Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

vel0city

macrumors 6502
Original poster
Dec 23, 2017
347
510
This is a thread for discussing everything about 3D rendering on Apple Silicon, M1 and beyond. Whether you're using a dedicated CPU renderer like Keyshot, a plugin like Redshift, Cycles, Arnold or Octane for C4D, Maya or Houdini, or even if you're rendering inside ZBrush with its native renderer, all 3D and related subjects are welcome in this thread. I'm sure we're all very eager and interested to read about performance, thermals and stability of our favourite engines on the new Macs.

So far, Jules Urbach of OTOY (Octane render) has announced that an M1 version of Octane is in development and that is it "FAST!"

If anyone has news of their favourite plugins for C4D being ported to Apple Silicon or any other 3D news of note, please feel free to post it up.
 

hifimac

macrumors member
Mar 28, 2013
61
40
I grabbed a M1 16GB 1TB MacBook Pro yesterday. In normal tasks (opening windows, switching apps, browsing the web) it is crazy fast. Like a different computer all together fast. I haven't been able to put it through it's paces yet on any creative apps.

Here's some quick take aways I've learned from initial use and digging around the web:
  • eGPU's are not supported on AS Macs. Not sure if this is a temporary limitation or a full stop. Drivers for AMD chips are not present at all. This is disappointing.
  • Cinema 4D is natively supported as is FCPX, Resolve and a few other creative apps.
  • I installed Redshift and Corona and neither plug-in loads on the MacBook. Redshift said they are working on support for AS. Seems like Corona is having problems on Intel Macs running Big Sur as well. They said they are working on Big Sur support, not sure if this means they will get AS support going soon as well. Have not tried Octane yet.
  • Cinebench scores are really promising.
    • MacBook scores 7476​
    • My 10c iMac Pro scores 11096​
  • Looks like it will be a LONG time before Adobe apps are fully ported. I'm guessing it will be the June release before we see any apps out of beta. Maybe a year or more before we see the entire suite ported. There is a M1 native beta for Photoshop, but it is missing a ton of features. Looks like Adobe will finally have to modernize their ancient code in all their Creative Apps to port them over. You can't even install freaking Acrobat in Big Sur.
  • After Effects in Rosetta seems to work. It boots in 8 secs on my M1 MacBook, 10 secs on my iMac Pro. I have not fully installed all of my plug-ins on the MacBook (this may be the speed difference) or loaded a complex project.
I've been hovering over a PC build for a long time and almost pulled the trigger, but now I have serious hope for the future of the Mac in mograph. We're FINALLY getting metal support in the renderers. AMD is launching cards in spitting distance of Nvidia. Maxon seems all in on AS, both C4D and Redshift devs were on the M1 launch reel, and the R23 launch reel heavily featured C4D running on Mac OS.

I'm going to try to pick up a 6800xt today and put it in my eGPU for use on the iMac Pro. I figure that will be a good stop gap until we get a better look at what the future AS pro machines will look like. [HA,HA,HA- I couldn't even get the purchase page to load before they sold out. Damn bots.]

I'll report back once I have a chance to load a few projects and test the MacBook out more.
 
Last edited:

vinegarshots

macrumors 6502a
Sep 24, 2018
938
1,296
I've been hovering over a PC build for a long time and almost pulled the trigger, but now I have serious hope for the future of the Mac in mograph. We're FINALLY getting metal support in the renderers. AMD is launching cards in spitting distance of Nvidia. Maxon seems all in on AS, both C4D and Redshift devs were on the M1 launch reel, and the R23 launch reel heavily featured C4D running on Mac OS.


I did finally ditch my iMac Pro and pulled the trigger on a PC with a Nvidia 3080 (Blender user here), and the realtime raytraced viewport performance is mental with the Nvidia Optix optimizations and denoiser. It's like a multi-generational leap in performance. No regrets :D

Here's a video of it in action:
Nvidia Optix in Blender

I need to see even more impressive performance than that on the Mac side before Ill ever consider switching back.
 

skippermonkey

macrumors 6502a
Jun 23, 2003
620
1,522
Bath, UK
Bit of a waiting game for me now – Octane doesn't yet work on Big Sur, and I think it'll be a few weeks before the C4D plugin is ready. My dual 5700XTs perform pretty well in Octane but I'd like to see how 6800XTs perform in my Mac Pro. If anyone finds any info or a link, post it on here!
 

skippermonkey

macrumors 6502a
Jun 23, 2003
620
1,522
Bath, UK
Quick update: I'm now on OctaneX PR7, macOS 11.1. I unsubscribed from Octane for a while as PR5 was pretty much broken. There's now an issue with denoising which, thankfully, Otoy has acknowledged and s fixing. So fingers crossed, PR8 will be solid, fast and (hopefully) free of major issues. I think I'll try putting my RX580X MPX back in the MacPro and see how I get on with three GPUS!

