3D Rendering on Apple Silicon, CPU&GPU

[Xiao_Xi]

If Apple cant improve this, then their high end market will be limited only for video and photo.

An Apple engineer has said on the Blender forum:

1. The structure of the renderer is much aligned with the existing path taken for CPU and CUDA, with little leverage of Apple Silicon’s more unique architecture as of yet. We do leverage the unified memory architecture to avoid duplication of resources, but there’s much more to do on this front and we’re keen to see that leaned on in CPU+GPU rendering modes. There is certainly scope to use the Apple Neural Engine for denoise in the viewport too.
2. Correctness has definitely been a focus for us, with ensuring we get solid results and a renderer users can use and rely on. This is intended to be a tech demo - it is aimed to be a tool that users can use all day every day. Some of the early R&D we’ve done has resulted in render performance being more than doubled over where it is now, but taking these prototypes and productising them is another matter, and takes significant time. The avenue to performance on Apple Silicon means driving the GPU in the way that is most efficient for its architecture. Each GPU architecture is different though, and we need to be able to cleanly drive our GPUs more efficiently but without compromising the existing performance on other GPUs.
3. Optimisation is going to be an ongoing effort, rather than a task we tackle just the once, and I’m hoping the team can see some improvements land in every release. We have big ambitions.

Cycles Apple Metal device feedback

The structure of the renderer is much aligned with the existing path taken for CPU and CUDA, with little leverage of Apple Silicon’s more unique architecture as of yet. We do leverage the unified memory architecture to avoid duplication of resources, but there’s much more to do on this front and...

devtalk.blender.org

 

[mi7chy]

Blender community on Reddit believe that Apple's marketing is a lie and M1 will never compete with dedicated GPUs as M1 Max proves to be worse than 1660.

If Apple cant improve this, then their high end market will be limited only for video and photo.

Marketing vs reality. Shouldn't be a surprise if you've been following this thread for the last several months.

 

[l0stl0rd]

Marketing vs reality. Shouldn't be a surprise if you've been following this thread for the last several months.

yeah but who knows… the important thing is will this happen or not “Some of the early R&D we’ve done has resulted in render performance being more than doubled over where it is now, but taking these prototypes and productising them is another matter, and takes significant time.”

If the M1 Max ends up being twice as fast as it is now it sure would be closer to their marketing bg claim.

 

[JimmyjamesEU]

Marketing vs reality. Shouldn't be a surprise if you've been following this thread for the last several months.

Indeed. It'd be like quoting a benchmark where the pc uses hardware acceleration and the mac doesn't. What kind of person would do that?

 

[ader42]

Beware of the marketing term “up to”.

 

[ader42]

Is there a consensus for the best (local/offline) general 3d software on Mac (initial purchase price irrelevant) - is it Maya with RenderMan engine given that RenderMan utilises both CPU and GPU? Especially as we now have Mac Studio Ultra with 20 cpu cores and 64 gpu cores?

 

[sunny5]

Marketing vs reality. Shouldn't be a surprise if you've been following this thread for the last several months.

I wouldn't say marketing because M1 Max IS powerful. Optimization is a biggest problem especially for 3D since all of them uses Nvidia CUDA for a long time.

 

[Xiao_Xi]

M1 Max IS powerful.

Apple's marketing makes it sound like the M1 Max is as powerful as NVIDIA GPUs for all workloads, but it appears that it is only good for some workloads.

 

[sunny5]

Apple's marketing makes it sound like the M1 Max is as powerful as NVIDIA GPUs for all workloads, but it appears that it is only good for some workloads.

And most of software are heavily optimized for Nvidia GPU. Not even a fair comparison.

 

[Xiao_Xi]

And most of software are heavily optimized for Nvidia GPU. Not even a fair comparison.

A comparison that Apple started.

Fans of Nvidia and AMD may find it unfair that Apple has more money than those companies and can be the first to get access to TSMC's best node.

 

[JimmyjamesEU]

A comparison that Apple started.

Fans of Nvidia and AMD may find it unfair that Apple has more money than those companies and can be the first to get access to TSMC's best node.

Apple is comparing software that is optimised for both. Even then I’d wager the benchmarks are optimised even more for x86/Nvidia.

i would find it amusing if any Nvidia fan was hypocritical enough to complain about a company abusing its fiscal power.

 

[sunny5]

A comparison that Apple started.

Fans of Nvidia and AMD may find it unfair that Apple has more money than those companies and can be the first to get access to TSMC's best node.

