Tested: Thermal conditions in the 2018 i9 MacBook Pro dramatically hampering performance

1234568»

Comments

  • Reply 141 of 158
    Liljom said:
    Apple fans don't buy for performance. The brand is the most important. It was obvious before the first test, that the thin body won't be able to handle the heat of an i9. You can buy a real i9 laptop with 32 GB RAM, 6 TB SSD and GTX 1070! for about half price of most expensive MBP 15 config (4TB SSD, no GTX at all). Of course heavier, but you just can not expect a thin machine Wich handles i9. 
    That it's not a laptop, it's a transportable workstation, but not as powerful as a workstation. It's a gaming machine for who can only afford one machine.
    More than 4 Kg to transport, big machine. For what? Video is ridiculous. Detached from the wall does not survive a couple of hours.
    There are no professionals using that, just gamers and youtubers.
  • Reply 142 of 158
    tmaytmay Posts: 3,243member
    KITA said:

    The benchmarks have come under fire from critics, with them saying that there are "breaks" to smooth out the performance. However, that does not appear to be the case.

    While there are pauses in the test, they are done at the end of each test, not in the middle. So, individual tests are performed under full load, with no cool-down period during the test to keep processor speeds low.
    Geekbench 4 uses 2 second pauses are after each workload. Each workload only lasts for, in most cases, ~1 second and there are 50 workloads in total (25 for single core / 25 for multi core).

    Here's an example in Geekbench 4 multicore:

    MacBook Pro 15 (i9-8950HK) - 22,522
    Gigabyte Aero 15X (i7-8750H) -  19,073

    And here's the sustained performance in Cinebench R15 CPU for the Aero 15X:


    That's an average of ~1,100 for the Aero 15X.

    Meanwhile, the MacBook Pro, as per your testing, both the i7 and i9 only scored ~900 sustained.

    Clearly the Geekbench 4 results are completely misleading as even the best case scenario for the i9 MacBook Pro fell short of the i7 Aero 15X's worst run.
    Apple is hardly the only vendor dealing with i9 thermal conditions, and like we said, Premiere performs far better with Nvidia GPU silicon than AMD Radeon gear which explains most of the Dell ripping through the test. 
    Dave Lee used the same Gigabyte Aero 15X (i7 + GTX 1070 MaxQ) discussed above, not the Dell XPS 15 (i9 + GTX 1050Ti MaxQ). Although, I imagine the Dell might have also faired quite well.
    The most obvious solution right now is for Apple alter the peak speed of the processor by adjusting the power that the chip gets. Ironically, slowing the peak speed of the processor may allow it to finish tasks quicker, as it will slow down less to keep the CPU cool.
    On that topic, someone has undervolted a Core i9 Dell XPS 15 with some excellent results. Clearly the Core i9 is not an easy CPU for 4~4.5 lbs laptops to handle, but it might be possible to get some more performance out of it after all.

    Great work on the article! Lots of key points covered. Looking forward to more!





    "Don't forget, CB15 benchmark is based on Cinema 4D R15. Maxon is now on R19, so four gens old. It's probably missing a good number of code improvements vs the R19 software being used today. People complain when my GTX1080 is old, well this is a lot worse and people say nothing"

    "That's like firing a stone from Portugal into the Atlantic and saying wherever it hits is 'average' distance to the US. Optimization is all over the place, no one software piece will show it accurately, which is why real-world tests are important."

    Lies, damned lies, and benchmarks
    cornchip
  • Reply 143 of 158
    tzmmtztzmmtz Posts: 23member
  • Reply 144 of 158
    Our IT department just cancelled the order for the new i9 MacBook Pro's to see how this plays out. Guess I'll have to wait and see what happens with all of this. :-(
  • Reply 145 of 158
    nhtnht Posts: 4,308member
    pascal007 said:
    friedmud said:
    BTW - as a way of comparison...  Here's a quick run of our custom scientific simulation software using 4 cores in my 2017 MBP that with a 3.1GHz i7-7920HQ that can turbo to 4.1 GHz:



    As you can see it looks like it's thermal throttling - but it's actually not.  Those dips in frequency actually correspond to dips in the instruction intensity of our software.  This is software that gets run on the largets computers in the world (literally... later this year we're planning a run that uses all of the new Summit supercomputer at Oak Ridge National Laboratory... which is currently THE fastest computer in the world (according to Top500 anyway))... so it has been incredibly optimized... but even it still has dips and stalls that allow the processor to clock down slightly.

