3D Rendering on Apple Silicon, CPU&GPU

[singhs.apps]

Thanks for the benchmarks. Most likely is Rosetta bottlenecking performance as Nathan said. Looks like Redshift is the only serious contender for rendering right now.

I would really love to get an M1 version of Arnold above anything, I'm sure it would fly on the new chips. I've been using it today on a Mac Pro and it's just such a great looking, robust and featured renderer. Assets that I've used in Redshift like HDRIs and PBR textures just look that much more convincing in Arnold.

Both Arnold and Octane render very realistic looking images out of the box because they are unbiased renderers.
Sure you can get good results out of unbiased ones too, but in my experience, these two looked more convincing without any tweaking.
Arnold is a very robust, world class renderer. GPU mode can be flakey at times though, but then so is Octane.
On CPU it is very powerful and scales very well.

Redshift’s defaults look a wee bit artificial, but it has been the most stable amongst the GPU renderers I have used.

 

[vel0city]

Just a thought, probably nothing, but I sincerely hope there's nothing significant holding back all these renderers from being ported to AS. So far we have Redshift (in beta) on Metal running native on AS and, er, that's it? Octane plugin is still on Rosetta, is that correct? Has any other renderer been ported, either GPU or CPU?

I know that renderers are massively complicated and also niche. But you would think more than one would be available a year into the transition.

 

[vel0city]

Both Arnold and Octane render very realistic looking images out of the box because they are unbiased renderers.
Sure you can get good results out of unbiased ones too, but in my experience, these two looked more convincing without any tweaking.
Arnold is a very robust, world class renderer. GPU mode can be flakey at times though, but then so is Octane.
On CPU it is very powerful and scales very well.

Redshift’s defaults look a wee bit artificial, but it has been the most stable amongst the GPU renderers I have used.

Agree with everything you said - I absolutely love Octane's look, its lighting handling is beautiful, but the crashes are too frequent to make it worth it. You only have to look at the node editor to bring your entire system down. It's a real shame how unreliable it is.

Redshift's speed with Octane's look and Arnold's scalability would be just about the ultimate renderer.

Anyone tried Unity or Unreal on a new system?

 

[Macintosh IIcx]

Just a thought, probably nothing, but I sincerely hope there's nothing significant holding back all these renderers from being ported to AS. So far we have Redshift (in beta) on Metal running native on AS and, er, that's it? Octane plugin is still on Rosetta, is that correct? Has any other renderer been ported, either GPU or CPU?

I know that renderers are massively complicated and also niche. But you would think more than one would be available a year into the transition.

Well, Octane standalone is fully Metal, not sure about the plug-ins but then only Cinema 4D and Blender are ported to ARM so that complicates things. In any event, Apple have helped both Otoy Octane and Redshift getting Metal able to do 3d rendering and Apple themselves are making Metal for Blender, so things are moving, albeit slowly.

 

[AirThis]

Blender 3.1 Alpha is available for download. Has anyone tested if Cycles has already GPU support in this Alpha version?

I'd do it myself but the delivery date for my M1 Max is in 4 weeks.

https://builder.blender.org/download/daily/blender-3.1.0-alpha+master.39c11c03d0d7-darwin.arm64-release.dmg

 

[sirio76]

Has any other renderer been ported, either GPU or CPU?

As said native Vray is coming in a matter of weeks.
Writing a renderer is not complicated, having an efficient/stable/feature reach engine is another story.
Coding a GPU renderer is more challenging than a CPU renderer and is overall more difficult to debug and get stability, for example a driver update can easily mess up features/performance, something you never experience on CPU.

 

[jmho]

Blender 3.1 Alpha is available for download. Has anyone tested if Cycles has already GPU support in this Alpha version?

I'd do it myself but the delivery date for my M1 Max is in 4 weeks.

https://builder.blender.org/download/daily/blender-3.1.0-alpha+master.39c11c03d0d7-darwin.arm64-release.dmg

Doesn't look like it, the GPU compute option is still greyed out.

Plus I don't think Apple have committed anything after their initial commit which was just to get the shaders to compile. There's still a lot of work left to do on both cycles and the MoltenVK viewport.

 

[syncopy]

would you be willing to benchmark keyshot for us?

How to Use the KeyShot Benchmark Tool

The KeyShot Benchmark is a tool within the free KeyShot Viewer. With this benchmark, you can test your CPU and/or GPU to evaluate their performance based on a reference system.

blog.keyshot.com

Just for reference, Keyshot on M1 Max 32 Core with 32GB of memory renders slower over time for animations than my 2019 Core i9 5500M MBP. Rosetta is definitely holding it back quite a bit. Keyshot on my M1 can't even open a 64MB complex animation without crashing back to the desktop, whereas it takes some time but still loads properly on the Intel machine. For still renders or small projects, this probably isn't much of an issue.