And yet, Nvidia compared 3D software against M1 Max while none of them are truly optimized for Apple Silicon. How ironic.

 

[jujoje]

Is there a consensus for the best (local/offline) general 3d software on Mac (initial purchase price irrelevant) - is it Maya with RenderMan engine given that RenderMan utilises both CPU and GPU? Especially as we now have Mac Studio Ultra with 20 cpu cores and 64 gpu cores?

To paraphrase Dr Johnson, using Maya for 3D is like a dog walking on it's hind legs. It is not done well; but you are surprised to find it done at all.

But in all seriousness, there's no real 'best 3D software' on the Mac - they all have their strengths and weaknesses. Maya is good for animation and modelling, Houdini for FX and procedural things and Blender is… Blender.

Actually kind of feel Blender actually is pretty decent from a generalist point of view; personally I don't think it's really good at anything, but it's pretty decent at a lot of things. Although, on a bit of a tangent, the Blender foundations decision to stop following the vfx reference platform and go off on their own (because they wanted to use a new version of python of al things), has been one of the stupidest things I've seen for a while, and pretty much killed it's adoption in the Film industry. It's a shame as it was getting a fair bit of traction in the industry.

Back on topic, in terms of Apple Silicon, Blender and Houdini have native builds, the latter being in Alpha at the moment. I'm not sure what the timeframe is for an AS version of Maya though (knowing Autodesk, no time soon).

As for Renderman, the cpu side of things are ok, but non-native, however the xPu relies heavily on Nvidia libraries so won't be coming to Apple Silicon anytime soon.

Been getting some reasonable results with Karma, which is, once again, non-native, and the xpu version is a ways off.

Natively, Redshift, Octane and Cycles spring to mind.

TLDR - Really depends what work you're doing; there's no real consensus.

 

[jujoje]

An Apple engineer has said on the Blender forum:

Cycles Apple Metal device feedback

The structure of the renderer is much aligned with the existing path taken for CPU and CUDA, with little leverage of Apple Silicon’s more unique architecture as of yet. We do leverage the unified memory architecture to avoid duplication of resources, but there’s much more to do on this front and...

devtalk.blender.org

That was pretty interesting. Really cool that they could use the Neural Engine for denoising; hopefully it would be faster and better quality than Nvidia's denoiser (the Intel denoiser is good, but very much a post process).

Also was interesting to hear where they see development being; sounds like there's a lot of room for more architectural specific improvements. Very curious to see where things end up after that and a bit of optimisation. Sounds pretty promising though.

 

[Boil]

That was pretty interesting. Really cool that they could use the Neural Engine for denoising; hopefully it would be faster and better quality than Nvidia's denoiser (the Intel denoiser is good, but very much a post process).

Also was interesting to hear where they see development being; sounds like there's a lot of room for more architectural specific improvements. Very curious to see where things end up after that and a bit of optimisation. Sounds pretty promising though.

I liked this part of it all...

Optimization is going to be an ongoing effort, rather than a task we tackle just the once, and I’m hoping the team can see some improvements land in every release. We have big ambitions.​

 

[jujoje]

I liked this part of it all...

Optimization is going to be an ongoing effort, rather than a task we tackle just the once, and I’m hoping the team can see some improvements land in every release. We have big ambitions.​

It must be said that Apples Pro Software team are really knocking it out to the park in providing support and assistance to getting things up and running on Apple Silicon - see Redshift, Octane, C4D and now Houdini (and on the Houdini front a lot of the long standing Mac bugs have been fixed in H19). Was a bit dubious about how serious Apple was about making the Mac a serious tool for 3D, but feeling a lot more bullish about it these days.

 

[jujoje]

Back on topic, in terms of Apple Silicon, Blender and Houdini have native builds, the latter being in Alpha at the moment. I'm not sure what the timeframe is for an AS version of Maya though (knowing Autodesk, no time soon).

Speaking of C4D totally forgot to add it here; somewhat shows my bias towards film and TV

 

[singhs.apps]

Is there a consensus for the best (local/offline) general 3d software on Mac (initial purchase price irrelevant) - is it Maya with RenderMan engine given that RenderMan utilises both CPU and GPU? Especially as we now have Mac Studio Ultra with 20 cpu cores and 64 gpu cores?

Why would you need Renderman? Arnold is just as robust world class renderer and it’s virtually native in Maya.