EDIT: Curses, I'm on Octane Studio, limited to two GPUs. Hopefully I'll get my free years' license in 20201 – about two years later it was first mooted.
 
Last edited:
  • Like
Reactions: P_Himmel

soundboutiquesean

macrumors newbie
Dec 27, 2020
1
0
Quick update: I'm now on OctaneX PR7, macOS 11.1. I unsubscribed from Octane for a while as PR5 was pretty much broken. There's now an issue with denoising which, thankfully, Otoy has acknowledged and s fixing. So fingers crossed, PR8 will be solid, fast and (hopefully) free of major issues. I think I'll try putting my RX580X MPX back in the MacPro and see how I get on with three GPUS!

EDIT: Curses, I'm on Octane Studio, limited to two GPUs. Hopefully I'll get my free years' license in 20201 – about two years later it was first mooted.
Are you talking about on a M1 Mac?? if so, can you please share how you've been able to install it and get it to work?
I have been in touch with Otoy and the developers and have not been able to get any version of Octane installed and working in C4D on any M1 Mac, which includes MacBook Air, and Mac mini.
 
Last edited:

skippermonkey

macrumors 6502a
Jun 23, 2003
620
1,522
Bath, UK
Are you talking about on a M1 Mac?? if so, can you please share how you've been able to install it and get it to work?
I have been in touch with Otoy and the developers and have not been able to get any version of Octane installed and working in C4D on any M1 Mac, which includes MacBook Air, and Mac mini.
Hi. No. I‘m on C4D R23 using Octane on Metal with GPUs. Since I wrote that I’m now on PR11 running all three GPUS. Seems to work fine, though I’m not doing much CG at the moment.
 

hifimac

macrumors member
Mar 28, 2013
61
40

Since GPU lineally scale, can someone help figure out what a 128 Core GPU would look like based on current M1 benchmarks? What would its Nvidia equivalent be?

 

joema2

macrumors 68000
Sep 3, 2013
1,644
861
...Since GPU lineally scale, can someone help figure out what a 128 Core GPU would look like based on current M1 benchmarks? What would its Nvidia equivalent be?..
If you extrapolate linearly from the published Geekbench 5 compute numbers for the M1, that implies the hypothetical M1X with 32 GPU cores might be at least 30% faster than the 16GB Vega 64 in the iMac Pro.

If you keep projecting along that line, it would imply the rumored Apple Silicon Mac Pro with a 128-core GPU might be roughly 4x that fast, or about 350,000 on Geekbench 5 compute. That might be very roughly 70% faster than an RTX-3090.

That assumes no further per-core enhancements and also discounts any performance advantage of using tile-based deferred rendering, which is possible with the Metal 2 framework.

Those performance levels are so far beyond the current M1 they could be way off. It's like trying to measure 20 feet with a one-inch ruler. We will know a lot more in the apparently near future when the "M1X" is released and the 32-core GPU is benchmarked.

There are other issues besides Apple Silicon GPU benchmark performance. If Apple holds to the unified memory design, that should greatly reduce overhead in copying GPU memory buffers back and forth to main memory. In some software it's not unknown for an app to actually slow down when GPU acceleration is enabled. Some past versions of Lightroom did this, possibly due to inefficiently copying data between GPU and main memory. In theory even a poorly written app might have better real-world performance on an Apple Silicon GPU than a traditional GPU having higher benchmark scores.

The problem is the only current example we have is the M1 with 8 GPU cores, so it only covers the lower performance echelon, also not all popular applications are yet available in native Apple Silicon format. The "M1X" will enable a better assessment higher up the performance scale.
 

joema2

macrumors 68000
Sep 3, 2013
1,644
861
On ATP Siracusa's "back of the envelope math" said 128 GPU cores would be just slightly faster than a 3090. He's got some interesting theories on how the next AS Mac Pro will stack up.
From the podcast, his idea on the GPU performance of the upcoming Apple Silicon seems based solely on a theoretical GPU "teraflops" spec. But there is no need for extrapolating from specs -- actual Geekbench compute benchmarks exist for people using RTX-3090s in a Hackintosh. Those produce from about 190,000 to 220,000. We also have the Geekbench 5 compute numbers for the M1 with 8 GPU cores, which is around 22,000.

