raise core temperatures 20+ degrees. In my case I've seen temps rise into the high 80c's, and in hot weather, the the low 90's. Hardware Monitor reports 95C as the shutdown point. (FYI - See several posts below from the Author of Temperature Monitor with answers to questions on TM's reporting/explanations of methods used/info from Intel's docs on the CPU, etc.-M) In this tortured link, you'll see two Macs that have been disabled apparently because of this. And don't think this is the only link of it's kind there, they started when the 09s came out. (ref: Apple forum thread Temperature Rises with audio - thread no longer online.)
This is all the stranger because while the temps go higher, CPU activity shows practically nothing going on. (FYI - Don't just monitor "My Processes" in Activity monitor - monitor All processes.-M) And it seems to affect many audio apps. Two very curious things also are that it only affects Nehalems, not any earlier computers, and if one is running Windows on their Mac the temp rises don't appear when running through that.
All in all an odd situation, and one I personally find troubling given my history of three G5s. And Apple isn't going to address or acknowledge it until it gets some press.
I hadn't previously installed any OS X temperature reporting software on the 09 Mac Pro (dual 2.66) here but have used it for hours of encoding video using Handbrake over the last few months. (Some slight increase in fan speeds noticed but not loud and never had any KP's or other problems so far - personally I'd not been concerned about overheating CPUs.)
After getting the above reader email, I wrote the author of Temperature Monitor/Hardware Monitor on this - here's his reply (from Nov. 27th - there are other replies from him below with additional info/answers to later questions, etc.):
"I guess you are referring to the Digital Thermal Sensors (DTS) of the Intel Xeon E55xx series processors. (the Xeon 55xx series are in the 2009 Dual CPU Mac Pros - the single CPU 2009's use the Xeon 35xx series CPUs)
The internal arithmetic precision by which the CPU processes its sensor values is 1/64 degrees Celsius (0.015625 ° C), the actual accuracy of the sensors is +/- 1°C. The CPU uses internal low-pass filtering before making the temperature values available to applications or its own internal temperature monitoring.
Some people seem to misunderstand how the Turbo Boost feature of the Intel Core i processors is designed to work. It is an INTENDED hardware feature of the processor to overclock itself up to the point until a high but uncritical temperature has been reached, in all cases where the CPU detects load on one core but no load on the other cores.
(Yes, Turbo Mode is not news. I linked to Intel's Turbo Boost Technology page (as well as other intel docs/features of these CPUs) in my March 2009 article here on the 2009 Mac Pro. I'd think that's common knowledge by now.
And I noted in my windows tests below (screenshot from a Hwinfo utility that reports Core clock rates, nothing like that in OS X that I know of) that Turbo mode is shown as working (on all cores) in the screenshots I posted originally here. And note that HWinfo shows that ALL CORES are running in Turbo mode per that utility's reporting. Even after 10 repeated tests of CineBench R10's MP render test (that uses all cores) although I did see some cores clock down, none (on either CPU) ever lowered to the stock X5550's 2.66GHz rate. As I said in that post, I guess even 10 CBR10 MP render tests in a row was not enough to lower the core clocks to their 'rated' speeds. That was just a quick test, I've not run any long (hours) stress tests in windows. (The original reason I posted this was to compare temps from short term audio playback to see if the same rapid rise in temperatures were seen. I've seen 80C+ cores in Windows at times also, but not just from playing audio. And 80C range cores don't worry me - I haven't seen any core temps exceed the limits, nor any thermal shutdowns personally.) But per HWinfo's reported core clock rates, what I'm seeing on this 2009 Mac pro is NOT just turbo mode when "the CPU detects load on one core but no load on the other cores" - CineBench R10's MP render test uses all cores and after a string of repeated MP render tests the cores still were reported as running at over the rated 2.66GHz. (Some did drop to just over 2.8GHz during that series of runs. A longer term test I suspect would lower them further.) But is Hwinfo's reported speeds accurate? Is there an issue with the 09 Mac Pro's Turbo support? I can't say. I'm just reporting what I saw.-M)
The CPU will return to normal operation as soon as there is more load on the other cores. 80-90 ° C is not a critical temperature for the models with higher clock frequencies.
I should add that people in high-performance computing forums are discussing the somehow "opposite" topic "how can we force the new CPUs to always run at the HIGHEST possible temperatures?" to get the maximum performance for single-threaded applications.
(I had also asked if any of his software had support for the LCD displays in Logitech keyboards (like the G15 keyboard or G13 gamepad)
When these keyboards came out - I think in 2007 - Logitech offered no Mac OS X drivers at that time. (There is now, although like their LCC software for OS X, many are leery due to bad experiences in the past)
I didn't check if this has changed since. Due to missing drivers, Hardware Monitor currently has no support for these LCD displays. If I can be of further assistance, please let me know.
