Author Topic: Mac M1 LOW POWER MODE bug lives on.  (Read 10330 times)

2023-02-20, 06:37:21
Reply #75

YURII

  • Active Users
  • **
  • Posts: 76
    • View Profile
    • YURII SUHOV
Sarcasm here? It will be same result

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?

2023-02-20, 09:06:46
Reply #76

James Vella

  • Active Users
  • **
  • Posts: 534
    • View Profile
Sarcasm here? It will be same result

Not sarcasm, just genuinely curious. It would be a good way to test/compare the throttle theory.

2023-02-20, 12:49:21
Reply #77

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5412
    • View Profile
YURII, as has been explained multiple times, 100% CPU is not the same thing, it all depends on which parts of the CPU software is using. Corona drives the CPU harder, or some parts of it, or makes the CPU require more consistent draw of power, than other software. It's as simple as that - the fact that some high level approximation says "100% CPU" does not mean you are comparing apples to apples.

Corona is known to use the CPU in a much more demanding way - ie taking full advantage of what the CPU can do - than other software, because we are so well optimized to use the CPU to its full advantage for rendering. Again, the evidence is in the fact that a 2 minute render on high power will run faster than a 2 minute render on low power mode, but after 5 minutes suddenly everything drops in speed - which can only be thermal throttling of some part of the motherboard or CPU. This then comes down to the design not being able to keep everything below throttling thresholds when the CPU is exploited so well.

This is not unique to Macs - if a PC had inappropriate cooling on it, it would do the same. If a motherboard had inappropriate cooling in any machine, it would do the same. It is impossible for software to ask a CPU to do more than it was designed to do - but it's entirely possible for a cooling system or motherboard design not to be able to support the CPU doing its maximum for an indefinite period.

So again, there is nothing we can do other than "cripple" Corona to NOT use the CPU to its best advantage - which you can already do yourself, with less threads, or low power mode.

I really don't know how else to explain it, so I guess if you choose to believe otherwise and have some explanation for why high power mode works at high speed for a period of time then slows down, then we will have to leave you to that belief.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-20, 12:54:35
Reply #78

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5412
    • View Profile
"you guys don't even have your own set of M1/M2 Macs to test all these hypothesis in house." - it has been answered already that we have access to the machines to do our own testing in house.

"Doesn't happen with other software" - other software doesn't access the CPU in the same way as we do. e.g. on my PC when denoising kicks in, I can hear my fans increase in speed, because it is known that denoising fully exploits the CPU more than anything else I have on this machine, with no spare clock cycles where the CPU gets any chance to cool down for microseconds. Remember, it's not just using the CPU, it's what parts of the CPU, and how consistently they are used without breaks, gaps, or pauses. Every software is different (but in the high level, approximate / summary, all will still say "100%" even though that represents different access to the CPU).
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-02-20, 12:59:02
Reply #79

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5412
    • View Profile
"crippling it by setting Mac to 'Low Power mode' gets better results than by using your very optimised algorithm, does it makes sense?"

No, for short renders (before heat has time to build up), Low Power will be slower, as expected. After a period of time where the CPU working so hard causes heat to build up, performance will drop as thermal throttling kicks in. Your 2 minute render will be faster in full power mode, in fact every render up until the CPU or motherboard hits the thermal threshold will be faster. For the render time after that, performance will be worse than low power mode, meaning that for long renders you may be getting a better average performance with low power mode  vs. high power mode for a time then thermally throttled performance after that time.

Again, this change in performance after rendering for a period of time is what points to thermal throttling - if it was poor optimization, high power performance would be worse even at the first few minutes of rendering.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-03-01, 09:22:08
Reply #80

lamfadel

  • Active Users
  • **
  • Posts: 16
    • View Profile
I bought m1 and here are the results in the rendering of one scene.
It's very slow...
My iMac 2020 is faster...
The old i7 makes 16 threads and renders the same scene twice as fast!!!
The new M1 Max has 10 threads…

« Last Edit: 2023-03-02, 07:08:41 by lamfadel »

2023-03-02, 12:18:24
Reply #81

Stefan-L

  • Active Users
  • **
  • Posts: 492
    • View Profile
not sure where the data shows the speed, number threads etc not show any about speed.
but overall:

the intel i7 has hyper-threading, so shows 16 threads but they are actually only 8 real ones.

the m1 has different cores, where as far i know not all are made for the same tasks, there are very specialized ones included. some are for have tasks like rendering, some for other tasks, or for light things, video encoding, or to save electricity/battery etc., so maybe i woudl not be even sure if it would be actually correct that all m1 cores are all time at 100% while rendering, i woudl expect only the one suited for render like tasks?

for me in some tests the m1/2 macs feel nice fast at GUI, and 2d graphics, video editing  etc, but not so in any 3d cpu app. Not sure if this is by missing optimization yet or by design of the chips.
« Last Edit: 2023-03-02, 16:55:28 by Stefan-L »

2023-03-02, 14:56:26
Reply #82

Philw

  • Active Users
  • **
  • Posts: 267
    • View Profile
Yep a 16 core iMac is absolutely going to be faster than a 10 core M1. M1 is good but not magic :-(

2023-05-16, 23:09:06
Reply #83

YURII

  • Active Users
  • **
  • Posts: 76
    • View Profile
    • YURII SUHOV
Im rendering now on same M1 Max and I dont see a difference in CPU load while looking at Activity Monitor in both - low and high mode. I think latest Apple updates did something, could be? Can anyone confirm it?

2023-09-19, 13:14:42
Reply #84

habber

  • Users
  • *
  • Posts: 2
    • View Profile
Im rendering now on same M1 Max and I dont see a difference in CPU load while looking at Activity Monitor in both - low and high mode. I think latest Apple updates did something, could be? Can anyone confirm it?

Hi,

I just stumbled upon this post as I discovered this issue by chance. Just been testing it and I am still experiencing this issue.

Another interesting fact I found out today is that render times vary significantly between Corona 10 and Corona 9 in my case (M1 Max 64GB on Ventura 13.5.2) I only recently updated to Corona 10 and was shocked by significantly higher Rays/s total, literally like up 5 to 10 times more rays/s depending (at first I was even thinking that I am miscounting the zeros lol) on the scene but surprisingly at the same time about half the absolute render times - very strange.

Just "downgraded" back to Corona 9 and even though it shows me lower rays/s my render times are split in half. Guess I'll stick with good old Corona 9 :)

The problem with lowpower vs highpower still continues though but in my case the difference is something I can live with. I never really realized because I was so happy with render times in general :)

Cheers!

2023-09-19, 13:24:27
Reply #85

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5412
    • View Profile
Did the scene that was slower on 10 use the Triplanar map?
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-09-20, 11:52:50
Reply #86

habber

  • Users
  • *
  • Posts: 2
    • View Profile
Did the scene that was slower on 10 use the Triplanar map?

Hi Tom,

No, it didn't use any Triplanar map. Only UVW + Cubic mapping throughout the whole scene.

In any case I was mostly astonished about the difference in rays/s compared to total render times in v9 vs. v10