Author Topic: Mac M1 LOW POWER MODE bug lives on.  (Read 11952 times)

2023-02-09, 19:33:37
Reply #60

Chules

  • Active Users
  • **
  • Posts: 32
    • View Profile
I have done many tests and always the Mac Book Pro M1 Max was much slower than the iMac i9. Always the M1 Max rendered at an average of 750-800% and the iMac i9 at 950%. Apparently, the M1 with better ventilation do not have the %CPU drop problem... I have raised the Mac Book Pro from the work surface leaving the lower central space free and cool and the render times have improved... it takes 1 hour with an average of 950% CPU. I'm doing long-term render tests and I'm buying it with iMac i9 and it's starting to improve a lot in performance. A little below the M1 Max. When I carry out more tests I will be able to throw a percentage or comparative times. I hope I can shed some light on this matter because my workflow is mac and corona and I am so comfortable that I would not want to change it. From what I read, the mac studio does not have this performance drop and it seems to be a very powerful processor. I have the problem on an iMac M1 that I don't know what I can do to improve its cooling or CPU performance. Thanks to all users and the corona team. Thank you for bringing this wonderful rendering engine to mac. A greeting from Spain.
« Last Edit: 2023-02-09, 20:05:15 by Chules »

2023-02-10, 16:10:06
Reply #61

rafaz

  • Active Users
  • **
  • Posts: 102
    • View Profile
I have done many tests and always the Mac Book Pro M1 Max was much slower than the iMac i9. Always the M1 Max rendered at an average of 750-800% and the iMac i9 at 950%. Apparently, the M1 with better ventilation do not have the %CPU drop problem... I have raised the Mac Book Pro from the work surface leaving the lower central space free and cool and the render times have improved... it takes 1 hour with an average of 950% CPU. I'm doing long-term render tests and I'm buying it with iMac i9 and it's starting to improve a lot in performance. A little below the M1 Max. When I carry out more tests I will be able to throw a percentage or comparative times. I hope I can shed some light on this matter because my workflow is mac and corona and I am so comfortable that I would not want to change it. From what I read, the mac studio does not have this performance drop and it seems to be a very powerful processor. I have the problem on an iMac M1 that I don't know what I can do to improve its cooling or CPU performance. Thanks to all users and the corona team. Thank you for bringing this wonderful rendering engine to mac. A greeting from Spain.

I have my macbook pro on a notebook stand with coolers on the botton and makes no diference.

I'm starting to build a PC :( and will be using it as a render server. Anyone here use this kind of setup, how's the workflow? I'd like to model on mac and render on the PC

2023-02-10, 16:48:07
Reply #62

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Not sure if you are a long time Corona user or not rafaz, but you would need at least one extra render node for that to be an efficient workflow. If you have been with us from before Corona 8, you may have free render nodes already, but if not you would have to ensure you add one to your subscription. Just as a heads up!

Other than that it should work fine, you can use Team Render to send things over to another machine (or network of machines) for rendering.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-10, 17:49:49
Reply #63

rafaz

  • Active Users
  • **
  • Posts: 102
    • View Profile
Not sure if you are a long time Corona user or not rafaz, but you would need at least one extra render node for that to be an efficient workflow. If you have been with us from before Corona 8, you may have free render nodes already, but if not you would have to ensure you add one to your subscription. Just as a heads up!

Other than that it should work fine, you can use Team Render to send things over to another machine (or network of machines) for rendering.

i've been with corona since the early betas.

This is it right?


2023-02-10, 19:28:30
Reply #64

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Perfect, yes, as a long time user you have your 3 extra free render nodes :)
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-11, 16:19:25
Reply #65

masterzone

  • Active Users
  • **
  • Posts: 157
  • CGI Generalist since 1998
    • View Profile
    • ZuccherodiKanna
I have done many tests and always the Mac Book Pro M1 Max was much slower than the iMac i9. Always the M1 Max rendered at an average of 750-800% and the iMac i9 at 950%. Apparently, the M1 with better ventilation do not have the %CPU drop problem... I have raised the Mac Book Pro from the work surface leaving the lower central space free and cool and the render times have improved... it takes 1 hour with an average of 950% CPU. I'm doing long-term render tests and I'm buying it with iMac i9 and it's starting to improve a lot in performance. A little below the M1 Max. When I carry out more tests I will be able to throw a percentage or comparative times. I hope I can shed some light on this matter because my workflow is mac and corona and I am so comfortable that I would not want to change it. From what I read, the mac studio does not have this performance drop and it seems to be a very powerful processor. I have the problem on an iMac M1 that I don't know what I can do to improve its cooling or CPU performance. Thanks to all users and the corona team. Thank you for bringing this wonderful rendering engine to mac. A greeting from Spain.