BTW - I've noticed lately booting (from power off) to 10.6.2 HD in 2009 Mac Pro that I'm getting startup log entries (repeated 16 times - one for each "virtual" core):
kernel WARNING: ACPI_SMC_CtrlLoop::initCPUCtrlLoop - turbo enabled but no turbo P-state found
I reset the SMC (shut down, disconnect AC power cord, etc.) and re-applied 10.6.2 using the combo updater from D/L with no change. (I don't remember seeing those warnings with a 10.5.8 boot, but want to check again to confirm.) [I'm not alone in seeing the above from a power-up boot w/Snow Leopard - other 2009 Mac Pro owners wrote they have also.]
(Any idea why I'm seeing the above console warnings? (repeated 16 times - one for each "virtual" core)
"Not really. I see the same messages on my system as well. The SMC is the top trade secret of Apple (because it plays a certain role in binding Mac OS X to original Apple hardware), so there is no documentation about it, and all parts of the kernel referring to the SMC have been "censored" in the source code.
The message appears while the kernel is initializing its platform plug-ins. The ACPI_SMC_CtrlLoop is the part of the kernel which organizes power and thermal management of the hardware, always running, and linking the power management parts built into the firmware to their counterparts in the operating system. A "P state" is basically a firmware table which lists at which combinations of CPU voltage, and clock frequencies the CPU is allowed to run and which effects each combination will have on power consumption and temperature. The message suggests that Intel Turbo Boost is enabled, but the corresponding control table for turbo mode is missing in the firmware. (If it matters this April 2009 shipped model has .B03 boot rom, vs .B04 in current shipments. Although the SMC version is the same.-M)
This looks like a bug, but I currently have no information whether this is a critical or harmless defect, or if it may have any negative influence on turbo boost operation.
Best regards, Marcel"
(All the discussion above from Friday finally made me make time to install Temperature Monitor on mine.)
My Initial Test Results (OS X vs Windows):Temperature Monitor Readings after 3 straight iTunes songs played:
Over the weekend (Nov. 28, 2010) I installed (freeware) Temperature Monitor in OS X 10.5.8 on a Dual 2.66GHz 2009 Mac Pro and used it to monitor core temperatures during iTunes audio playback and Handbrake video encoding/conversion.
After playing 3 songs in a row in iTunes (less than 15min total time) some cores reported as high as 82°C. (Not exceeding the limit listed in TM, but much higher than reported by a Windows utility (HWinfo32) in Bootcamp/Vista 64 after playing music (a sample file in WMP) for the same duration - which showed a max core temp of 62°C during that time - see screenshots and additional info below.) [FYI - Room temperature was appx 72F (winter here) and the Mac Pro was on a carpeted floor under a home-made desk close to the wall. Not great airflow but have never seen a problem in months of use. The Ambient sensor in the Mac Pro during testing reported 80-84F, much higher than avg room temp.)
By default in TM (unlike Windows utilities), 'virtual' cores (even numbers) are also shown (there's a pref setting where you can remove them from a display list and set TM to not acquire readings from them, but they still appeared in the window listing, just w/o any values. I didn't notice that option when I first ran these tests.) I'd prefer to have only real cores shown. (And BTW - later tests of the above but for about 35 minutes resulted in similar readings - no cores had gotten significantly higher than the 15 min test.) Note the CPU A (and B) Temperature Diode reading is a few degrees C higher than after 25 min of HandBrake video encoding (screenshot below) with much higher CPU/cores load. And as I've seen before, the cooling fans didn't ramp up much for this (even after 30min of iTunes playback like this with cores in the 80's, fan rpms still reported as under 1200rpm. Some have used SMCFancontrol to tweak fan speeds (on many models), but I've never installed it personally.)
(FYI - For an explanation of why a core above has "---" vs a value see Marcel's explanation below.) The Intel thermal specs doc for the 5500 series Xeons (PDF file) notes normally Tcase Max (that's not the core temps - pg 103/para 6.1.2 notes Tcase is measured at the geometric top center of the processor integrated heat spreader) should not be reached as TCC (thermal control ckt) should kick in. (Some may have confused Tcase max with CPU core limits. At first I thought maybe the "CPU Temperature Diode" TM monitors might be related to Tcase, but Tcase max references are for system designers to measure for ensuring a proper/adequate Heatsink/thermal design in products that use this CPU.) The Intel PDF has several charts on various 5500 series (5500 series are in the Dual Processor 2009 Mac Pros - the single CPU models use 3500 series CPUs). For the "standard" 5500 series that doc's Fig 6-3 shows 76C Tcase max but regardless Tcase max is not referring to the core(s) temperature. I saw a post from a Mac Pro owner claiming Intel lists low 70's C as the max Nehalem (_Core_) temperatures - I don't know where they're getting that info from and it's clearly not correct.
The questions I had (see Marcel's answers below) were a) Where/how is TM getting its "specified upper limit' for CPU core temps and b) Is TM reading a delta from its listed upper limit or an absolute value to determine the reported core temps? (And if a delta, the number reported would depend on the value of the max/limit used.)