    I would expect something like a rendering program to have even more... it probably has to do plenty of file i/o... at least writing out the frames!  Every time it goes to the filesystem the CPU frequency will drop because the processors don't have enough to do.  This will average out to something less than spectacular looking.
    If it was only the CPU taking a breather, as you say, the task would still be performed faster. Yet, as Lee showed, the task is performed faster (or as fast) using a i7 CPU compared to the i9, so there is a real performance problem when heat-stressed.
    There is overhead costs in using more cores and each core adds less efficiency especially for premiere.  Puget did an analysis in 2015 where you saw premiere working 96% efficient for 4 cores and dropping to 25% efficient after 8 cores for complex timelines and 35% efficient after 4 cores for simple timelines.  Meaning you generate 100% of additional heat for 25%-35% improved mpeg2 encoding speed.  There were similar breakpoints for other tasks like exporting 4K to 1080p h.264.

    Simply exporting 4K H.264 say only 40% efficiency after 5 cores.

    https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-Multi-Core-Performance-698/

    Puget has a lot of interesting articles for those folks making themselves an editing workstation.

    "What is surprising is simply how consistent the results were between the different codecs and resolutions. The amount of time it took in seconds to export each was vastly different, but the speedup per core was very similar. From our data, it appears that there is little to no benefit to having more than four or five cores with a single high-end GPU when exporting to 1080p. In fact, we even saw a drop in performance when we added a second physical CPU when using the RED 4K footage.

    Adding a second GPU helped a bit and made it so that it was beneficial to have roughly six CPU cores instead of just four or five. With dual GPUs, however, we saw three instances where having two physical CPUs was worse than just having one (H.264 4K, CineForm 4K, and RED 4K)."

    https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-2015-Multi-Core-Performance-Update1-806/

    The more recent tests are here:

    https://www.pugetsystems.com/labs/articles/Premiere-Pro-2017-Intel-Core-i7-7700K-i5-7600K-Performance-884/

    For folks that think Intel isn't doing anything:

    Starting with the Core i7 8700K, we saw huge performance gains compared to the Core i7 7700K. A 31% gain when rendering previews and a 36% gain when exporting is a staggering increase in performance. Even warp stabilize was on average 24% faster than the Core i7 7700K! Live playback didn't improve much, but this makes the Core i7 8700K a terrific CPU for those on a budget. There are certainly higher performing CPUs available like the Core i7 7820X, Core i9 7900X, and Core i9 7940X, but considering the price point of the Core i7 8700K this is a very impressive result.

    https://www.pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-1-2-CPU-Performance-Core-i7-8700K-i5-8600K-i3-8350K-1047/

    Testing this sort of stuff in macOS is going to be annoying but the same general behavior should be expected.  Windows allows you to lock tasks to cores but macOS treats them as hints and you have to do it programmatically.

    It strikes me that the new i9 MBP isn't geared toward video editors but those developers who have been whining about running a dozen VMs at the same time.  Typically you want to reserve 2 cores for the host OS and that sure doesn't leave a lot for VMs even when counting logical cores (aka hyper threading).  Most of the time, at least during the testing I've seen, the load across all VMs aren't terrible all at the same time as some machines end up computational bound while others are IO bound.  

    That should allow the i9 MBP to handle more VMs than the i7 without ending up in thermal overload and throttling down.