This will be my last post about this issue because is very difficult to explain here. More test must be done but not from us but from developers. Try to run a Cinema4D native render scene (not corona or Vray). You will able to reach the 995% of the CPU for hours and also you can place the MacBook on a pillow without any issue. You will find the same Cinebench of an iMac i9 10K generation.
Alex
« Last Edit: 2023-02-11, 16:24:59 by masterzone »
| www.c4dzone.com
| Italian Maxon Training Center
| www.zuccherodikanna.com
| Italian based CGI Studio

2023-02-11, 17:35:27
Reply #66

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12711
  • Marcin
    • View Profile
This will be my last post about this issue because is very difficult to explain here. More test must be done but not from us but from developers. Try to run a Cinema4D native render scene (not corona or Vray). You will able to reach the 995% of the CPU for hours and also you can place the MacBook on a pillow without any issue. You will find the same Cinebench of an iMac i9 10K generation.
Alex

I'm glad that you have finally read the explanations that we wrote multiple times on why the issue occurs with Corona and not with some other software and that finally the issue does not require any further comments.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2023-02-11, 23:27:50
Reply #67

frv

  • Active Users
  • **
  • Posts: 337
    • View Profile
Tx TomG
I used sketchup till about 6 or 7 years ago. I don't know of any architect who still does. It's all Archicad, Vectorworks or Revit, the last one not on a Mac. Architects work with BIM models so all the modelling in Sketchup became redundant.
So for me no Enscape still. It's really only Coronarender for me considering the Mac Ultra.

2023-02-19, 16:13:26
Reply #68

YURII

  • Active Users
  • **
  • Posts: 76
    • View Profile
    • YURII SUHOV
Issue occurs with Corona ONLY BECAUSE it is poorly optimised for Mac silicone! Simple as that! How you can be so blind and try to convince us that it is hardwares fault? That M1 Mac cant handle 100% CPU load for extended period of time on Corona, even though it can with other software? Clearly other guys did something right that you guys did wrong! Its like blaming a tutor for not being able to submit test at 100% while other students did that just fine.

Maybe call Apple and ask them for help? Call Maxon and ask how they were able to optimise their app and native renderers? This - We cant do anything - just kills me!

Im seriously thinking of switching to some other renderer, even though Ive ben with you guys since first ever version.

Do something -

This will be my last post about this issue because is very difficult to explain here. More test must be done but not from us but from developers. Try to run a Cinema4D native render scene (not corona or Vray). You will able to reach the 995% of the CPU for hours and also you can place the MacBook on a pillow without any issue. You will find the same Cinebench of an iMac i9 10K generation.
Alex

I'm glad that you have finally read the explanations that we wrote multiple times on why the issue occurs with Corona and not with some other software and that finally the issue does not require any further comments.

2023-02-19, 17:35:38
Reply #69

James Vella

  • Active Users
  • **
  • Posts: 542
    • View Profile
Issue occurs with Corona ONLY BECAUSE it is poorly optimised for Mac silicone! Simple as that!

I dont work for Corona but didn't you see the graphs and comments in this thread?

The take home points I see is - just because a native render engine 'claims to be using 100%' of the CPU doesn't mean it is. Is there any proof? Can you show how the native render engine is utilizing the CPU to its fullest? Are you sure the native render engine isn't reducing consumption so those % numbers are higher while reducing the actual performance of the CPU to ensure the CPU doesn't throttle?

I see results from the Ultra Studio in another thread which show no reduction due to higher thermal regulation, so if the Mac laptop is experiencing this isn't this a direct enough comparison to throttling? What are you trying to achieve, as TomG has noted if you down throttle the CPU its the same as using low power mode correct?

Here is an example:
I use a 3080RTX GPU in a laptop, I understand due to power/thermal limitations that my 3080 is about half the performance of a desktop 3080. In this situation my preference to be mobile is the limiting factor, not my expectation that it should match the desktop performance of a 3080.

Example 2:
Imagine you have 2 identical computers, First computer is at room temperature and when it hits 90degrees centigrade it throttles to keep the components cool, thus you have the first computer reducing CPU power even though it stays at 100% the entire time = slower render time.

Second computer is kept in room temperature of -20 degrees. No matter how hard you push that CPU its going to stay at 100% the entire time and not throttle because it doesn't ever reach 90 degrees centigrade. Second computer is going to finish the render faster.

This is oversimplified due to how you cool your CPU etc however I think the point is clear.

Also you can check on youtube for yourself search "mac pro throttle" and filter by the last year, you will see plenty of videos with similar issues regardless of 3D.
« Last Edit: 2023-02-19, 20:09:02 by James Vella »

2023-02-19, 20:41:59
Reply #70

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
The main, simple thing to take away here is that a CPU is NOT a uniform single thing - it has different parts that make it up. Different software will call upon those different parts to different amounts, and those different parts may result in different amounts of heat being generated. The example most commonly given is the AVX instruction set.

Not all software will call those different parts in the same way, so even though 2 different software says "100% being used", it may NOT mean the same parts are being used. Corona is very optimized for performance, and when it uses 100% that can indeed cause more heat than other software - but doing anything else would mean Corona would be slower, even before thermal throttling happened.