There's a note in the Intel 5500 PDF (linked above) that Tcase_Max is a thermal solution design point and in actuality, units will not significantly exceed Tcase_Max_A due to TCC (Thermal Control Circuit) activation. Section 6.1.1 (Thermal Specifications) has more info on thermal profiles and mgmt and notes "Systems that implement fan speed control must be designed to use this data". (The question some of us have is if the OS X boot log warnings on no Turbo P-state info found have any effect on the higher than expected core temps reported w/light loads like itunes playback. Does Apple's fan control/thermal mgmt even monitor core temps? or just the "CPU Temperature Diode" (that Temperature Monitor also monitors) and heatsink temps?) And I still wonder why the 'booster' fans (in the heatsinks) never seem to spin up very much off the default speeds. (On my dual 2.66, even with cores in the 80's C the booster fans are reported to only be a hundred rpm or so higher than idle on my system.)
But IF the 09 Mac Pro thermal mgmt is working properly, the CPU should not exceed the Thermal max limits (TCC should kick in if nothing else) and per the Intel Doc, if any junction temps are over limits, ThermTrip is activated and shuts the CPU down. (And they note to further protect the processor, its core voltage must be removed after ThermTrip is asserted.)
Before this subject came up (concern over reported core temps, etc. last Friday), I always assumed the cores were not running near max as I would have expected the cooling fans to have sped up noticeably (although I suspect apple is keying off other sensors like the CPU case and heatsink temps, not the cores for their fan control) - or I'd see the usual signs of overheating (i.e. kernel panics, system shutdowns, etc.). Granted the Mac Pro's fans are quiet compared to what I'd seen with my air cooled G5 tower (where you could clearly tell when the system was taxed...). But I can't recall ever getting a KP (or shutdown) with this 09 Mac Pro even after many hours straight of heavy use. (Granted I hadn't played a long series of itunes songs with it - and was a bit surprised at 82C core temp after only 3 songs played in OS X... That's higher than I'd like/expected... but it's well within the reported limits. The reason appears to be Turbo mode (overclocked core/light load), as explained below.)
Anyway, moving on I also checked core temps after appx 25 min of encoding/converting video using Handbrake (v0.94 32bit) - the highest CPU core temperatures reported by Temperature Monitor during that was one core at 82°C and another at 79°C. (Other cores ranged from low 60's to 77°C. Note the CPU Temperature Diode and Heatsink temps differ of course. If any 2009 Mac Pro owner has seen the CPU case temp (which I assume is the CPU Diode Temperature) exceed 80C let me know.)
Temperature Monitor Readings after appx 25min of Handbrake Video Encoding:
BTW - I was focusing on Core readings and didn't expand the TM window enough to show all the sensors which include memory dimms, PS, NB heatsink and (SMART) Hard Drives.
[Update (Jan 18th, 2010): The EFI firmware update 1.4 release from Apple didn't affect this as I noted in the news page that day but then I didn't expect it to really. I've heard OS X 10.6.3 however will not have the console/log warnings on boot regarding turbo p-state (notes from a beta tester).
And just as a FYI - after monitoring during several encoding sessions in a row (each 30-35Min) with Handbrake (w/quality setting increased a bit over std presets) on my dual 2.66GHz 09 Mac Pro, I've seen some cores at 90-92C although no KP's or shutdowns yet. And the (CPU heatsink) booster fans ramped up a bit - to appx 1350 rpm.]
Comments from Temperature Monitor's programmer: - I wrote Marcel (author of TM/Hardware Monitor) today (Dec. 2nd) regarding questions I had over TM (regarding the "specified upper limit" for cores and if the core readings were absolutes or deltas from a max value, etc.)
(from Dec 2nd reply, updated Dec 3rd w/more comments)
"(Temperature Monitor lists 95°C as the "Specified Upper Limit" for CPU Cores (ref: above screenshot) when run on the 09 Mac Pro here - where does that figure come from?)
You might be referring to the specified upper limits for Digital Thermal Sensors (DTS) displayed by Temperature Monitor
when the program is running on an Intel Xeon E5520, or similar processors.
All Intel Core i processors have a feature to report their own minimum Thermal Control Circuit (TCC) activation temperature if they are asked to do so. The TCC temperature is the absolute value to which the DT Sensors report negative temperature offsets. This means this value is equivalent to the upper limit for any DTS measuring on that individual processor model.
Temperature Monitor and Hardware Monitor of course make use of that feature to individually determine these limits for each DTS. The value is the upper limit the programs report for the DTS entries. Although Intel won't guarantee that the
TCC activation temperature is perfectly calibrated for each processor on an absolute scale (there is part-to-part variation), the value is guaranteed to be the *minimum* absolute temperature for DTS measurements on each chip. This means actual absolute temperature values could be *higher* than the values reported as DTS readings.
