Heavy workload on multi-core systems can lead to less than expected performance

Discussion in 'Mac Pro' started by bxs, Apr 14, 2014.

  1. bxs macrumors 6502

    Joined:
    Oct 20, 2007
    #1
    This is a phenomenon I've been curious about and my intuition has always told me this should be expected. Today I did a quick test to see if my tuition had any merit. This has been brought about because some people have been saying the newer laptops such as the MacBookAir can out perform some older MacPros.

    My test was done on my MBP8,3 (17-inch late 2011). It's rated at 2.5 GHz and has an Intel Core i7 processor with 4 cores.

    The question being explored here is can this MBP8,3 maintain at least its 2.5 GHz when placed under a heavy workload. The processor is sensitive to heat and its hardware architecture is designed to keep it within a certain power/heat envelop.

    So I started the Intel Power Gadget with just a light load and slowly increased the workload to cause all 8 core threads to be in operation.

    You will note that as this heavy workload comes on board the processor hardware will step in to reduce the core clock frequency below the rated 2.5 GHz.

    This shows that a laptop such as the MBP8,3 will not be able to achieve a constant 2.5 GHz of compute power when placed under a heavy workload of any reasonable duration.

    [​IMG]

    Now for a real workstation such as the MacPro (old and new ones) this should not be such an issue and will run a similar test later on using a new MacPro6,1 12-core with a stated processor clock rate of 2.7 GHz.

    As soon as the 8 cores in my MBP8,3 start operating flat out the two fans very quickly ramp up to their max RPM of 5500. They stay at this RPM so long as the cores are running flat out and only reduce their RPM as the cores start idling down. Thus, max cooling is employed for the processors when they are under constant heavy use. My test shows the MBP8,3's cooling capacity allows my processor to maintain around 2.0 GHz when under extreme workloads.

    Thus from a performance standpoint when deciding just how much work the MBP8,3 can perform when dealing with a heavy workload I should be using 2.0 GHz as a point of reference for getting work done, and not the rated 2.5 GHz.

    It will be of interest to me to see how this type of test pans out on my MP6,1 12core later today.

    ...later

    The test that represents the screen shots posted above was for a heavy workload running for maybe 10 minutes. I wanted to show in the Intel Power Gadget's display what the GHz were for a light load and then bringing onboard a very heavy workload. I allowed the heavy workload to run for 30 minutes after and the result was as shown in my OP. The GHz simply was being maintained at below the rated 2.5 GHz for the processor.

    I have now been able to run a similar test on my MP6,1 12-core with a 2.7 GHz processor and 64GB RAM. Here are the results with a light workload transitioning to a very heavy workload. This transition is clearly seen in the top graph showing the MP6,1's power consumption. Note that the TDP for the MP6,1 is 130watts.

    Under the light load the power consumption was level at around 40 watts, fan spinning at 790 RPM, and as the heavy workload was brought on this power consumption stepped up quickly to around 100 watts and the fan speed steadily rose up to around 1000 RPM.

    Note that as the power transitioned from 40 watts to 100 watts the Processor cores stayed reasonably level at around 3.0 GHz and just above the processor's rated 2.7 GHz.

    I left the heavy workload running for 30 minutes. The Processor kept chugging along at an average of 3 GHz with an occasional dip to 2.85 GHz. The fan RPM stayed constant at around 1000 RPM. All 12 cores were running flat out at 100%. Processor temps held at around 160ºF. I also had heavy i/o active at around 320 MBytes/sec to a RAID-0 LaCie 2big Thunderbolt-1 device.

    This test and the results are a testament to Apple's excellent design for this MP6,1 and tells me it's a workstation system that can run a heavy workload all day long, all week long and maybe all year long without taking a breath.

    No Apple laptop can do this type of computing workload without reducing its CPU operations to stay within its TDP and will in fact run at a performance level less that its stated Processor GHz.

    I am pleased to see the MP6,1 12core processor rated at 2.7 GHz can in fact run all day long (I assume this all day long even though my test ran for 30 minutes) at around 3.0 GHz and not some lessor value below 2.7 GHz.

    The MPs are real work horses for sure and can run all day long doing heavy work, and hopefully you will agree that the laptops simply cannot match this type of sustained performance.

    ...Oh, and BTW the heavy workload I ran on the MBP8,3 and the MP6,1 were very very similar... as I used the same applications in each testing scenario.

    [​IMG]

    Here's a performance set of data as the heavy workload was brought on... the Processor temp and fan speed were starting to ramp up...

    [​IMG]

    ...and here is what happened after the test completed with the heavy workload gone and returning to a light load.