With regards to best…it depends and you can’t go wrong with Maya, Cinema 4D and Blender if you are a generalist, with the latter two getting a slight edge, being slightly more solo artist friendly, but Maya has almost caught up now in that dept.

Houdini is great for simulation and it has been making amazing strides in other areas.

If you intend to work for, or with large VFX houses, then you can invest in either Maya or Houdini.

 

[ader42]

To paraphrase Dr Johnson, using Maya for 3D is like a dog walking on it's hind legs. It is not done well; but you are surprised …

TLDR - Really depends what work you're doing; there's no real consensus.

Thanks for the great input. I’m an old dog that used to use Lightwave last century - I did use C4D a bit too.
Recently tried Blender but found the UI to be pretty terrible - it reminded me of the pain I used to feel with 3DS Max at work - when I had to use that.

My main intent is ZBrush modelling and then taking work from there - i was leaning towards Maya for character rigging & blendshapes type of work but given how ZBrush is Maxon owned now I think I should try out C4D as well as Maya and see which gives me the features/performance I want.

As for why RenderMan, it was purely that I thought it might have better render times being xPu but that sounds like an error on my part.

 

[singhs.apps]

If Your area of interest is character based, Maya is still king amongst all the commercially available apps, bar none.
If any major character animation based tech comes along, chances are it will support Maya first.

Maxon buying zbrush doesn’t make it the defacto choice for zbrush support for characters. Maya is still preferred in that area…

Speaking of zbrush, Autodesk should reinvest in Mudbox..with Apple showing keen interest in 3D, Mudbox would perform really well with its silicon, unified memory and all

 

[darngooddesign]

Indeed. It'd be like quoting a benchmark where the pc uses hardware acceleration and the mac doesn't. What kind of person would do that?

It's a valid benchmark as far as completing a task is concerned. If you work is primarily renders, and hardware accelerated/optimized computers render quicker, then that is valid as far as renders are concerned. Doesn't Apple have video encoder/decoder acceleration built into the M1? We're fine with M1 video benchmarks being great.

 

[Boil]

I’m an old dog that used to use Lightwave last century...

EIAS here, up until I lost my license dongle in a move...!

Speaking of zbrush, Autodesk should reinvest in Mudbox..with Apple showing keen interest in 3D, Mudbox would perform really well with its silicon, unified memory and all

I would really, Really, REALLY love to see an Apple first-party 3D/DCC software suite someday...!

 

[JimmyjamesEU]

It's a valid benchmark as far as completing a task is concerned. If you work is primarily renders, and hardware accelerated/optimized computers render quicker, then that is valid as far as renders are concerned. Doesn't Apple have video encoder/decoder acceleration built into the M1? We're fine with M1 video benchmarks being great.

I feel you're missing a great deal of history. My comment refers to specific things mi7chy has done. They (several weeks ago) showed a video with the pc having acceleration turned on, and the mac having acceleration deliberately turned off. Not that the mac didn't have it...it was turned off. This was put forward as proof of how bad the m1 was.

 

[jujoje]

If Your area of interest is character based, Maya os still king amongst all the commercially available apps, bar none.
If any major character animation based tech comes along, chances are it will support Maya first.

Maxon buying zbrush doesn’t make it the defacto choice for zbrush support for characters. Maya still is preferred in that area…

Speaking of zbrush, Autodesk should reinvest in Mudbox..with Apple showing keen interest in 3D, Mudbox would perform really well with its silicon, unified memory and all

Yeah definitely agree with that; if you're looking at character stuff, particularly animation and rigging with a bit of modelling/retopo, Maya is definitely the one to go for (RIP XSI). Hopefully KineFX and the character tools in Houdini will offer a bit of competition and spur Autodesk into innovating a bit more in that space, but it's a ways off. Most of my grumpiness with Maya comes from how slow and staid development has become since Autodesk bought it.

That said there have been some interesting new things in animation from physics based characters to ML stuff that have been cropping up with increasing frequency lately, so perhaps it's an area of the the industry where we'll see some exciting new things in soon.

Speaking of Autodesk, it's such a shame they failed to develop Mudbox, pretty much at all; had such potential as more artist friendly zbrush, but they just let it languish.

As for why RenderMan, it was purely that I thought it might have better render times being xPu but that sounds like an error on my part.

It's a bit frustrating that Pixar has gone all in on Nvidia rather than a more cross vendor API; the Renderman xPU looks great, although still a bit limited.

 

[-narcan-]

Unfortunate to see the ultra is going to be no good for 3D, but knew it all sounded a bit too good to be true.