Author Topic: Corona for Apple Silicon M1?  (Read 36768 times)

2022-02-24, 04:30:28
Reply #60

BigAl3D

  • Active Users
  • **
  • Posts: 879
    • View Profile
That's fast! So that's not a Threadripper, but a lesser CPU? Can't imagine what the 64-core Threadripper does to that scene.

2022-02-24, 10:17:44
Reply #61

ficdogg

  • Active Users
  • **
  • Posts: 112
    • View Profile
Not a threadripper, just a 5th gen Ryzen with 12c/24t, but it has some really high boost on the cores.

2022-02-24, 11:18:34
Reply #62

ianosss

  • Active Users
  • **
  • Posts: 83
    • View Profile
That's fast! So that's not a Threadripper, but a lesser CPU? Can't imagine what the 64-core Threadripper does to that scene.

I just tried that scene on my threadripper 3960X 24-Core  :) ...  1min 19secs. I think the 64-core is roughly x3 times faster than that even!. :)

2022-02-24, 15:29:25
Reply #63

BigAl3D

  • Active Users
  • **
  • Posts: 879
    • View Profile
@ianoss

FIRE!

2022-02-25, 02:20:40
Reply #64

jojorender

  • Active Users
  • **
  • Posts: 241
    • View Profile
I ran the scene on a 2 year old 10c/20t hack - I9 10850K with a gentle 4.7G overclock.
3:35 min - not on fire but might be the closest to a M1max?

I still think a Corona scene could be more useful.
For example, does M1 native in Corona scale the same as “M1 native” in physical renderer?
Do updates to the corona core, shaders (C-bitmap), etc make the engine faster?
How well does TR scale in different configs.
This would be more than just a pure “machine power” comparison, but would mean people need to be clear about how they tested, including screenshots of VFB, cor version, Mac specs, etc.
This could potentially get messy and maybe not as helpful as I think...
Tried to 1-click convert the scene to corona - no go, unless you like black grapes.

2022-02-25, 09:49:16
Reply #65

Philw

  • Active Users
  • **
  • Posts: 267
    • View Profile
3:07 on M1 Max 64GB

2022-02-25, 13:57:16
Reply #66

jojorender

  • Active Users
  • **
  • Posts: 241
    • View Profile
Thanks Philw
I assume this is M1 native?
Shows us how far off the corona benchmark is regarding M1 (rosetta)
Ran the bench with same I9-10850K

2022-02-25, 14:39:25
Reply #67

Philw

  • Active Users
  • **
  • Posts: 267
    • View Profile
Yep R25 and latest 8 Daily build

2022-02-25, 17:05:30
Reply #68

BigAl3D

  • Active Users
  • **
  • Posts: 879
    • View Profile
@Philw's render time of 3:07 is good, but barely faster than my 2017 iMac Pro. Granted, this thing cost a lot back then. Maybe $,8000. Can't remember.

The M1 is way cheaper and the first generation so I'm hoping the new Macs coming up this year and next will show a big improvement in rendering power. I'm waiting to pull the trigger, but it has to be a "Wow! That's much faster!" moment.

Maybe the Devs or one of us could put together a sample scene. Something real-world like a archvis interior, exterior, a car scene and maybe a motion graphics type scene. I do like that Blender provides nice scenes available to all. The benchmark tests just don't translate into the real world IMHO, but if I see a scene that could be a job I might do and I see that render time to compare, that means more to me.

2022-02-25, 17:12:21
Reply #69

Philw

  • Active Users
  • **
  • Posts: 267
    • View Profile
Yep - the "Almost" Mac Pro that is rumoured with the M1 SoCs stitched together will be the only way they achieve it right now. Minor speed bumps in the cores won't make a huge difference. First iterations of M1 were always going to be about power consumption rather than being uber-powerful.
« Last Edit: 2022-02-26, 07:54:06 by Philw »

2022-02-25, 20:31:09
Reply #70

jojorender

  • Active Users
  • **
  • Posts: 241
    • View Profile
@Philw's render time of 3:07 is good, but barely faster than my 2017 iMac Pro. Granted, this thing cost a lot back then. Maybe $,8000. Can't remember.
Let’s not forget we are talking about a laptop here. I still think it’s pretty impressive.
I’m also waiting for that Mac Pro (mini) with several SOC’s glued together, and knowing how well this scales from a M1max or how it compares to a threadripper in corona is useful info.
All my intel macs are currently heating my workspace, which I can also clearly see on my electric bill. So, yes please give me something less power hungry.
Just for reference, that i9-10850k/64GB was a $1k build (no gpu - only render node)

There is this very old free scene on the blog https://blog.corona-renderer.com/corona-c4d-a1-1/ that might be useful.
It popped up a missing plugin warning but strangely didn’t ask to convert any corona assets to new core.
- delete vray bridge and GI from render settings
- changed the pass limit 9999 to 5% noise
- turn off bloom and glare in corona/ camera tab!!!
Could this be good scene for testing?

2022-02-25, 22:17:50
Reply #71

BigAl3D

  • Active Users
  • **
  • Posts: 879
    • View Profile
Ha. That was one of the first scenes I downloaded to test with Corona Alpha. Since I render lots of cars, I modified it just a little. When I showed people my render and they asked how did they get that car inside an apartment? I knew Corona might be a good choice for me, plus Octane and Redshift didn't run on Macs unless you had an e-GPU setup.

Here's my test run. I actually saved 4 seconds after I closed After Effects, Media Encoder, Photoshop and Firefox. This iMac Pro holds up pretty well I think.

As far as a test scene, it could work. It's freely available for anyone. The only concern, is it challenging enough for a real test? I mean we both got pretty fast speeds and that 64-core Threadripper will demolish those times. So it's more difficult to get a feel if the times are 3 seconds for us one day, and 1.5 seconds for the Threadripper. Yes, it's 50% faster, but do I care about 1.5 seconds? See what I mean?

2022-02-26, 19:06:58
Reply #72

jojorender

  • Active Users
  • **
  • Posts: 241
    • View Profile
This scene was also my first contact with Corona and blew my mind back then…
“The only concern, is it challenging enough for a real test?”
I know exactly what you mean and of course you don't care about 1.5 sec, but if 50% faster means 2 hrs instead of 4 I'm sure you do care.
I ran the scene again with a 2% noise level and it rendered in 16:22. That might be a bit long for the casual tester.
If Philw is up for it and wants to test this on his M1 and Threadripper and establish a noise level  on the M1 around the 8 min mark… Maybe that’s a good compromise?
Unfortunately this scene doesn’t render in TR as is, need to remove
“WhiteGlass” > vray comp tag.
Would love to see how fast M1 + ripper renders this scene to 2% noise

I mostly render 6-8K stills to 3% noise and 4-6hrs is "normal" for me. If I can get this down 50% for a reasonable price, I'd be more than happy.

2022-02-26, 20:50:40
Reply #73

Philw

  • Active Users
  • **
  • Posts: 267
    • View Profile
I’ll give that a go. No problem.

2022-02-27, 04:10:59
Reply #74

BigAl3D

  • Active Users
  • **
  • Posts: 879
    • View Profile
Don't worry about Team Render. That brings with it too many other variables to be useful as a "how good is my CPU at rendering a real job?" question.