    As someone who actually travels with a MBP and does coding in the field (usually in a desert in a conex or trailer) I still prefer the current MBP design and wouldn't go with a huge 7-10 lb laptop (like the Dell 7720 or Alienware).  Anytime I need one of those I'm better off FedExing a real workstation (or iMac) in a pelican case and bringing a Honda generator.  

    cgWerks
  • Reply 146 of 158
    cgWerkscgWerks Posts: 1,759member
    Liljom said:
    Apple fans don't buy for performance. The brand is the most important. It was obvious before the first test, that the thin body won't be able to handle the heat of an i9. You can buy a real i9 laptop with 32 GB RAM, 6 TB SSD and GTX 1070! for about half price of most expensive MBP 15 config (4TB SSD, no GTX at all). Of course heavier, but you just can not expect a thin machine Wich handles i9. 
    Well, Apple fans used to buy for performance... maybe that has changed.
    (Though, I'm not so sure about that as a generalization, because while many of those other laptops/desktops have parts like i9, or a 6TB SSD, or 32GB of RAM, they often aren't really 'apples to apples' when you get down into the details. For example, a typical SSD in some PC might not be the same as the faster SSD in a Mac.)

    But, what is most important isn't the brand, but the OS and eco-system. That why some of us are so concerned when we see what we think are cracks or direction changes in that.

    tzmmtz said:
    That was a pretty stupid article, though, even if it turns out to be correct. They did nothing to show the problems... it was just all conjecture and saying Apple is probably more reputable than a YouTube'r. My favorite line was, 'He even told me that some of the tasks his company does “would not have been possible” five years ago.' Um, we're comparing 2017 to 2018 here, not 5 years ago. Heh.

    Again, if accurate.... if Lee gained performance back by putting the laptop in a freezer, then the problem is heat-related, not optimization related. The software almost certainly ALSO needs to be optimized, but that might not matter.

    nht said:
    "What is surprising is simply how consistent the results were between the different codecs and resolutions. The amount of time it took in seconds to export each was vastly different, but the speedup per core was very similar. From our data, it appears that there is little to no benefit to having more than four or five cores with a single high-end GPU when exporting to 1080p. In fact, we even saw a drop in performance when we added a second physical CPU when using the RED 4K footage.

    Adding a second GPU helped a bit and made it so that it was beneficial to have roughly six CPU cores instead of just four or five. With dual GPUs, however, we saw three instances where having two physical CPUs was worse than just having one (H.264 4K, CineForm 4K, and RED 4K)."

    Yeah, performance on multiple cores is quite software dependent, so I'm sure that plays a huge role... but there is still that 'put it in the freezer' issue. So, maybe it's a mix of both.

    nht said:
    It strikes me that the new i9 MBP isn't geared toward video editors but those developers who have been whining about running a dozen VMs at the same time.  Typically you want to reserve 2 cores for the host OS and that sure doesn't leave a lot for VMs even when counting logical cores (aka hyper threading).  Most of the time, at least during the testing I've seen, the load across all VMs aren't terrible all at the same time as some machines end up computational bound while others are IO bound.  
    Yeah, or if you're just running a few higher-demand apps instead of just measuring 1 app. But, the VM thing is big for a lot of people.
  • Reply 147 of 158
    Guys, I have a real theory about what happened here - bear with me.

    Did you notice that even before the 2018 Macbook Pro was released, all the Wintel companies launched laptops with the Core i9.  (All of which had proper thermals).
    I don't think Apple EVER PLANNED to offer this upgrade.  But maybe they felt if they didn't, the Macbook Pro would look really out of date compared to all the competition.  But as you see, Apple wasn't ready with a new design - this is the SAME DESIGN from 2016, and it isn't capable of handling a Core i9.  For all I know, the next big redesign won't even have an Intel chip.

    I speculate that Apple made a very last minute decision to just drop in 6-core processors just so they could "keep up" with existing Wintel laptops and not look bad.  If they were caught off guard, they might have really rushed the process.  Maybe Intel even told them it was just a drop in replacement.