If Corona was poorly optimized, we'd not see the graphs and user reports where it runs faster FOR A WHILE than in low power mode, THEN runs slower. The only explanation for this is that the design could not dissipate the heat and it built up until the CPU decided to throttle itself. If Corona was poorly optimized, it would always run slower and that speed would be consistent and not change over time.

Thus, there is no way to "optimize" Corona here, only to "cripple" it to avoid so much heat being generated - and you can already do this yourself by setting Corona to use less threads, or by swapping to low power mode.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-19, 20:50:18
Reply #71

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
PS - I should also add there are other components on the motherboard that can thermal throttle too, e.g. when the CPU demands more power to work and parts of the board have to start supplying that extra juice, so overall design has to be good to ensure those can dissipate heat.

Either way, "starts faster in full power mode then gets slower than when in low power mode" is a sure sign of thermal throttling somewhere in the machine. The decision is then whether it's faster to have it run at max capacity for a while and then swap to something slower than low power mode, or just having it all in low power mode from the beginning - short renders may complete before throttling occurs, for instance, so you can just do those in full power; but very long ones may benefit from just starting in low power mode for final overall render times.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-19, 21:02:18
Reply #72

YURII

  • Active Users
  • **
  • Posts: 76
    • View Profile
    • YURII SUHOV
Yes, and as a matter of fact, crippling it by setting Mac to 'Low Power mode' gets better results than by using your very optimised algorithm, does it makes sense?

Im just saying, that it can't be that everyone is doing it wrong and you guys are doing it right, if you take into account scale of corporations like Apple, Maxon, BlackMagic, Adobe etc. You guys don't even have your own set of M1/M2 Macs to test all these hypothesis in house.

So, that is what is not making sense to me and few other people in this thread, who tried to tell you same stuff but in a bit more calmer tone. It just doesn't make sense!

Maybe, we should write to Chaos about it? Maybe they have extra budget for these types of scenarios?

The main, simple thing to take away here is that a CPU is NOT a uniform single thing - it has different parts that make it up. Different software will call upon those different parts to different amounts, and those different parts may result in different amounts of heat being generated. The example most commonly given is the AVX instruction set.

Not all software will call those different parts in the same way, so even though 2 different software says "100% being used", it may NOT mean the same parts are being used. Corona is very optimized for performance, and when it uses 100% that can indeed cause more heat than other software - but doing anything else would mean Corona would be slower, even before thermal throttling happened.

If Corona was poorly optimized, we'd not see the graphs and user reports where it runs faster FOR A WHILE than in low power mode, THEN runs slower. The only explanation for this is that the design could not dissipate the heat and it built up until the CPU decided to throttle itself. If Corona was poorly optimized, it would always run slower and that speed would be consistent and not change over time.

Thus, there is no way to "optimize" Corona here, only to "cripple" it to avoid so much heat being generated - and you can already do this yourself by setting Corona to use less threads, or by swapping to low power mode.

2023-02-19, 21:28:55
Reply #73

YURII

  • Active Users
  • **
  • Posts: 76
    • View Profile
    • YURII SUHOV
So, I did a Cinebench R23 benchmark just now (16", M1 Max, 64RAM), started with Low Power mode, ran it for 10:00:00 and the score was 10834, then, immediately, after this test, I ran it again, this time in High Power mode and **SHOCKER** it scored 12108.

So why is that? Why Corona does it backwards? We want performance the same way it is when running Cinebench - CPU at 99%, Low Power Mode uses less power and scores lower. High Power Mode ramps up the fans and scores higher! I can hear fan ramping up, because it is not letting it to throttle as you guys claim! CPU is at 98% as we speak. Why Corona is not ramping up fans the same way as Cinebench is?

Again, question - why you guys think that everyone else got it wrong (or are faking CPU usage of 100%) and you guys got it right? Numbers speak for themselves!

PS - I should also add there are other components on the motherboard that can thermal throttle too, e.g. when the CPU demands more power to work and parts of the board have to start supplying that extra juice, so overall design has to be good to ensure those can dissipate heat.

Either way, "starts faster in full power mode then gets slower than when in low power mode" is a sure sign of thermal throttling somewhere in the machine. The decision is then whether it's faster to have it run at max capacity for a while and then swap to something slower than low power mode, or just having it all in low power mode from the beginning - short renders may complete before throttling occurs, for instance, so you can just do those in full power; but very long ones may benefit from just starting in low power mode for final overall render times.

2023-02-19, 22:17:43
Reply #74

James Vella

  • Active Users
  • **
  • Posts: 542
    • View Profile
So, I did a Cinebench R23 benchmark just now (16", M1 Max, 64RAM), started with Low Power mode, ran it for 10:00:00 and the score was 10834, then, immediately, after this test, I ran it again, this time in High Power mode and **SHOCKER** it scored 12108.

Interesting! Can you try a 60 minute test and post the results?