Author Topic: CRITICAL SPEED ISSUE WITH APPLE M SYSTEMS  (Read 13336 times)

2023-10-19, 10:37:11

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
Dear Community and Mac-Users. Please read this if you own a new Apple M-System.

Since around 12 years I work as a professional visualization artist. I have been using CoronaRender for years on Mac computers and with the latest upgrade to Apple M-chipsets I would like to share my experience, as the issue I am having could be very much relevant and critical for every Apple ARM user.

1 Introduction
I work in the field of architectural visualization. In the past 3 years all my renders were based on a setup using my Apple iMac i9 and Corona v9. It has been a great, reliable setup, but I wanted to upgrade. I skipped the first batch of Apple M-systems, but with a fully maxed out Mac Studio M2 Ultra I stepped into the game. So once bought, I updated all my systems to C4D 2023 and Corona V10 to test speed and have a comparison. Using the latest Corona Benchmark and running Cinebench it turned out the new system everywhere speeds up 2.2x compared to my previous setup. So the numbers were as I expected and I got excited.

2 Reality
The reality is very much different and it took me quite a while to notice it. When opening existing scenes, I sensed that the new Mac Ultra somehow did not feel as fast as expected. So I tested both computers with a basic scene, including some models, a few textures and a simple light setup. Of course all systems are updated, run the same software and have the scene saved as project-file with assets. It turns out that the Mac Studio easily gets beaten by a three year old Apple Intel system in real conditions. Depending on many different scenes I tested, the new Mac Studio was not twice as fast, but instead up to 3 times slower than my old Intel iMac ! So in fact it is up to 5 times slower as it should perform. Same is true for parsing times, which are not significantly lower than on the i9. I also noticed the same results using C4D 26 and/or trying Corona V9. Also a friend of mine had similar results on his Apple M1 Ultra. That is beyond imagination, a disaster and I have not found any explanation yet.

3 Issue

So I would like to address this issue to the community. Attached you will find my simple test-scene (DropBox link below) and also a screenshot of the VFB of my Apple iMac i9 and my Mac Studio M2 Ultra. I would be interested if you face the same troubles rendering that scene and hopefully detect something that dramatically slows down the M-chips. If this turns out to be a general issue, using CoronaRender with Apple M-Chipsets does not make any sense at all. It is a failed investment.

Hopefully we can find a solution, as my prediction is, that users who bought a new Apple M system might not even be aware that their computers are not performing to the fullest. You might have paid a lot of money for limited power and not even noticing it.

All best – ASIMO
_

DropBox Link to file : https://www.dropbox.com/scl/fo/rv6cwapzmy7qfvhbf31x4/h?rlkey=elghyfyv4lqzmw4o49ex3haqg&dl=0

2023-10-19, 11:06:04
Reply #1

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12804
  • Marcin
    • View Profile
Thank you for a thorough report. We will look into it and share our findings with you.

Using the latest Corona Benchmark and running Cinebench it turned out the new system everywhere speeds up 2.2x compared to my previous setup. So the numbers were as I expected and I got excited.
Can you confirm that you are getting the expected performance with Corona Benchmark and Cinebench? If so, that would mean there is some issue either with Corona in C4D or with the specific scene.

I understand that you may be reluctant to this, but we would really appreciate it if you double check if the issue is not related to overheating and thermal throttling. We had a similar thread in the past. Please see details here:
https://forum.corona-renderer.com/index.php?topic=38220.msg209024#msg209024

It could be also an interesting idea to check with V-Ray. You could use the trial license for this.



Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2023-10-19, 12:36:00
Reply #2

frv

  • Active Users
  • **
  • Posts: 354
    • View Profile
I have a Mac Studio 20 core 128Gb ram and a 2Tb harddisk. My experience is actually very good in general.

It's important to understand that the M1 or M2 don't have very high numbers in terms of megahertz per CPU. This could result in some operations not being all that faster than a older  i7 or even slower.
Apple had to increase the core count on the Mac Studio m2 otherwise the speed difference compared to the m1 would have been zero. A 28 core i9 with hyperthreading will be at least twice as fast in multicore tasks than anything Apple (no hyperthreading) has to offer.

I do notice significant speed improvements because of much faster hard disk (built in SSD) access and RAM for loading maps and so on,  so parsing is much faster compared to my older systems. Generally modelling and all other stuff aside from rendering speed is very fluent and crash free on my Mac Studio. If you do images that take hours to render really a PC costs much less and is a lot faster. I am thinking of going for a render PC and a Mac laptop in the future. Best of both worlds and C4D and CR runs on both as well.