    So that's my theory - we would normally have seen ordinary quad core CPUs if the other vendors like Dell, Gigabyte, etc, hadn't all launched i9 based laptops that were all 4 pounds and had 4K screens, etc....
  • Reply 148 of 158
    cgWerks said:
    DuhSesame said:
    Well, if you want to bring an i9 to maximum performance, you got to have the size of a gaming laptop to begin with.  I don't think many realized that in the first place, that they think "If we just make it thicker, it will be no problem" -- they did not realize how much "thicker" it needs to be to achieve that performance!  And if you do get a little thicker (or you can go all the way to Unibody), sure it will be better, but still throttles significantly, in the end you're not solving problem with it.
    There is no way other than thickness? But yeah, going back to the unibody would be find by me, especially for a high-end version of the 15".

    elpopo64 said:
    Why not, it is just one more choice. If you don't like you take the i7. For the real pro it is ok, when you needs sustained load, you are working on long computation on the desk. If you spend all that amount for the i9 version, you can also add 20-30 bucks for a good cooling pad to put under the macbook.
    A lot of people do it already, mostly in the PC world and most of these pads can be carried around also in case you need. 
    True... now that we know. But, this probably ticks people off who just go to the Apple Store and pick maxed out configs.
    Then this guy will never have problems as it is just the casual buyer who only has the money for the top, but don't need it.

    For the unibody: this model is still unibody in the construction. You may mean to go before the unibody, but still the thickness was just a bit more and (I still have them stored around my office) they are not better in the thermal part. The power was much lower.... And I do not compare with the plastic ones before.

    The real solution is just better ventilation. It is only connected to thickness by the fact that you can increase fans sizes and apertures with a thicker construction. But you can do the same with an external cooling pad. The advantage of the last solution is that you increase the thickness only when you need it and maintain high portability.

    I am going to order an i9 model but I would like to see some tests with a cooling pad or different cooling conditions (like the macbook not just standing flat on a wood or plastic desk), but these days you only found amateurish testers, youtubers (like the guy that put the computer in the freezer, or the one who compares the 2017-2018 by using only 2 cores and concludes the 2017 is faster), no one doing the right testing to understand the extend of the problem.
  • Reply 149 of 158
    teedeeteedee Posts: 10member
    elpopo64 said:
    cgWerks said:
    DuhSesame said:
    Well, if you want to bring an i9 to maximum performance, you got to have the size of a gaming laptop to begin with.  I don't think many realized that in the first place, that they think "If we just make it thicker, it will be no problem" -- they did not realize how much "thicker" it needs to be to achieve that performance!  And if you do get a little thicker (or you can go all the way to Unibody), sure it will be better, but still throttles significantly, in the end you're not solving problem with it.
    There is no way other than thickness? But yeah, going back to the unibody would be find by me, especially for a high-end version of the 15".

    elpopo64 said:
    Why not, it is just one more choice. If you don't like you take the i7. For the real pro it is ok, when you needs sustained load, you are working on long computation on the desk. If you spend all that amount for the i9 version, you can also add 20-30 bucks for a good cooling pad to put under the macbook.
    A lot of people do it already, mostly in the PC world and most of these pads can be carried around also in case you need. 
    True... now that we know. But, this probably ticks people off who just go to the Apple Store and pick maxed out configs.
    Then this guy will never have problems as it is just the casual buyer who only has the money for the top, but don't need it.

    For the unibody: this model is still unibody in the construction. You may mean to go before the unibody, but still the thickness was just a bit more and (I still have them stored around my office) they are not better in the thermal part. The power was much lower.... And I do not compare with the plastic ones before.

    The real solution is just better ventilation. It is only connected to thickness by the fact that you can increase fans sizes and apertures with a thicker construction. But you can do the same with an external cooling pad. The advantage of the last solution is that you increase the thickness only when you need it and maintain high portability.

    I am going to order an i9 model but I would like to see some tests with a cooling pad or different cooling conditions (like the macbook not just standing flat on a wood or plastic desk), but these days you only found amateurish testers, youtubers (like the guy that put the computer in the freezer, or the one who compares the 2017-2018 by using only 2 cores and concludes the 2017 is faster), no one doing the right testing to understand the extend of the problem.

    I’ve got the i9 model and my friend got the i7 higher end model. I did some quick test as I was considering refunding my i9 for an i7. The test isn’t very comprehensive as it’s for my own reference, but it gives a rough idea of i7 vs i9 CPU performance.

