3D Rendering on Apple Silicon, CPU&GPU

[mi7chy]

There is no real reason to put them in a ring either. You’d be better off connecting them all to a network.

Clarify what is meant by network since it could be Thunderbolt 4, 10Gb ethernet or quasi combo of both? Continue down the road of using Thunderbolt 4 but there aren't any 20/30-port TB 4 switches that comes up on search so it's still point-to-point with unknown overhead between nodes. Or, shelving TB 4 which is intended more for connecting peripherals rather than networking at scale and using 10Gb ethernet switching star topology?

 

[sirio76]

Sure, I have been used C4D Team Render with couple of MacPro's and iMac's several times but never used VRay for this. Also there is another problem shiny magnificent superior magical ''Subscription Model'' . For Maxon/Redshift you had to buy another license for each machine no matter what. I have no idea how it is work with VRay DR, if it is reasonable price it could work in logical budgets.

C4D Team Render is very inefficient in my experience and is good mostly for animation, not for rendering large still. Last time I use it for an animation with Corona I loose about 40% of the performance of my Threadripper slave.
Performance may vary and beside Vray DR there are other third party distributed rendering systems, but in general no matter what you use you need to pay extra for standalone license. Don’t know if it still like this but Arnold a few years ago was charging the same cost of the software for additional standalone licenses, Vray, C4D and other charge only a fraction.

 

[jmho]

Clarify what is meant by network since it could be Thunderbolt 4, 10Gb ethernet or quasi combo of both? Continue down the road of using Thunderbolt 4 but there aren't any 20/30-port TB 4 switches that comes up on search so it's still point-to-point with unknown overhead between nodes. Or, shelving TB 4 which is intended more for connecting peripherals rather than networking at scale and using 10Gb ethernet switching star topology?

Say you run the render script on your lead machine and it tells every node to sync its scene files. You already have most of your assets in a shared asset library on each machine, so let’s say that downloading the main scene file takes each machine a second over 10Gb ethernet.

Your render node then renders for 35 hours, and then produces a couple of gigabytes of images which it then uploads to a shared drive that takes again a handful of seconds.

I'm not sure it's worth fretting over like a handful of seconds when you're rendering for 35 hours. You could use a thumb drive and load all the files manually and it still wouldn't make a huge difference in the grand scheme of things.

 

[aytan]

C4D Team Render is very inefficient in my experience and is good mostly for animation, not for rendering large still. Last time I use it for an animation with Corona I loose about 40% of the performance of my Threadripper slave.
Performance may vary and beside Vray DR there are other third party distributed rendering systems, but in general no matter what you use you need to pay extra for standalone license. Don’t know if it still like this but Arnold a few years ago was charging the same cost of the software for additional standalone licenses, Vray, C4D and other charge only a fraction.

Actually I do not remember the exact performance over Team Render, I have use for rendering out animation scenes overnight with couple of machines. Redshift/Arnold demands standalone license for each machine, unless if you have a business as a studio you could get an agreement for bulk licenses. Interesting side is if you use Octane for your workflow with any number of Macs there is no cost, you can use each Mac with octane. Because by M1/2 Ultra or max seen as single GPU unit and you can not add another GPU but it is free. On the other side for number of PC's you had to get licenses ( in the past there was render nodes for Octane but I do not know how it works today ) for each of them.
For Redshift and Arnold Prior number of standalone bulk license price will very upon the number of machines you have however each of them will be cheaper than standalone license. Last time I check Redshift it was that way, maybe it is different today I have no idea.
Anyway basic issue is using 3-4-5 machines for rendering is not cheap for Freelancers. Maybe 2 machine could be use with 1 license. For rendering which one is faster you could use it for rendering and at the same time you could work with other machine for texturing modeling or compositing. This is how I use my M1 Ultra and PC for now because Maxon One is not cheap today and will not be cheaper in the near feature.

 

[aytan]

Actually I do not remember the exact performance over Team Render, I have use for rendering out animation scenes overnight with couple of machines. Redshift/Arnold demands standalone license for each machine, unless if you have a business as a studio you could get an agreement for bulk licenses. Interesting side is if you use Octane for your workflow with any number of Macs there is no cost, you can use each Mac with octane. Because by M1/2 Ultra or max seen as single GPU unit and you can not add another GPU but it is free. On the other side for number of PC's you had to get licenses ( in the past there was render nodes for Octane but I do not know how it works today ) for each of them.
For Redshift and Arnold Prior number of standalone bulk license price will very upon the number of machines you have however each of them will be cheaper than standalone license. Last time I check Redshift it was that way, maybe it is different today I have no idea.
Anyway basic issue is using 3-4-5 machines for rendering is not cheap for Freelancers. Maybe 2 machine could be use with 1 license. For rendering which one is faster you could use it for rendering and at the same time you could work with other machine for texturing modeling or compositing. This is how I use my M1 Ultra and PC for now because Maxon One is not cheap today and will not be cheaper in the near feature.