F
« Last Edit: 2023-10-19, 12:54:22 by frv »

2023-10-19, 14:35:37
Reply #3

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
Thank you for a thorough report. We will look into it and share our findings with you.

Using the latest Corona Benchmark and running Cinebench it turned out the new system everywhere speeds up 2.2x compared to my previous setup. So the numbers were as I expected and I got excited.
Can you confirm that you are getting the expected performance with Corona Benchmark and Cinebench? If so, that would mean there is some issue either with Corona in C4D or with the specific scene.

I understand that you may be reluctant to this, but we would really appreciate it if you double check if the issue is not related to overheating and thermal throttling. We had a similar thread in the past. Please see details here:
https://forum.corona-renderer.com/index.php?topic=38220.msg209024#msg209024

It could be also an interesting idea to check with V-Ray. You could use the trial license for this.

Hello Maru

Thank you for your investigations. I will reply using 4 points to address the issues.

1
I can confirm the performance using Cinebench and the Corona Benchmark. Please see the attachments.

2
As far as I understand, this issue has nothing to do with thermal throttling. It is a desktop Mac. Also while rendering the scene again I monitored the cores. They seem to perform nearly to 100% power all way through. Please see screenshots with timestamps. 4min. 8min. 10 min.

3
The scene all-over is very basic. There are no errors, misbehaviors or problems in loading and rendering the scene on both Macs. The problem is the huge difference in performance. So yes – There might be something happening in the scene that is addressed differently depending using an Intel-chip or an M-chip with CoronaRender.

4
To add something important. I tested both Macs in rendering the known living-room sample-scene. Turns out the performance is perfectly balanced. Both Macs perform exactly the way they should do, having the M2 Ultra with 24 cores 2.2x faster than the Intel computer. So again the question is, what is CoronaRender doing different than in my sample scene and almost every other scene I tested. I also want to point out, that the difference in speed depends on the scenes. While with some files the Mac Ultra performs alright, in other scenes the M-chip is way slower.

Best – A

2023-10-19, 14:42:26
Reply #4

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
I have a Mac Studio 20 core 128Gb ram and a 2Tb harddisk. My experience is actually very good in general.

It's important to understand that the M1 or M2 don't have very high numbers in terms of megahertz per CPU. This could result in some operations not being all that faster than a older  i7 or even slower.
Apple had to increase the core count on the Mac Studio m2 otherwise the speed difference compared to the m1 would have been zero. A 28 core i9 with hyperthreading will be at least twice as fast in multicore tasks than anything Apple (no hyperthreading) has to offer.

I do notice significant speed improvements because of much faster hard disk (built in SSD) access and RAM for loading maps and so on,  so parsing is much faster compared to my older systems. Generally modelling and all other stuff aside from rendering speed is very fluent and crash free on my Mac Studio. If you do images that take hours to render really a PC costs much less and is a lot faster. I am thinking of going for a render PC and a Mac laptop in the future. Best of both worlds and C4D and CR runs on both as well.

F

Hello

Thank you for your response. Of course a 28 core i9 is way faster than the recent M-chips. In my comparison it is a 10 core i9. So something really different. The speed itself is not the issue. I can prove that my M2 Ultra is indeed performing 2.2x faster than my iMac i9 of 2020. In regular files and in the benchmarks.

The question is, why the performance still is scene based. Why is CoronaRender performing up to 5 times faster using an Intel-Chip based on the exact same file. This is not understandable for me.

Best – A

2023-10-19, 19:07:21
Reply #5

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5522
    • View Profile
"the known living-room sample-scene" - is this some very old scene that would have been created in a Corona several versions old? That might point to things that are new, like bucket rendering, 4K cache, etc. so it would be good to know how those "some scenes work as expected, some don't" split out, whether it is older vs. newer scenes. TY for the info!
Tom Grimes | chaos-corona.com
Product Manager | contact us

2023-10-20, 09:31:34
Reply #6

davetwo

  • Active Users
  • **
  • Posts: 302
    • View Profile
I'm probably not adding much here  - but for what it's worth it rendered at 7:52 on my M2 MPB/Ventura/64gb (with a couple of other apps running too).  I'm not sure why the ultra speed is so slow in comparison.

Renders in 5:02 on my old threadripper, which has a worse cinebench score than the MBP on paper - but better performance IRL.

The scene does contaiin all legacy materials, plus native c4d compositing tags, which may or may not have some sort of effect. (I dont see why it should TBH).