    ========

    Cinebench CPU Test:
    i7 Turbo Boost Enabled with force Max Fan: 921
    i7 Turbo Boost Disabled: 835
    i9 Turbo Boost Enabled with force Max Fan: 935
    i9 Turbo Boost Disabled: 945

    On the i9 the thermal throttling is making it difficult to even maintain its base clock speed of 2.9Ghz due to it keep boosting to approx 3.7-4Ghz right after cooling, before severely throttling again after 1-2 seconds. I figured that if I disable Turbo Boost, I can make it maintain at 2.9Ghz as it doesn’t boost and overheat. True enough, it maintained at 2.9Ghz at approx 80 degrees celcius. In fact as seen from the above results, the i9 without turbo boost performs better than with turbo boost enabled.

    ========

    HandBrake 4K Encoding Test:
    Both the MacBooks are throttling to about the same GHz range. The estimated time on both MacBooks are very similar, with the i9 having slightly better estimate time (~1-2mins out of 28mins). I did not time the entire encoding as I didn’t have the patience to wait for 28 minutes. However from my observation after performing this test several times, it shows that the i9 is very slightly faster than the i7 model.

    I also performed this test with turbo boost disabled expecting better estimated time. However the estimated time remained the same...

    ========

    After these quick test, I feel more assured that the i9 does not perform worse than the i7. As a result, I will not be exchanging my i9 to i7. It seems that Apple can improve the performance of i9 by simply updating the firmware to manually lower the max turbo boost speed so that the CPU doesn’t get excessively hot and throttle. 

    The only disappointing thing is that I was expecting the handbrake encoding speed to be much better. My base mid 2012 rMBP with i7 quad core has estimated encoding time of 50-60 mins. Intel’s latest hexacore CPU is only performing 2X better than a 6 year old quad core. I was expecting more...
    M.PaulCezanne
  • Reply 150 of 158
    I want to thank AI and all of you for your detailed and well informed comments.  However, I’m now experiencing analysis paralysis and can’t decide which to buy.

    It seems like the decision is easier for those who were on the fence about needing the i9.  I fall into that category. However, now I can’t decide if I should get the 15 or max out a 13.  Probably thinking too much.

    🤔
  • Reply 151 of 158
    Intel® Power Gadget's Mac download was removed from Intel's website recently. Many think this was in response to the i9 throttling controversy. However, it is now back online and updated to version 3.5.3.
  • Reply 152 of 158
    tmaytmay Posts: 3,243member
    I want to thank AI and all of you for your detailed and well informed comments.  However, I’m now experiencing analysis paralysis and can’t decide which to buy.

    It seems like the decision is easier for those who were on the fence about needing the i9.  I fall into that category. However, now I can’t decide if I should get the 15 or max out a 13.  Probably thinking too much.

    ߤ䦬t;/div>
    Watch this video. If you aren't using Premiere Pro, then consider the i9 over the i7;
     


    John Poole's twitter feed

    https://twitter.com/jfpoole

    geekbench blog

    https://www.geekbench.com/blog/2018/07/macbook-pro-mid-2018-throttling/

    Bottom line.

    Premiere Pro uses both CPU and GPU, which is why it throttles so badly, and compared to FCP or DaVinci Resolve, is an awful implementation by Adobe for i9 MBP.

    Had people waited until there was more real world benchmark's, I doubt there would have been such a fuss.
    edited July 21 cgWerks
  • Reply 153 of 158
    cgWerkscgWerks Posts: 1,759member
    elpopo64 said:
    For the unibody: this model is still unibody in the construction. You may mean to go before the unibody, but still the thickness was just a bit more and (I still have them stored around my office) they are not better in the thermal part. The power was much lower.... And I do not compare with the plastic ones before.

    The real solution is just better ventilation. It is only connected to thickness by the fact that you can increase fans sizes and apertures with a thicker construction. But you can do the same with an external cooling pad. The advantage of the last solution is that you increase the thickness only when you need it and maintain high portability.