Regardless, I doubt Keyshot is going to be optimized for M1 for some time. It took an entire version (9 -> 10) to simply support running with Rosetta. Native CPU is probably their next "priority", and then maybe we'll see native GPU rendering by the time Apple is releasing M8 Max chips. I can't justify using Keyshot anymore if Luxion is going to move at a glacial pace and give us nothing. I've already started watching Redshift tutorials.

 

[FurtherForm]

Doesn't look like it, the GPU compute option is still greyed out.

Plus I don't think Apple have committed anything after their initial commit which was just to get the shaders to compile. There's still a lot of work left to do on both cycles and the MoltenVK viewport.

Apple Blender dev mentioned upcoming work with cycles two weeks ago:

Is there interest in supporting a Metal version of Cycles for MacOS?

When will the first test build available? I’m eager to test it with the M1

devtalk.blender.org

"We’ll bring a feature branch up with a fully Metal-enabled version over the coming weeks, as we tidy up the code and break it up into atomic commits that make sense for code review.

The first commit was the most invasive, touching all function signatures that took an undecorated pointer.

I expect our next commits will add the metal_compat.h and they remainder of the kernel changes needed to compile as MSL, along with the main kernel entry point wrappers. Then we’ll add the host side device."

I'd wager that it'll be well in to Q1/22 until we get well working patches for Cycles/Metal @ Blender.

 

[Rashy]

Blender enthusiast here. I was very delighted as soon I heard about the announcement of Apple supporting the Development fun as full Patron, just a few days before the release of those amazing MBPs. Right now I am running 2.93 on my 2017 iMac and 2015 MBP, you can imagine how painfully slow it is (and I barely use Cycles, only Eevee).

M1 Pro + Blender 3.1 sound like a promising upgrade for me to go in a few months. No more need for me to even consider the Dell XPS 15.

 

[nathan_reilly]

Is Redshift really a viable alternative to Keyshot? I really enjoy Keyshot and have used it since version 2.

 

[singhs.apps]

I suspect we will see one or two production renderers running natively on Apple silicon by the time the new Mac Pro arrives next year. Apple would love to display the muscle of an Uber version of their SOC.

3D is where Apple is weakest - software support wise - on the M1 systems

 

[hefeglass]

Blender 3.1 Alpha is available for download. Has anyone tested if Cycles has already GPU support in this Alpha version?

I'd do it myself but the delivery date for my M1 Max is in 4 weeks.

https://builder.blender.org/download/daily/blender-3.1.0-alpha+master.39c11c03d0d7-darwin.arm64-release.dmg

yes I tried it..its still not using the GPU yet.

 

[Boil]

I suspect we will see one or two production renderers running natively on Apple silicon by the time the new Mac Pro arrives next year. Apple would love to display the muscle of an Uber version of their SOC.

3D is where Apple is weakest - software support wise - on the M1 systems

Hoping to see demos of the new M1 Max Duo & M1 Max Quadro powered Mac Pro products (32" iMac Pro / Mac Pro Cube / Mac Pro) & assorted Metalified DCC software come WWDC 2022...!

 

[singhs.apps]

Is Redshift really a viable alternative to Keyshot? I really enjoy Keyshot and have used it since version 2.

Depends on the workflow of the host app that you will have to use if you plan to try redshift.
Render quality wise it’s competitive.

 

[nathan_reilly]

Hoping to see demos of the new M1 Max Duo & M1 Max Quadro powered Mac Pro products (32" iMac Pro / Mac Pro Cube / Mac Pro) & assorted Metalified DCC software come WWDC 2022...!

I've been wondering this myself. How well does the M1 architecture suit itself to being "doubled" or "quadrupled" in the sense that you could stick two xeons in my Mac Pro or just 1 depending on your budget.

 

[diamond.g]

I've been wondering this myself. How well does the M1 architecture suit itself to being "doubled" or "quadrupled" in the sense that you could stick two xeons in my Mac Pro or just 1 depending on your budget.

Wouldn't it depend on how much latency is introduced moving the ram off SOC?

 

[quarkysg]

I've been wondering this myself. How well does the M1 architecture suit itself to being "doubled" or "quadrupled" in the sense that you could stick two xeons in my Mac Pro or just 1 depending on your budget.

Does the current Mac Pro support 2 CPUs? According to WikiPedia, only the 1st gen Mac Pro sports a dual CPU design, as well as the Power Mac G5.

I don't think current macOS support NUMA tho.