    [​IMG]

    Here's a test (no i/o was involved -- purely CPU) showing clearly how turbo boost kicks in when a single core is heavily used on my MBP8,3 running with a stated 2.5GHz Processor that has a TDP of 45watts.

    I started the single core activity and then terminated it after a short while so that the graph showed how the boost kicks up the GHz.

    Note that the single core does not reach it highest turbo boost value which should be 3.6 GHz. It instead goes up to 3.3 GHz only..

    [​IMG]

    [​IMG]

    I also ran this same single core test for around 30 minutes and the MBP8,3's two fans spun at 5500 RPM constantly and were able to keep the processor temperature at around 92ºC, thus allowing the turbo boost for the single core to be maintained. The power wattage was averaged 30 watts with an occasional spike to 45 watts.

    [​IMG]

    I was never able to get the single core turbo boost to kick the clock rate up to 3.6 GHz, even with every thing else terminated. I simply suspect some processes were demanding some CPU attention.

    I ran another test for well over an hour with all 8 threads being used 100% CPU. The core rates were reduced to an average of 2.2 GHz, the power held at around 30 watts, two fans constantly running at 5500 RPM and Processor temperature steady at 90ºC. This tells me that if I want to run heavy CPU intensive workload on my MBP8,3 I should reckon on not more than 2.2 GHz for the cores.

    ========

    Also, here's some additional information on TurboBoost.
    http://www.intel.com/content/www/us/en/architecture-and-technology/turbo-boost/turbo-boost-technology.html
    which does have the statement:

    Intel® Turbo Boost Technology 2.01 automatically allows processor cores to run faster than the rated operating frequency if they’re operating below power, current, and temperature specification limits.

    Intel Turbo Boost Technology 2.0 is activated when the Operating System (OS) requests a frequency higher than the rated frequency of the processor. Whether the processor enters into and the amount of time the processor spends in the Intel Turbo Boost Technology 2.0 state depends on the workload and operating environment.

    Maximum turbo frequency indicates the highest possible frequency achievable when conditions allow the processor to enter turbo mode. Intel Turbo Boost Technology frequency varies depending on workload, hardware, software and overall system configuration.

    Due to varying power characteristics, some parts with Intel Turbo Boost Technology 2.0 may not achieve maximum turbo frequencies when running heavy workloads and using multiple cores concurrently.
     
  2. DJenkins macrumors 6502

    DJenkins

    Joined:
    Apr 22, 2012
    Location:
    Sydney, Australia
  3. AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #3
    Good analysis and reporting.

    I doubt, though, that anyone would be surprised that MacBooks do thermal throttling. They have a well-earned reputation for running hot.
     
  4. Riwam macrumors 6502a

    Riwam

    Joined:
    Jan 7, 2014
    Location:
    Basel, Switzerland
    #4
    Very interesting

    ......................
    Thank you!
     
  5. ssls6 macrumors 6502a

    ssls6

    Joined:
    Feb 7, 2013
    #5
    Nice test Barry. It would be interesting to see if power throttling occurs as well. Basically, if the two GPU's were loaded down would the system slow the CPU to balance to the total power to be below the 450W limit. My guess is the GPUs would only be impacted if any but that's just a guess.
     
  6. Sym0 macrumors 6502

    Sym0

    Joined:
    Jun 6, 2013
    #6
    Heavy workload on multi-core systems can lead to less than expected performance

    Seems obvious conclusion to me. However intel design those chips not Apple and I regularly run my 15rmbp flat chat with 8-threads and it maintains 3.2-3.4gHz and peaks at 3.7 when some of the threads drop off. It can do that for 14hrs without drama but that is in windows 8.1.

    I watched the temps and when single thread stuff is running the cores a cycled as it reaches 105c and cools to 80-90 while the other three are cycled.
     
  7. bxs thread starter macrumors 6502

    Joined:
    Oct 20, 2007
    #7
    When you run flat out for 14 hours I would assume the fan(s) are running at full speed to keep temps below the threshold. That of course will wear the fan(s) out that much quicker than typical use of a laptop.

    I can run my MBP8,3 (2.5 GHz 17-inch late 2011) flat out but both fans HAVE to run at 5500 RPM constantly to keep the processor temp just below 100ºC. That is way hot IMO, and if continued day after day the MBP will burn itself out eventually.
     
  8. flowrider macrumors 601

    flowrider

    Joined:
    Nov 23, 2012
    #8
    I have a stupid question. Why are we talking about this on the Mac Pro forum. Why didn't the OP post his findings over the the MBP Forum:confused: The Mac Pro is designed to be used as a desktop for Power Users and has cooling and components designed for that purpose. The MBP is not and does not.