So based on those, an Apple Silicon Mac Pro with a 128-core GPU might produce around 16x 22,000, or 350,000 on the Geekbench 5 compute benchmark. That would be far more than "just slightly faster than a 3090".

Siracusa also can't believe Apple could put that in a single package, because he reasons the equivalent RTX-3090 burns so much power (actually up to 350 watts). He can't envision that much GPU horsepower (and heat) in the same package as Apple's 40-core CPU.

Based on that reasoning you could look at the Radeon Pro 560X in a 2019 MacBook Pro 15 (which burns 60-80 watts and produces a GeekBench 5 Metal score of about 18,000), and conclude it would be impossible for Apple to make an M1 chip which has better GPU performance yet consumes a fraction of the power. Yet that product is shipping today.

So he is thinking in terms of past and present products. The 3090 uses 8 nanometer fabrication, whereas it's likely the future Mac Pro "System in Package" will be 4 nanometers or better.
 

diamond.g

macrumors G4
Mar 20, 2007
11,099
2,438
OBX
From the podcast, his idea on the GPU performance of the upcoming Apple Silicon seems based solely on a theoretical GPU "teraflops" spec. But there is no need for extrapolating from specs -- actual Geekbench compute benchmarks exist for people using RTX-3090s in a Hackintosh. Those produce from about 190,000 to 220,000. We also have the Geekbench 5 compute numbers for the M1 with 8 GPU cores, which is around 22,000.

So based on those, an Apple Silicon Mac Pro with a 128-core GPU might produce around 16x 22,000, or 350,000 on the Geekbench 5 compute benchmark. That would be far more than "just slightly faster than a 3090".

Siracusa also can't believe Apple could put that in a single package, because he reasons the equivalent RTX-3090 burns so much power (actually up to 350 watts). He can't envision that much GPU horsepower (and heat) in the same package as Apple's 40-core CPU.

Based on that reasoning you could look at the Radeon Pro 560X in a 2019 MacBook Pro 15 (which burns 60-80 watts and produces a GeekBench 5 Metal score of about 18,000), and conclude it would be impossible for Apple to make an M1 chip which has better GPU performance yet consumes a fraction of the power. Yet that product is shipping today.

So he is thinking in terms of past and present products. The 3090 uses 8 nanometer fabrication, whereas it's likely the future Mac Pro "System in Package" will be 4 nanometers or better.
Why would you not need 160 watts to power 128 cores if you only need 10 to power 8 cores?
 

joema2

macrumors 68000
Sep 3, 2013
1,644
861
Why would you not need 160 watts to power 128 cores if you only need 10 to power 8 cores?
I don't think we know how much power each 8-core GPU consumes on the M1. The chip inside the package is a monolithic, single-die design, combining CPU, GPU and lots of other logic. Without GPU-specific power consumption you can't accurately predict the power consumption of a 128-core Apple GPU, or the equivalent such as 4 x 32-core GPU dies or chiplets.

Siracusa's point was since an nVidia RTX-3090 consumes a vast amount of power (up to 350 watts), he assumed an Apple Silicon GPU with similar performance would consume roughly similar wattage, so how could Apple possibly put that inside the same package as a 40-core CPU?

But he was thinking of power and heat in terms of current CPUs and GPUs. It's definitely true if you tried to cram the silicon dies of a 40-core Xeon and an RTX-3090 into a single soldered-in package, it would be a problem. A 36-core Xeon 8360Y has a TDP of 250 W an the RTX-3090 is about 350 W, so you'd be putting 600 W inside a single package.

The future Apple Silicon SoC for the Mac Pro will probably not consume nearly that much power. If it were, say, 160-220 watts total between CPU and GPU, that wouldn't be bad at all. It could easily be handled by a conventional cooling solution. Siracusa's point was based on the RTX-3090, the Apple Silicon SoC for the upcoming Mac Pro would consume vastly more power than that. I'm not sure that will happen.

Either way we will likely know within about a year. Even the upcoming "M1X" could reveal a lot more, and that announcement may only be a few weeks out.
 