Note that this limit applies to DTS usage (per-core temperature directly on the chip die) only. Other limits, for example the Thermal Design Power (TDP) limit or the maximum junction temperature (Tjmax) of mobile processors, as specified in data sheets of Intel, apply to other locations on chip and are measured based on standardized procedures defined in Intel's Thermal Design manuals. For the Xeon E5520, the published maximum TDP limit is 76°C, for example. (see http://processorfinder.intel.com/details.aspx?sSpec=SLBFD)
(Also can you tell me if the (core readings) TM uses are actuals or delta's from a max? (upper limit/max))
If you are referring to DTS readings, the values will be computed by deltas relative to TCC activation, because this is by definition how the DTS values must be used.
(I saw the notes on TCC info (in 5500 series thermal specs doc I linked earlier/above) but I'm still now clear on how/if Tcase Max (center of the CPU's integrated heat spreader) is monitored. At first I thought maybe the "CPU Temperature Diode" TM monitors might be related to Tcase, but I'm not clear on that. I need to make time to read the entire intel doc.-M)
The TCC activation point is a temperature value individually calibrated for each processor part which defines the threshold when the CPU will begin to take countermeasures in case the chip die is about to overheat.
It corresponds with Tcasemax and Tjmax limits but cannot be directly compared to them. The temperature diode is the closest sensor to estimate the chip case temperature, but it is not guaranteed to reflect the exact sensor location where temperatures mentioned in the spec sheets are being measured. (Tcase per the intel doc is measured at the center of the CPU's integrated heat spreader.)
The TCC activation point reported by the CPU can be equal or less than the true calibrated TCC activation point. The reported value is the absolute offset for temperature values reported by the DTS sensors.
- calibrated TCC: 97.4°C
- TCC reported by CPU: 95°C (shown as upper limit in TM)
- raw delta reported by DTS: -40°C
- DTS core temp assumed by CPU: 95 - 40 = 55°C (shown as DTS core temp in TM)
- actual calibrated core temp: 57.4°C (not disclosed by CPU)
(I think some are confusing (IMHO) Tcase limits with Core readings/limits.)
Yes, and it is even worse: They are mixing up the processor families, comparing apples and oranges. (He's referring to some posts in an apple forum to PC site threads on various CPUs/temp reporting, etc) Although all post-Pentium processors have core sensors, their values must be interpreted differently for each of the following CPU models:
- Intel Core processors
- Intel Core 2 processors
- Intel Desktop Core 2 processors with 45 nm technology
- Intel Mobile Core 2 processors with 45 nm technology
- Intel Core i processors
In the argumentation that the Nehalem processors operated near their upper limits, the absolute temperature values wouldn't play any role.
Let's assume a processor would actually operate 5 centigrades below its TCC limit. This means the DTS would internally report a value of -5. If the program assumed an incorrect max threshold of 125°C, it would report "core temperature is 120°C, upper limit is 125°C". Regarding any cooling questions or doubts, this has the same meaning as saying "core temperature is 90°C, upper limit 95°C".
(Yes, but in the first example where an incorrect max threshold of 125°C was assumed, the reported number (based on delta from the incorrect assumption) would have owners very concerned (i.e. 120°C in that example - despite meaning it was still under the limit by the same delta). Although the reasons why have been explained (turbo mode, etc), it's the typical owner's perception/reaction to seeing a high number under light load (and quickly) that brought this entire subject up (in Apple's forums). This was my reason for asking where/how the specified upper limit was obtained and if a delta from it was used. (Hopefully that explanation makes sense.) You've answered that, thanks.
And maybe I'm not thinking clearly but since TM currently reports values (a temperature) [that has sometimes concerned users as far as the number] would it be better to just display the delta below max instead? (i.e. core 2 is 5°C below max). But then I'm sure users would want to see a temperature (number) instead of a delta.-M)
I now read your reports about your Windows tests. It would be interesting to run iTunes for Windows and to check CPU loads with the Windows System Monitor.
(I used WMP in Vista as I wanted to see how it fared using its native player. Although I would have liked to run iTunes in Vista also, I don't have it installed in Vista (or Win7) on that machine.
(Update/FYI: A Quad 2.93GHz owner did run an iTunes audio playback test in Windows 7 - he said temps were stable at 47C vs 88C seen with same itunes test in OS X. His test summary was added to the reader reports below.)-M)
To verify if Turbo Boost kicks in, you might like to download the official "Intel Turbo Boost Technology Monitor" which is a sidebar gadget for Vista:
(It lists support for Core i7/i7 Extreme and Core i5. (I assume Xeon 5500 and 3500 series also.) Per the PDF release notes for v1.0.3 (Nov 2009) it also supports Win 7 (32/64bit) and includes revision history/known issues.
But I did see higher than the 5550's 2.66GHz clocks already reported in HWinfo32 - see first screenshot below taken shortly after booting Vista (and allowing a minute to settle/connect to wireless network, etc.). I'm not a big fan of windows "Gadgets" (I turned off sidebar gadgets day 1 - and no fan of OS X "Widgets" either), but I might install it later.-M)
Note that Windows uses a different process scheduler and power management than Mac OS X, so the ratios of the different P states and C states per time slice could be different, resulting in different temperature behavior.