    Lou
     
  9. bxs thread starter macrumors 6502

    Joined:
    Oct 20, 2007
    #9
    Because it's a comparison between laptop and desktop.... so I suggest to you it can be in either forum... It was my choice to post here... sorry if you disagree with this.
     
  10. tomvos macrumors 6502

    tomvos

    Joined:
    Jul 7, 2005
    Location:
    In the Nexus.
    #10
    Interesting experiments you did there. I’m curious—what does happen on the nMP once the thermal stress is increased? Because the cooling system is designed to cope with the energy output from one CPU and two GPUs.

    Can you rerun the test on the nMP with additional load on one GPU / on both GPUs? This might push the system against the bounds of its thermal envelope. I would like to know if the cooling system limits the nMP in the same way as the MBP—however the limit likely be much higher (and perhaps only relevant for benchmarks but not real life usage scenarios).
     
  11. bxs thread starter macrumors 6502

    Joined:
    Oct 20, 2007
    #11
    I've only run 24 CPU intensive threads on the MP6,1 12core with dual D700s. When I did the fan speed never wavered from running at 790 RPM and I recall the power was at a steady 100watts. The average core GHz was at around 2.95 GHz or maybe a bit lower. I was impressed with this as the stated processor GHz for this 12core is 2.7 GHz.

    The Intel stepping for this processor is 3/3/3/3/3/3/3/4/5/6/7/8

    Meaning, with one core active the GHz can be boosted by 8x 100 MHz. So this would technical push/boost the active core to run at 2.7 + 0.8 or some 3.5 GHz. Mind you, it's always a challenge to actually hit this boosted value in the real world.

    Another example for comparing my test result running 24 CPU intensive threads that gave around ~2.9 GHz would be using the 'stepping' value of 3x 100 MHz. Thus 2.7 + 0.3 is 3.0 GHz. So my observed ~2.9 GHz is pretty darn close to the max possible.
     
  12. Sym0 macrumors 6502

    Sym0

    Joined:
    Jun 6, 2013
    #12
    Well i have done over 400 of these processes and she is good as the day it was taken out of the box. And we have two year warranty here so if it does I'll get a new one or have it repaired.

    Besides my business pays for these toys so if it dies it apart from repairing I'd simply get another one. But I can tell you it's not going to die, as much as dooms day predictions like to think they will, they won't unless there is manufacturing fault.

    I have cheap PCs fro 7 years ago that still are fully functioning, including the cheap CD drives and 5400rpm HDD, they have been doing work duties all that time without a hiccup. I have no doubt this fine machine will fail to better that.
     
  13. pertusis1 macrumors 6502

    Joined:
    Jul 25, 2010
    Location:
    Texas
    #13
    This is a great writeup. Without gathering all of the data that you did, I have empirically noticed the same thing. I had the same issue with a MacMini. The benchmark scores are great, but during stress testing, it did not fare so well.

    If you really want to go overboard, I have another experiment that you could run. I noticed last summer that my mac mini's performance seemed to suffer when I let my house's temperature rise to 78 F to save money on air conditioning. During colder weather (we let our house go down to about 67 F), it seems to improve. I wonder how ambient temperature affects the performance of the laptops and minis... Part of me thinks it must have been my imagination, but who knows. :eek:
     
  14. peabo macrumors regular

    Joined:
    Feb 14, 2008
    #14
    I've run a stress test on the same specced system using both the CPU and GPUs at max capacity. It maintained 100W and between 2.8 and 3Ghz for about 20 minutes. When the CPU temp hit around 98-99 degrees, the wattage fluctuated from 90-100 rapidly to keep the temperature down just a fraction more and the speed of the CPU dropped to between 2.7 and 2.9Ghz, but it kept going in this way until I stopped the test.

    I've never seen the system go over 100W at any point as that seems to be some kind of soft cap below the 140W limit.

    But I've never seen any kind of sharp drop in performance and it's actually VERY hard to force the machine down to its base clock of 2.7GHz.
     
  15. bxs thread starter macrumors 6502

    Joined:
    Oct 20, 2007
    #15
    I agree. The laptop is designed primarily for dealing with short bursts of heavy workloads and even then are likely to not provided workload throughput expected from reading the hardware specifications. Whereas, the desktop hardware IS designed for sustained workloads and can in fact provide workload throughput beyond the hardware minimum stated specifications.
     
  16. Macsonic macrumors 65816

    Macsonic

    Joined:
    Sep 6, 2009
    Location:
    Earth
    #16
    True. Laptops were not meant for long hours of heavy usage. Years ago I was using a Macbook Pro and it ran very hot. The feedback I got from a Mac technician is they get more laptops for repairs, usually busted logicboard likely due to intense heat from prolonged usage. They seldom get Mac Pros for repairs.
     

Share This Page