The only imaginery solution is ( for MacOS side ) using number of Macs, 1 of them is master machine and daisy chain other number of macs ( does not important their M1/2 models, air, mini, pro mini, max, ultra can be functional ) as slave GPU's. This is the only way to use MacOS for distributed rendering setup. Which shown by Apple for AI systems with number of Ultra's work as slave GPU's for Master Ultra machine.

( Xiao_Xi has posted the video from WWDC )​

 

[ader42]

Render Farm Software | CrowdRender

Crowdrender is free software you can use to build your own render farm then combine it's power with the cloud, your friend's computers to make distributed compute clusters

www.crowd-render.com

I bet even a bunch of M2 Minis would make a good render-farm for those doing mostly animations.
Obviously cloud rendering is ofte the best option, but it’s not always an option due to confidentiality.

 

[aytan]

Render Farm Software | CrowdRender

Crowdrender is free software you can use to build your own render farm then combine it's power with the cloud, your friend's computers to make distributed compute clusters

www.crowd-render.com

That is great for Blender projects 👍 Thanks for the link

 

[Xiao_Xi]

That is great for Blender projects

There were two sessions on distributed rendering at the last Blender Con.

 

[senttoschool]

MacOS is great on local desktop. When running things in the cloud you often don’t even know (or care) what OS you are running on. How would that be a selling point?

3 categories:

1. Software that works well with macOS such as Metal-optimized renderer or some task that requires a huge amount of VRAM. No UI needed. Just terminal.

2. Actually virtualizing macOS and use it as a remote desktop to do all your work on an Extreme SoC.

3. Accelerating local applications that are optimized for Apple Silicon chips through an Apple Silicon cloud.

 

[senttoschool]

This is a fair point. Very possible that you are right. But it's also possible that by offering a compelling product for a reasonable price a certain niche can be carved out. Apple's technology would work well in this scenario. Whether they are interested is a whole different question.

Apple doesn't compete on price, it offers features that no one else has. What could Apple offer that no one else does? What would make an Apple render farm a success?

Exactly. I don't think there is anything.

My argument this whole time is that it doesn't make economic sense to make an Extrem SoC for a niche product. Therefore, you need a bigger market for an Extrem SoC than just in a Mac Pro that will sell in the tens of thousands each year. Apple needs to sell hundreds of thousands each year and have the market grow each year instead of shrink.

How Apple can be competitive in the Cloud is a different question and one where we can debate.

 

[Boil]

M3 Extreme in the Cloud running 3D/DCC apps via Vision Pro thin client...

 

[senttoschool]

M3 Extreme in the Cloud running 3D/DCC apps via Vision Pro thin client...

I’m not sure if you’re joking but I can see that happen.

 

[Xiao_Xi]

3 categories:

1. Software that works well with macOS such as Metal-optimized renderer or some task that requires a huge amount of VRAM. No UI needed. Just terminal.

2. Actually virtualizing macOS and use it as a remote desktop to do all your work on an Extreme SoC.

3. Accelerating local applications that are optimized for Apple Silicon chips through an Apple Silicon cloud.

1. How many workloads require metal-optimized software that requires a lot of VRAM and outperforms CUDA-optimized software in terms of performance/price?

2. AWS has something similar for small studios called Amazon Nimble Studio, but I don't know if it's very popular. How many interactive workloads would be more responsive on a remote Extreme-based macOS than on an Ultra-based macOS?

3. Is that possible? How does it work?

My argument this whole time is that it doesn't make economic sense to make an Extrem SoC for a niche product. Therefore, you need a bigger market for an Extrem SoC than just in a Mac Pro that will sell in the tens of thousands each year.

Nvidia has taken over the fast cloud computing market. What could Apple offer to take market share away from Nvidia?

 

[Boil]

Nvidia has taken over the fast cloud computing market. What could Apple offer to take market share away from Nvidia?

M3 Extreme in the Cloud running 3D/DCC apps via Vision Pro thin client...

 

[Boil]

But for less lag...

Vision Pro with M3 Extreme Mac Pro Cube...

Use M3 Extreme(s) in the Cloud for remote rendering...