    I am going to order an i9 model but I would like to see some tests with a cooling pad or different cooling conditions (like the macbook not just standing flat on a wood or plastic desk), but these days you only found amateurish testers, youtubers (like the guy that put the computer in the freezer, or the one who compares the 2017-2018 by using only 2 cores and concludes the 2017 is faster), no one doing the right testing to understand the extend of the problem.
    Yeah, what I meant is that I was fine with MBP sizes back in the mid-2000s as a pro user. I like the smaller sizes, now, but not if it means giving up too much. Maybe I'm wrong in that there wasn't more cooling going on with those machines... but I would think there would have been.

    Yes, I've used stands to lift the laptop off surfaces and have tried blowing fans over/under them. I even built a stand with several magnetic bearing fans (super-quiet) blowing under it constantly.

    I'm pretty sure the YouTuber knows it's bad to run it in the freezer, but he did that to prove it was thermal throttling.
  • Reply 154 of 158
    It’s all due to the new silicone membrane in the keyboards.
      :D
  • Reply 155 of 158
    tmay said:
    I want to thank AI and all of you for your detailed and well informed comments.  However, I’m now experiencing analysis paralysis and can’t decide which to buy.

    It seems like the decision is easier for those who were on the fence about needing the i9.  I fall into that category. However, now I can’t decide if I should get the 15 or max out a 13.  Probably thinking too much.

    ߤ䦬t;/div>
    Watch this video. If you aren't using Premiere Pro, then consider the i9 over the i7;
     


    John Poole's twitter feed

    https://twitter.com/jfpoole

    geekbench blog

    https://www.geekbench.com/blog/2018/07/macbook-pro-mid-2018-throttling/

    Bottom line.

    Premiere Pro uses both CPU and GPU, which is why it throttles so badly, and compared to FCP or DaVinci Resolve, is an awful implementation by Adobe for i9 MBP.

    Had people waited until there was more real world benchmark's, I doubt there would have been such a fuss.
    Thanks for the link. Jonathan’s point actually supports my thinking.  All this testing is mainly quantifying what we already know: laptops have thermal limits and the i9 is (probably) almost always faster than the i7 - maybe only slightly sometimes and maybe it depends on the GPU in use, but over time the differences might be substantial enough to justify the relatively small extra investment in the i9.  Maybe not. Heavy pro users take your pick, but this “controversy” makes it easier for those of us who didn’t really need the i9 to begin with, but might have considered it if it crushed the i7 convincingly in every way possible.

    Maybe this is a question for a 13” discussion, but as an occasional pro app user, my decision now becomes base 15 or maxed 13. Yes, the obvious difference is screen size, but the 13 really is more portable and easier to write with.  There are technical differences too with the hardware - just not sure how much they impact real-world performance because I haven’t seen anyone putting the maxed 13 through the same battery of tests.

  • Reply 156 of 158
    ikirikir Posts: 71member
    It is a non-issue, i9 gets hot it is normal, Apple will release an uptake firmware which will fans run faster/before and probably most of the issue will be gone.
  • Reply 157 of 158
    ikirikir Posts: 71member
    emoeller said:
    We now have more than one data point, and this is very disappointing news.    My new MBP has the i9 and I will need to decide if I need to return and wait until next year when Intel will hopefully have their laptop chips out (and Apple utilizes the new DRAM spec).  For me $5k+ is a lot of money, but more importantly I needed that extra computational capability for field 3d renderings of block models.
    I cancelled my order...
    No need to panic, the machine is incredible fast anyway. Apple will update firmware for faster/sooner fans. Or just order the i7 6-core version which is very powerful.
  • Reply 158 of 158
    DHFDHF Posts: 1unconfirmed, member
    I returned mine even after the software update; it did not resolve the issue. With VMWare, I could not get the device to stay below 85c under decent load while running a few other apps. I could see the slowing of performance as I routinely run other OSs in VMs which kept the CPU load up under normal use. If the VM performed any load for longer than 30 secs, I could visually see the throttling impact the applications. After all, the 32Gig of ram and CPU is meant to give us the horsepower to do this. I also struggled to get anything above 60% utilization and when the power draw was high, it spiked and held 95-100c which was worse than my i7. I appreciate Lee drawing attention to this. It is time Apple made pro machines that provided the performance the components are capable of..
    edited August 3 cgWerks
Sign In or Register to comment.