Best regards, Marcel"
One thing I did notice with Windows/HWinfo32 - sensor readings (opening HWinfo's sensor panel) are almost instantly read for all sensors. However calling up HWinfo's sensor panel while playing audio would stutter the playback for a second or so. (I understand they're running under different OS's, different code, etc. - that was just an observation.) Here's Marcel's explanation as to why TM sometimes shows a "---" instead of a core's value:
"Mac OS X intentionally doesn't support "processor affinity" which is a feature found in most other operating systems. This means Apple does not allow any process to select on which processor core and by which scheduling strategy a process should run. As a consequence, code which determines the temperature of processor core X will have to wait until the OS happens to let this code coincidentally execute on core X. No program and not even a kernel extension can influence when this will happen.
If total load on the system is low, the power management of Mac OS X will additionally switch off the unneeded cores, powering them down by entering HALT state. You will notice this in Temperature Monitor when at first the hyper-threaded virtual cores (even core numbers), and in a two-package CPU system, an entire processor (lower or upper half core numbers) are shut down. These cores are sleeping, so they won't return temperature readings any longer which is correctly reflected in Temperature Monitor by "---" readings. (In the first TM screenshot above (iTunes audio playback), only Core 4 was shown as "---", although at other times I'd seen more depending on load/usage/conditions.-M)
The behavior of HWinfo32 could even be seen as a bug because this program might interfere with the objects it monitors, indirectly influencing its own measurements. If the program forces all cores to run monitoring code in regular time intervals, it will hereby create additional load, waking all processors and actively modifying the scheduling policy of the system.
When I first ran HWinfo32 (first screenshot below) it shows all cores at 2.92xGHz with this 2.66GHz 5550 CPU. (CPU#0 selected in screenshots, but selecting CPU#1 also showed similar 'turbo' speeds.) See below for updated (Dec 4th) notes from later tests including monitoring both CPU's reported core speeds and temperatures during repeated CB R10 MP rendering tests. But as you mentioned, it's not an apples to apples comparison on many levels. (HWinfo32 summary/monitoring shows only the physical cores (8 in this system) although the processor features list does recognize there are 16 "Logical processors". And running apps like CineBench does report/recognize 16 'virtual' cores.)
iStat Menus 2.0 Observations: (Dec 5th notes)
I used Temperature Monitor for the above OS X tests since that was what those reporting the concern originally used (and I didn't know of any other freeware that reported core temps in OS X), but I know some readers (for quite awhile) have used iStat Menus. After getting a mail from a 2009 Dual 2.66 owner that used it saying he'd never seen over 65°C reported (heavy loads) in iStat Menus, I decided to give it a try. (There's no Core temp reporting in iStat although it has many other features including wattage/amps draw, network stats, etc. but all I focused on today was CPU temperatures.)
Anyway I installed iStat Menus 2.0 (current version) in 10.5.8 (same drive used for above tests - the only difference was I'd installed the Java update on Thursday night), installed the iStat sensor module (single user install) and ran the same iTunes song playback (only app running) test. There's no reporting of Core Temperatures available (clicking on cores shows usage, etc but not its temperature) so I monitored the reported "CPU A" temperature shown. I assumed their CPU temperature reading would be the equivalent of TM's "CPU Temperature Diode". However the iStat menu reported CPU A temperature during the iTunes audio (only) test was much lower than even the CPU Temperature diode readings I'd seen in TM. (Same boot HD/same itunes playlist, no other apps running.)
Then I launched TM (iTunes still playing) to compare readings. iStat menus was reporting CPU A at 60°C - TM however showed the CPU A Heatsink Sensor temperature at 61°C. And the Heatsink temperature of course is much lower than the CPU Temperature Diode reading (73°C at that time) or the cores (as in earlier test, some at 80/81°C). I wrote iSlayer/Bjango to ask what sensor they use for iStat Menus CPU temperature reporting but based on this comparison, I suspect it's the CPU Heatsink. (Update - Here's their reply:)
"Re: iSlayer Support - Question on istat menus sensor used for CPU temperatures
I think you're right that we are reporting the heatsink for that model. We will try and fix it in a future version.
(I asked if they were going to change to using the "CPU temperature diode" or also include core temperatures.-M)
We will try and get the (CPU) diode temp added in a future update.
Cheers, Bjango Support Team"
One other thing I noticed - with iStat Menus 2.0 installed, Temperature Monitor (v4.8) would no longer report my 4 hard drive's temperatures. (Removing iStat Menus restored TM's reporting on them.)
Windows/Bootcamp Tests: (Updated notes on Dec. 4th)
There was also some question about windows support for "turbo" mode (which I think is a hardware function and not OS dependent), so I also took screenshots of that info from the windows HWinfo utility that was recently updated (v3.30 as of late NOV 2009) and supports these CPUs - HWiNFO32. It shows Turbo mode info (speeds) for the 5500 CPU's (2.66GHz) in this Mac Pro. The screenshot below was taken shortly after booting to Vista 64 Ultimate (idle - no other apps running). It shows the (2.66GHz) 5500 cores at 2.92xGHz and lists turbo mode (3066.7MHz) support. (And IIRC, I've seen higher than the 5550's 2.66GHz speed reported by some apps/utils in Windows at times. And CPU core clocks are not the only variable ones, but that and temperature reporting are the focus here.)