« Last Edit: 2023-10-20, 09:48:23 by davetwo »

2023-10-20, 09:50:54
Reply #7

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
I'm probably not adding much here  - but for what it's worth it rendered at 7:52 on my M2 MPB/Ventura/64gb (with a couple of other apps running too).  I'm not sure why the ultra speed is so slow in comparison.

Renders in 5:02 on my old threadripper, which has a worse cinebench score than the MBP on paper - but better performance IRL.

The scene does contaiin all legacy materials, plus native c4d compositing tags, which may or may not have some sort of effect. (I dont see why it should TBH).

Hello

Thank you for testing. Also very interesting, that your M2 MBP already is 2min faster than my M2 Ultra. That also is very strange.

A

2023-10-20, 09:54:42
Reply #8

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
"the known living-room sample-scene" - is this some very old scene that would have been created in a Corona several versions old? That might point to things that are new, like bucket rendering, 4K cache, etc. so it would be good to know how those "some scenes work as expected, some don't" split out, whether it is older vs. newer scenes. TY for the info!

Hello Tom

Yes the scene I am referring to is known, as it was used frequently to compare speed. It is this scene created in 2014.

https://blog.corona-renderer.com/making-of-corona-sample-c4d-interior/

This scene renders perfectly on my Mac Studio, also having the expected result in comparison to my Intel Mac, which is about 2.2x faster.

Thank you for your investigations

Asimo


2023-10-20, 11:51:55
Reply #9

frv

  • Active Users
  • **
  • Posts: 354
    • View Profile
Just tested your scene on my m1 20 core 128gb 2tb system using CR11 (daily build), OS Sonoma and C4D 2024.
See attachment.  11 min 36 seconds.
When I convert all materials to PB it takes slightly longer at 12 min and 3 seconds.

The difference between a 24 core m2 and a 20 core m1 is not worth it to me then.
If I can half render times with a pc for half the money I will no longer bother for a Mac. A dedicated pc with a Mac laptop for general work will be my next set. I have a Mac Studio borrowed for now. Thanks Asimo to bring this under the attention. A new Mac Studio m2 is 6229€ incl. tax. You can get a 24 core i9 PC with a 4090 GPU for about 3800€ incl. tax.
« Last Edit: 2023-10-20, 12:20:02 by frv »

2023-10-20, 11:58:44
Reply #10

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
Just tested your scene on my m1 20 core 128gb 2tb system using CR11 (daily build), OS Sonoma and C4D 2024.
See attachment.  11 min 36 seconds.

Thank you for testing and adding your result

So same issue. Based on the benchmarks and tests with old scenes your M1 system should render this file in no longer than 3min. So your system is 4 times slower on this scene as it should be. It is mysterious.

Best – A

2023-10-20, 12:56:52
Reply #11

Philw

  • Active Users
  • **
  • Posts: 268
    • View Profile
Thanks for the detailed updates.

As you've mentioned here and on the other thead - there is definitely some missing link between the M1/M2 in real life and how it scores on benchmarks. I suspect its something to do with the Intel Embree Apple Silicon implementation still being quite "new".

2023-10-20, 13:05:15
Reply #12

ASIMO

  • Active Users
  • **
  • Posts: 81
    • View Profile
Thanks for the detailed updates.

As you've mentioned here and on the other thead - there is definitely some missing link between the M1/M2 in real life and how it scores on benchmarks. I suspect its something to do with the Intel Embree Apple Silicon implementation still being quite "new".

Yes and No.

You have to understand, that in some scenes the M-Chip performs perfectly and according to the Benchmarks ! So the performance really is there !

It is just that in certain scenes (In my case almost every scene) there are massive performance drops happening, sometimes they are more dramatic, sometimes they are less dramatic. There is something happening, maybe it has to do with Textures, Shaders, Objects, Settings, which causes this problem on M-systems.

Hopefully we have some answers soon.

Best – A

2023-10-20, 13:07:02
Reply #13

Philw

  • Active Users
  • **
  • Posts: 268
    • View Profile
Absolutely. The devs can hopefully listen and investigate in depth.

2023-10-20, 14:25:49
Reply #14

frv

  • Active Users
  • **
  • Posts: 354
    • View Profile
Just copied pasted the scene in to a new file. Result 14:26
A lot of difference...
I don't know but maybe there is something with this file.
Maybe good to check with something more recent before coming to conclusion on the Mac Studio.
I don't know if intel chips have as well big differences depending on scene type. You test scene hardly makes use of render instances.