 

[Xiao_Xi]

M3 Extreme in the Cloud running 3D/DCC apps via Vision Pro thin client...

How would it work? Would Apple force other developers to use its Extreme-based cloud computing, or would Apple's apps outperform any third-party apps?

 

[senttoschool]

1. How many workloads require metal-optimized software that requires a lot of VRAM and outperforms CUDA-optimized software in terms of performance/price?

I will ask the same question to you for an Extreme chip in a local Mac Pro.

2. AWS has something similar for small studios called Amazon Nimble Studio, but I don't know if it's very popular. How many interactive workloads would be more responsive on a remote Extreme-based macOS than on an Ultra-based macOS?

Amazon uses Mac Minis. They're mostly for remote software builds? I don't know. But Amazon does not have anything more than the base SoC. We're talking about Extreme SoCs.

3. Is that possible? How does it work?

Let's say you have a very large software project. You can build the software much faster with an Extreme SoC. On your local xCode, press the "Cloud Build" button and an Extreme SoC will build it for you in the cloud and then your local machine would download the build.

 

[Xiao_Xi]

I will ask the same question to you for an Extreme chip in a local Mac Pro.

While it makes sense that Mx Extreme's business viability is tied to cloud computing, I don't see the point of either. The more powerful the SoC, the less competitive advantage Apple Silicon has.

Amazon uses Mac Minis. They're mostly for remote software builds? I don't know. But Amazon does not have anything more than the base SoC. We're talking about Extreme SoCs.

Nimble Studio provides a collaborative environment for small studios using GPU workstations. Why would Amazon use Apple hardware for that purpose?

Let's say you have a very large software project. You can build the software much faster with an Extreme SoC. On your local xCode, press the "Cloud Build" button and an Extreme SoC will build it for you in the cloud and then your local machine would download the build.

How many Apple-specific software projects would take advantage of an SoC with more than 24 cores at compile time? Is LLVM so good that it can parallelize compilation on so many cores?

 

[jmho]

How many Apple-specific software projects would take advantage of an SoC with more than 24 cores at compile time? Is LLVM so good that it can parallelize compilation on so many cores?

It has more to do with the language.

Building C/C++ can be highly parallel because of the explicitly defined interfaces and self contained nature of most functions and classes.

Something like Swift is a lot harder to parallelize because the compiler is doing a lot more work implicitly type checking interfaces which means a lot more blocking and waiting for other dependencies to be built.

The LLVM backend that actually turns the intermediate representation into code is likely very easy to run in parallel.

I remember hearing that there are projects inside Apple that they will never convert to Swift just because the code base is so large that compilation would take way too long.

 

[senttoschool]

How many Apple-specific software projects would take advantage of an SoC with more than 24 cores at compile time? Is LLVM so good that it can parallelize compilation on so many cores?

It would allow someone sitting at a coffee shop using a Macbook Air 15" to compile like an Extreme SoC.

 

[jmho]

It would allow someone sitting at a coffee shop using a Macbook Air 15" to compile like an Extreme SoC.

Personally I think some kind of cloud based Xcode sounds awesome, but there is also the famous mail from the Epic lawsuit where someone in Apple asked Craig Federighi about cloud gaming and Craig almost bit his head off saying that Apple is all about providing "high performance local compute" and that streaming apps "might make sense for our competition, but not for Apple"

It's also quite reassuring that Craig said that though, because it hopefully means that Apple are all about "high performance local compute" and the current somewhat weak GPU situation is temporary.

 

[ondioline]

Is LLVM so good that it can parallelize compilation on so many cores?

As of this month yes, the linker has been rewritten to be highly parallel:

https://twitter.com/i/web/status/1665929739155177472

 

[Xiao_Xi]

Today Blender 3.6 has been released.

3.6 LTS — blender.org

Blender Foundation and the online developers community proudly present Blender 3.6 LTS!

www.blender.org

Apple Silicon GPUs now support NanoVDB for Metal, reducing memory usage of volumes.

Cycles: add NanoVDB support for Metal on Apple Silicon · 02c2970983

Contributed by Yulia Kuznetcova at Apple. NanoVDB is patched to give add address spaces required by Metal. We hope that in the future Metal will support the generic address space. For AMD and Intel this is currently not available since it causes a performance regression also on scenes without ...

projects.blender.org

 

[jmho]

Interesting that they only support NanoVDB on Apple GPUs.

 

[Xiao_Xi]

Could any of Intel's algorithms make it to macOS?

Intel's Latest Research for Graphics and Generative AI

Find out about Intel's latest research for graphics and generative AI.

www.intel.com