[FYI: As of late December there's a later version of HWinfo32 (v3.32) than I used for tests here (v3.30). I will recheck/compare results with v3.32 when I can get some free time.]
HWinfo32 Summary screenshot (after Vista 64bit boot/Idle)
(Screenshot cropped to remove GPU/other info)
The above was taken right after booting/loading Vista and allowing a minute for it to settle, connect to the wireless network, etc. (Idle means no user apps running, of course there's background processes in Vista Ult and NIS2010 is also running, although CPU usage isn't high.) Note the 2.66GHz 5550's cores are reported as running at 2.92xGHz (turbo mode). [If I switched to CPU#1, similar speeds also shown.] This was repeatable (rebooted, checked again. Cores reported at 2.92GHz speeds. Speeds did vary in some later tests (notes below) but apparently I've not pushed it hard enough for long enough yet to have any core reported as running below 2.8GHz.]
Below is a screenshot from HWinfo after appx 15 min of Windows Media (very small window) sample song playback (w/visualizer):
HWinfo32 Sensor Readings after appx 15min of Vista/WMP song playback
BTW - I repeated the above test a 2nd time, this time w/o any visualizer and with WMP minimized. The config was set to open the sensor panel automatically but I had closed it before running WMP and did not open the sensor panel until after 15min of audio playback with WMP minimized had completed. Temperatures were within a degree or two of the above - within what I'd consider normal run/run variation. (And just as a FYI - based on a timing of the sensor window's "CPU Digital Thermal Sensor" (DTS) heading text flicker, the sensors were being read/refreshed appx every 3 seconds.) During a 3rd run (w/similar results) norton's background scan kicked in. (Although no change in core speeds, etc. reported from that.)
I also checked core temps after appx 5 minutes of UT3 play:
HWinfo32 Sensor Readings after exit from appx 5min of UT3 gaming
BTW - I don't think the Mac Pro's fan speeds reported above are accurate (by a large margin - like 4x. I also see insanely (beyond the fan's rating) high fan speeds reported by it when run on a Macbook Pro also. Even when the system is idle/fans clearly not running at high rpm). For instance in OS X my 2009 Mac Pro (dual 2.66GHz) system's booster fans are usually reported as running under 1200rpm even with cores in the 80s. (As mentioned earlier they don't seem to spin up very much from the default speeds.) Core temperatures reported however seem comparable to what I've seen with another utility (RealTemp) in Windows.
CineBench R10 MP Render Tests (all cores used):
I then watched HWinfo32's cores/speed (on main panel) while running CineBench R10's MP rendering test (which uses all CPUs/cores). The reported core speeds varied during that render test (on both CPU's) - from a low of 2.85xGHz to a high of 2.97xGHz on CPU#0 although CPU#1 seemed to never fall below 2.91xGHz. (Observations during four MP render runs.) After the render was complete, within a few seconds the cores settled back to 2.92xGHz. After running the CB MP Render test 10 times, the highest core temp reported as 82°C on one of CPU#0's cores (80°C on other CPU#0 cores - CPU#1 cores were shown as having reached 72-74°C).
And BTW - of course I can't prove any of the above readings are 100% accurate. (I've seen differences in reporting with different utilities over the years. HWinfo and EVGA Precision don't agree exactly on GPU temps either. I can't say there's no bugs in the software, errata in the hardware, etc. But I did find this short exercise interesting.)
Although I didn't see core temps exceed the 80s yet (but only recently installed temperature monitoring software) and haven't had any kernel panics, overheating/thermal shutdowns personally over the last 6 months of use, some OS X users said they have. [Update: As I mentioned in a later update above - after repeated HB encodings I've seen some cores at 90-92C.] At least one apple forum user said he'd seen as high as 96C core temps in some tests with his single 2.93GHz CPU 2009 Mac Pro and the system had shut down (thermal shutdown). (Update: See his tests summary below in the other reports section.)
I later ran the iTunes audio (only) test for about 35 minutes (2.66GHz dual CPU system) and still didn't see core temps any higher than 81/82C (no real rise over what was seen after 15min). But after several repeated long sessions of Handbrake encoding I've seen some cores reported at 90-92C. (I've done that many times over the last 6 months or so - at times 12 hours of a day spent doing that. But HB was the sole app running - no audio playback during that.) I did expect the Booster (CPU heatsink) fans to spin up more than they do - they seem to stay within a hundred rpm or so of default even with cores in the 80's (and CPU diode temp in the mid-70's C). (Update - with some cores at 90-92C, my 2 booster fans were reported as appx 1350 rpm - a couple hundred rpm higher than when the cores were in the 80's.)