  • Like
Reactions: JMacHack

leman

macrumors Core
Oct 14, 2008
19,174
19,003
I don't think we know how much power each 8-core GPU consumes on the M1.

We know it exactly because the diagnostic tools report precise power statistics for each compute cluster. M1 GPU (all 8 cores) tops at 10W, always. Never saw it go any higher.
 
  • Like
Reactions: jdb8167

jdb8167

macrumors 601
Nov 17, 2008
4,724
4,415
We know it exactly because the diagnostic tools report precise power statistics for each compute cluster. M1 GPU (all 8 cores) tops at 10W, always. Never saw it go any higher.
This is what I see too. 100% at 1.278 GHz for all 8 GPU cores peaks at just above 10000 mW on my M1 8 GPU core MacBook Air. So about 1256 mW/core peak. Being an MBA, it almost immediately throttles so the average is obviously less but for most M1 Macs it probably should stay around peak. If anyone has an M1 mini, MacBook Pro, or M1 iMac they can test this with: sudo powermetrics -i2000 --samplers gpu_power,cpu_power. I used the 3DMark Wild Life Extreme GPU test.

This is representative of my MBA GPU peak:
Screen Shot 2021-05-23 at 6.15.36 PM.png
 

joema2

macrumors 68000
Sep 3, 2013
1,644
861
We know it exactly because the diagnostic tools report precise power statistics for each compute cluster. M1 GPU (all 8 cores) tops at 10W, always. Never saw it go any higher.
OK, thanks. If next year's Apple Silicon Mac Pro used that exact same GPU built on the exact same fabrication, it might indeed be 16 x 10W, or 160 watts for the 128-core GPU.

However it appears that Apple may be using 3 nanometer fabrication by then, which is considerably more power efficient: https://www.computerworld.com/artic...for-3nm-silicon-iphones-and-more-in-2022.html

The iMac Pro max power consumption is about 370 watts, which is handled well in a fairly slim case. Even IF the Apple Silicon Mac Pro in a smaller case consumed a similar amount of power, I don't see that as problematic.

Siracusa speculated that based on the RTX-3090's power consumption, the upcoming Apple Silicon Mac Pro could require an enormous cooling system. I don't see that as likely, but we will all know within about 12 months.
 

BootLoxes

macrumors 6502a
Apr 15, 2019
745
858
Substance painter and marmoset toolbag 4 render perfectly fine. Slower than my desktop but thats to be expected since its rosetta. All rendered in toolbag 4 with ray tracing turned on. I am by no means a pro but I am happy with how well my air handles it
 

Attachments

  • DBC6FF57-498F-443C-B848-E0B99C3D189B.jpeg
    DBC6FF57-498F-443C-B848-E0B99C3D189B.jpeg
    309 KB · Views: 222
  • A766B949-02A3-4723-960D-968DDF8C8C52.jpeg
    A766B949-02A3-4723-960D-968DDF8C8C52.jpeg
    177.9 KB · Views: 225
  • 58EB6CEF-A561-4D85-91AE-EA68FFA317B4.jpeg
    58EB6CEF-A561-4D85-91AE-EA68FFA317B4.jpeg
    120.9 KB · Views: 257

vel0city

macrumors 6502
Original poster
Dec 23, 2017
347
510
Moved from a fully maxed out 5,1 Mac Pro with an Nvdia 1080ti GPU to a 16GB/2TB Mac Mini.

Overall the Mini is noticeably and enjoyably faster for everything apart from final rendering - so that includes Photoshop, working in C4D, Zbrush, Final Cut Pro. I've kept my 5,1 in my studio for final rendering/team rendering.

Thought I'd post my findings with C4D and third party renderers so far using C4D R23. R23 because that's the last version I can run on my 5,1 and I'm using Team Rendering so both versions need to be the same. My main work is stills and short loops so be aware I'm not outputting movies. My render settings are usually 4000x4000px and I always utilise AOVs for masks, depth and lighting. My images usually make use of displacement maps, PBR textures, particles, Megascans assets and Substances.

Onto the renderers.

C4DtoA (Arnold) Running in Rosetta. this is a CPU renderer but thought I'd throw it in here. Absolutely usable and workable on the M1 Mini. IPR speed is not far off my 5,1 though that was never what you'd call fast. Using it for lookdev and setting up scenes is absolutely fine. Rendering speed is actually faster than my 5,1 using the same scene and settings. Team rendering works as expected and it's nice to see all those additional buckets helping out.