Apple most likely will stay with a single SoC design and stitch multiple M1 Max together via a high speed interconnect fabric to take care of cache coherencies between all the M1 Maxs chiplets. Probably faster than two M1 Max as separate SoCs tied together externally and requiring macOS to support NUMA.

 

[ader42]

What about CrowdRender in Blender - to have multiple M1 Macs working on a single render? That is what I have been considering, maybe a MBP for portability with a few M1 Minis on the desktop to plug in to assist with final rendering.

Does anyone know if CrowdRender works with both CPU and GPU in Blender?

Do any other render engines support multiple Macs for a single render (a still image not an animation)?

 

[Macintosh IIcx]

What about CrowdRender in Blender - to have multiple M1 Macs working on a single render? That is what I have been considering, maybe a MBP for portability with a few M1 Minis on the desktop to plug in to assist with final rendering.

Does anyone know if CrowdRender works with both CPU and GPU in Blender?

Do any other render engines support multiple Macs for a single render (a still image not an animation)?

To be honest, I’m starting to think it might make more sense to have a GPU solution local that it fast enough for look development and then hand off the finished render job to a cloud rendering service. Not sure if the M1 Max 32 GPU would be good enough for look dev, but I guess that depends on complexity and the render.
Octane might be the best option if one want to go in that direction, but then Octane might not be the best option of animation renderings.

 

[singhs.apps]

Wondering aloud:

Considering that macOS currently doesn’t support more than 64 threads and AS doesn’t have hyper threading (essentially one thread per core ), it still leaves Apple enough headroom to max out its Mac Pro AS system (instead of the current - and most likely - 40 core speculation) ?

Potential combinations for a M1 Mac Pro max :

14 firestorm + 2 ice storms x 4 SOCs = 64 cores (56 firestorm + 8 icestorm )

12 firestorm + 4 ice storms x 4 SOCs = 64 cores (48 firestorm + 16 icestorm ) = 16 LP cores are a waste in the Mac Pro IMO.

12 firestorm + 4 ice storms x 2 SOCs + 16 firestorm x 2 SOCs = 64 cores ( 56 firestorm + 8 icestorm )

16 firestorm x 4 SOCs = 64 cores ( full fat 64 firestorm ) = who needs LP cores in wall powered workstations



Of course I am pulling this out of thin air and have no clue about chip designs …but hey… why not ?

 

[quarkysg]

Wondering aloud:

Considering that macOS currently doesn’t support more than 64 threads and AS doesn’t have hyper threading (essentially one thread per core ), it still leaves Apple enough headroom to max out its Mac Pro AS system (instead of the current - and most likely - 40 core speculation) ?

Potential combinations for a M1 Mac Pro max :

14 firestorm + 2 ice storms x 4 SOCs = 64 cores (56 firestorm + 8 icestorm )

12 firestorm + 4 ice storms x 4 SOCs = 64 cores (48 firestorm + 16 icestorm ) = 16 LP cores are a waste in the Mac Pro IMO.

12 firestorm + 4 ice storms x 2 SOCs + 16 firestorm x 2 SOCs = 64 cores ( 56 firestorm + 8 icestorm )

16 firestorm x 4 SOCs = 64 cores ( full fat 64 firestorm ) = who needs LP cores on wall powered workstations



Of course I am pulling this out of thin air and have no clue about chip designs …but hey… why not ?

Feeding those cores with data to process will be a tough challenge. Without a dramatic increase in L2 and SLC plus RAM bandwidth, I would think most of those cores will be waiting for data a lot. I suspect existing CPUs with high core count are actually data starved.

 

[singhs.apps]

Feeding those cores with data to process will be a tough challenge. Without a dramatic increase in L2 and SLC plus RAM bandwidth, I would think most of those cores will be waiting for data a lot. I suspect existing CPUs with high core count are actually data starved.

Would the 400 GBps of the current m1 max’s also be a bottle neck ?

Be that as it may, Apple has shown inclination towards GPU compute as also dedicated encoders/decoders on the SOC so it’s entirely possible it will try and focus more on these areas.

 

[quarkysg]

Would the 400 GBps of the current m1 max’s also be a bottle neck ?

Be that as it may, Apple has shown inclination towards GPU compute as also dedicated encoders/decoders on the SOC so it’s entirely possible it will try and focus more on these areas.

According to Anandtech’s article, the M1 Max CPU clusters tops out at 250GB/s. Although super impressive, it shows that the internal fabric may be bandwidth limited. It could be by design as GPUs also need bandwidth.

A single M1 core on the other hand could saturate the entire RAM bandwidth, although only at 60GB/s. So I think Apple is balancing the cost of their design. It’ll probably be too expensive to go high core counts. I do not know what’s the magic number tho. ?

 

[ponzicoinbro]

Does Luxmark have a native version?