And reading about some other user's problems/repairs reminded me that I should get AppleCare on this expensive beast after all - something I normally don't do. (I've only had AppleCare once - an xmas gift for the 2003 G5 tower.) Not that I feel like the CPUs are likely to fail with my usage patterns, but since I plan on keeping it for several years and the insurance might be worthwhile regardless. (Especially considering typical apple repair parts costs.) I've not seen any problems to date (even with many days of 12hrs total of video encoding), but I've not played iTunes (or other core audio) while encoding video but that's something I plan to do in the future just to see its effects on temperatures.
However I do wish Apple would comment on this and the "kernel WARNING: ACPI_SMC_CtrlLoop::initCPUCtrlLoop - turbo enabled but no turbo P-state found" seen during boot in Snow Leopard (10.6.2 used).
And I forgot to mention it here, but some have used SMCfancontrol to manually force higher fan speeds (at the cost of more fan noise). (One reader said he used it to tweak fan speeds a bit to eliminate some fan haromonics also.) I've not done that personally (I generally avoid addons like that) and don't plan to unless I see a thermal shutdown or other problem in the future.
Other 2009 Mac Pro owner replies:
Several other 2009 Mac Pro owners replied to the original (Nov. 27th) post.
[NOTE: Reports below were BEFORE the "Audio Update" for OS X 10.6.2 was released.]
ESI MAYA44e PCIe Audio card Tests: (Dec. 31st, 2009 - BEFORE Audio Update was released.)
You may have seen this already (for anyone following the Apple forum thread (no longer online.) on 2009 Mac Pro CPU temperatures discussion). I don't own one of these cards personally but posting as a FYI:
" Guys, I have a solution... kind of. I had the same symptoms like you guys. With audio playback CPU temps jump over 70 C sometimes, without (audio playback) CPU stays between 30 - 40 C all the time. I tested this with the internal audio and a MOTU Ultralite Firewire interface, same result.
(Not sure what he's using to monitor Temps (Temp. Monitor reports cores as well as CPU Diode and Heatsink temps - iStat menus v2.0 CPU temperature reported is just the Heatsink temp.) 80C+ cores are common after just several minutes of OS X audio playback as many have noted earlier. And a 2.93GHz owner has reported 96C cores/thermal shutdown in some tests.)
So, today I bought the ESI MAYA44e PCIe card because I don't need the MOTU anymore and I was never really satisfied with latency on Firewire interfaces. And yep, symptoms are gone - CPU stays way below 40 C all the time. I'm running just iTunes for an hour now and CPU is at 32 C! Before, I was able to rise the temperature within 2 minutes to over 60 C.
It seems that the Mac Pro doesn't have these problems with PCIe cards.
Greetings from Germany (excuse my bad language)
Mac Pro Quad 2,66 (Early 2009), Mac OS X (10.6.2)"
As of 2009, the ESI MAYA44e 4-in/4-out PCIe Audio Interface card product page (www.esi-audio.com/products/maya44e/) lists compatibility with Mac OS X 10.4.4 and higher (and Windows of course). Bundled with Cubase LE 4 from Steinberg they say. He bought it from Amazon for $149. ($199 list.)
2009 Quad 2.93GHz Owner's Tests (OS X vs Windows/Other Macs):
(added 12/29/2009 from Apple forum thread)
"I just ran a few tests again for confirmation of results. There are tests I have run that I haven't included, as some I need to repeat in order to have confidence in them.
Temperature Monitor 4.8
Windows 7 Ultimate
Real Temp 3.40
Intel Burn Test 2.4
09 Mac Pro (Quad 2.93GHz)
08 Mac Pro (Octo 2.8GHz)
17" Unibody Macbook Pro (2.8GHz)
27" i7 iMac
Test conditions are somewhat variable due to the fact that not all these machines were available for all tests, and they were used at different locations. Different ambient temperatures will cause some variance in the results.
Test 1 - Monitoring CPU heatsink temperature (IHS temp) with iStat Pro or iStat Menu when starting audio playback in OSX (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Temps increase from 32C to 65C within 5 minutes
08 Mac Pro (Octo 2.8GHz) / Temps remain constant at 29C
17" Unibody Macbook Pro (2.8GHz) / Temps remain constant at 49C
27" i7 iMac / Temps remain constant at 42C
Test 2 - Monitoring CPU core temperatures with Temperature Monitor 4.8 when starting audio playback in OSX (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Temps increase from 42C to 88C within 1 minute
Test 3 - Monitoring CPU current requirements with iStat Menus when starting audio playback in OSX
09 Mac Pro (Quad 2.93GHz) / Current increased from 3.88W to 46.97W within 5 seconds
Test 4 - Monitoring CPU core temperatures with Real Temp 3.40 when starting audio playback in WINDOWS 7 (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Temps remained constant at 47C
Test 5 - Conducting a Xbench CPU / Thread test with audio playing and without in OSX
09 Mac Pro (Quad 2.93GHz) / Without Audio = 351-355, With Audio = 273-276
(I've not monitored core usage with XBench but wonder how many cores it uses. Never considered it a great benchmark personally.-M)
Test 6 - Conducting a Geekbench test with audio playing and without in OSX (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Without Audio = 9144, With Audio = 8257
Test 7 - Converting an AVI movie to MP4 format (1:33min at constant 100% quality) in Handbrake with audio playing and without in OSX (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Without Audio Elapsed Time = 17:32min, With Audio = 20:47min
Test 8 - Timing an Intel Burn Test 2.4 loop in WINDOWS 7 with audio playing and without (I played an AAC file in iTunes)
Test settings were: 64bit, High Stress Level, 2048mb, 1 run, 8 threads
09 Mac Pro (Quad 2.93GHz) / Without Audio Elapsed Time = 71.653secs, With Audio = 71.651secs
Test 9 - Measuring CPU Core temps while converting an AVI movie to MP4 format (1:33min at constant 100% quality) in Handbrake with audio playing in OSX (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Temps increased from 42C to 96C within 5 minutes then system suffers from kernel panic and requires shutdown
Test 10 - Measuring CPU Core temps while converting an AVI movie to MP4 format (1:33min at constant 100% quality) in Handbrake with audio playing in WINDOWS 7 (I played an AAC file in iTunes)