Octane Running in Rosetta. Exciting to run Octane on a Mac without an Nvidia card, but it crashed so often I can't give much of an opinion on this. A real bummer because I love Octane's look, but even on a basic, basic scene, Octane brings C4D down needing a force quit. This has been my experience of Octane on the 5,1 under CUDA as well. Moving on.

Redshift running natively. Been using it a lot and found it absolutely stable, responsive and reliable. I've created two render settings to flick between, one very low quality for positional/setting up, then a medium quality setting for a better look at how things are looking. This is working out well and I'm getting some proper work done. It's all about optimise, optimise, optimise. Make use of freezing tessellation, turn on IPR undersampling, crank up your Unified Sampling Threshold to a level that's not too noisy and just eyeball it until you have something you can work with.

Final rendering in Redshift is of course slower than using a dedicated GPU. But the M1 Min remains silent and cool and just renders in the background with no impact on the responsiveness of the system, leaving plenty of resources free to work in Photoshop or Zbrush. Full disclosure - I wouldn't rely on this workflow 100% and am using my 5,1 for final rendering in most cases. I've created a workflow using Dropbox and working from a synced folder so that assets are shared across both machines which is working out well so far. Because you need a Redshift license for both machines, I'm just using the Redshift demo on my M1 and my licensed version on the 5,1 which gives me final renders with no watermark.

Looking forward to the Pro machines and will be all over them day one. But for now this is a workable transition (would be even better if we could use an eGPU but so it goes) that will get me through until the Apple Silicon Pro is available.
 

ader42

macrumors 6502
Jun 30, 2012
421
373
Substance painter and marmoset toolbag 4 render perfectly fine. Slower than my desktop but thats to be expected since its rosetta. All rendered in toolbag 4 with ray tracing turned on. I am by no means a pro but I am happy with how well my air handles it
Do you happen to know how ZBrush runs?
 

vel0city

macrumors 6502
Original poster
Dec 23, 2017
347
510
Do you happen to know how ZBrush runs?

I know you didn't direct this at me but from my experience so far, working on a high-density sculpt with a high subtool count and using GoZ to send it to and from C4D, Zbrush is working very well in Rosetta and I didn't have any stability issues - it was always stable and fast on my Mac Pro though and doesn't feel much different on the M1.

I 've also used Zbrush on a regular iMac and during any sculpting the fans would always ramp up and stay up - on the M1 Mini it remains silent, you genuinely wouldn't know it was on. Likewise with rendering in Arnold and Redshift - completely silent and nonintrusive. Working late at night and not hearing a Mac whirring away during a render is certainly a novelty.

Edit - Zbrush launches in 3.8 seconds in Rosetta, just timed it. Latest version.
 
  • Like
Reactions: bhodinut and ader42

BootLoxes

macrumors 6502a
Apr 15, 2019
745
858
Do you happen to know how ZBrush runs?
Well a detailed answer was given before I could reply but Zbrush runs amazingly. I am excited to see what a native m1 version will be like. I can easily hit several million (I think 7 mil) before seeing any serious lag. Though I never break more than 2 mil on my stuff so it has never been an issue for me
 
  • Like
Reactions: vel0city and ader42

vel0city

macrumors 6502
Original poster
Dec 23, 2017
347
510
Well a detailed answer was given before I could reply but Zbrush runs amazingly. I am excited to see what a native m1 version will be like. I can easily hit several million (I think 7 mil) before seeing any serious lag. Though I never break more than 2 mil on my stuff so it has never been an issue for me

Have you tried rendering in Keyshot with the Zbrush>Keyshot Bridge? Be interesting to see how a dedicated standalone CPU renderer performs on the M1. Keyshot has a C4D rendering plugin too, never tried it myself but would like to hear any reports on it.

I'm working on a dark gothic client illustration right now and just added 1000 monks to a scene using Redshift scatter, it's appreciably more responsive and providing much faster visual feedback than my Mac Pro. The Renderview updates instantly and of course the M1 is completely silent - just a joy to use. In the same scene I'm using volumetric light, Megascans assets, displaced textures, post effects and its still completely workable and responsive. Redshift definitely feels like the most robust and stable renderer on the M1 right now.
 
  • Like
Reactions: ader42
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.