09 Mac Pro (Quad 2.93GHz) / Temps increased from 42C to 85C within 5 minutes. System remained stable.
(apple forum username)"
I wrote him to ask if he'd use HWInfo32 in Windows to monitor his core clock speeds to a) see what speeds they're running by default and b) if during any stress tests they ever reduce down from "turbo mode" to the CPU's "rated" speed. (As I noted earlier above, after repeated CineBench R10 MP render tests (using all cores) my dual 2.66GHz's cores still didn't drop to less than 2.8GHz. I didn't monitor them during any long stress tests however - not had time to test that yet. They default to 2.92xGHz (on both CPUs) per HWInfo32.)
(comments below were just some replies to my original post on this in Nov. 27th news page)
(from Nov. 27, 2009 mail)
"I'm just writing in to report that my 2009 Mac Pro 2.66GHz Single Quad Core idles at about 43 - 46'C. When i'm playing a song in iTunes all cores jump up to 73 - 76C. The CPU load is extremely minimal so this doesn't make any sense. Never noticed it until I went to your site today and read that people are having this issue. Count me in to that group.
Model Name: Mac Pro (MacPro4,1)
Processor Name: Quad-Core Intel Xeon
Processor Speed: 2.66 GHz
Number Of Processors: 1
Total Number Of Cores: 4
L2 Cache (per core): 256 KB
L3 Cache: 8 MB
Memory: 8 GB
Processor Interconnect Speed: 4.8 GT/s
CPU usage is about 1.3 - 1.4. I'm using Temperature Monitor to monitor the heat. (I asked if he'd seen the turbo p-state not found console warning message) Yes, I am seeing that error message in my Console:
11/27/09 7:30:40 AM kernel WARNING: ACPI_SMC_CtrlLoop::initCPUCtrlLoop - turbo enabled but no turbo P-state found
I'm using 10.6.2 client. I'm gonna try to reinstall the Combo updater for 10.6.2 and i'll let ya know if anything changes.
I don't think that will help (I reset SMC and reinstalled 10.6.2 from Combo download - no change.)
(from Nov. 27, 2009 mail)
Same issue here. 2009 Quad Mac Pro that is cooking itself whenever audio is played. I have spoken to Apple about this at length and keep getting a "within limits" comment from them which is unacceptable to say the least. The machine gets so hot from simply playing an iTunes song that I have to run the air conditioning at full blast just to keep the room temperature at a reasonable level.
Thanks for highlighting this issue on your site as the only way Apple will actually acknowledge it is if this continues to get attention. The fact that the Mac Pro are not as prevalent as other Apple machines does not help our cause.
(I asked he also check for the "turbo enabled by no turbo P-state found" console warnings)
After a reboot I checked the console and did in fact find 8 instances of that message (I have a quad core 2.93). Interesting!
(from Nov. 27, 2009 mail)
I have not experienced this behavior on my 09 Mac Pro 2x2.26 machine. (Boot rom version MP41.0081.B03.) I do notice the temperature rises but the CPU readings never exceeds 60C and the max temperature I report on any component is 65C (north bridge) while playing music. (Temperature Monitor reporting.)
When I updated to 10.6, I installed from scratch. I've also been running developer releases throughout 10.6.x. I boot exclusively in 64 bit mode and never run in 32 bit mode. These may not be the issue or the reason that I'm not seeing the issue, but it never hurts to see if anyone is running this exact configuration and seeing the issues I'm not.
I do see the following in the console log after boot up: (ditto here w/10.6.x-M)
kernel WARNING: ACPI_SMC_CtrlLoop::initCPUCtrlLoop - turbo enabled but no turbo P-state found
I believe this to be a common occurrence and may or may not have any affect on the system (bug submitted).
I think it's common (one reader didn't see it but asked he check again (Console set to "all messages") after a power-up boot. (He then spotted the messages - one for each "virtual core" as I mentioned earlier. See notes above from Nov. 27th news page post.)