3D Rendering on Apple Silicon, CPU&GPU

[JanErik75]

The vanilla M2 was never going to be a 3D artist machine. Obviously I'd have loved to have seen RT cores, but it would also be kinda weird for them to release them on a MacBook Air.

There is a WWDC session on Thursday though called "Maximise your Metal ray tracing performance", so we'll see just how "hardware friendly" that is.

I'll only start freaking out when we see the Mac Pro and it doesn't have RT cores.

The Pro, Max and Ultra chips are all based on the basic M1 chip. Meaning what we see with the vanilla M2 is what we'll get with the M2 Pro, Max and Ultra, in the MacBook Pro, Mini, and Studio. Although with more cores of what's already there. No additional brand new cores.

Looks like we'll have to wait for the M3 or M4 or whatever for that. Although I'm not holding my breath. It seems that Apple is not going head-to-head with NVIDIA here, because Apple doesn't have a good enough solution to compete with OptiX and CUDA yet. And they won't have it for a long time, because the aforementioned technology is so mature.

If and when the ASi Mac Pro gets RT cores and the other good stuff, that's when the trickle down effect starts to happen on the other Macs.

Oh and one more thing, I predict the ASi Mac Pro chip won't even be part of the M family. It will have a completely new, separate naming scheme. Just like the Intel Xeon on Mac Pros versus Intel Core iX on the other Macs.

 

[neinjohn]

The Pro, Max and Ultra chips are all based on the basic M1 chip. Meaning what we see with the vanilla M2 is what we'll get with the M2 Pro, Max and Ultra, in the MacBook Pro, Mini, and Studio. Although with more cores of what's already there. No additional brand new cores.

The M1 Pro, Max and Ultra have ProRes support meanwhile the M1 doesn't. Apple silicon has some modularity to it be it on adding or not expanding (neural network cores are the same number everywhere).

We will have to wait and see what the A16 on the iPhone Pro models brings, or not, on RT specific hardware.

 

[jmho]

Apple's developer site's "What's new in Metal 3" page says:

New Ray Tracing features​

The latest advancements in Metal Ray Tracing mean less GPU time is spent building acceleration structures, work like culling can move to the GPU to reduce CPU overhead, and both intersection and shading can be optimized with direct access to primitive data.

That sounds pretty nice and quite hardware-friendly.

 

[altaic]

Apple's developer site's "What's new in Metal 3" page says:



That sounds pretty nice and quite hardware-friendly.

The WWDC22 Platforms State of the Union video talks about Metal 3 starting at ~52 minutes in. Unfortunately they don’t really talk about ray tracing, though.

 

[jmho]

There's also a new Raytracing sample, but it's hybrid rendering, so more gaming focused. I guess the first place we'll see just how good Metal 3 is for pure path tracing is in Blender.

Also I'm not 100% sure, but looking at the Metal 3 shader language spec it does look like they've got partial ulong / 64 bit atomic support. Will be interesting to see if that helps Epic port UE5's nanite to macOS.

 

[Xiao_Xi]

Blender has just released its latest version, any improvements for macOS?

3.2 — blender.org

Blender Foundation and the online developers community are proud to present Blender 3.2!

www.blender.org

 

[exoticSpice]

Boost performance with MetalFX Upscaling - WWDC22 - Videos - Apple Developer

Discover MetalFX, a new API that provides platform optimized graphics effects for Metal applications. With MetalFX Upscaling, your...

developer.apple.com

This very good. One of the huge additions in Metal 3

 

[altaic]

Blender has just released its latest version, any improvements for macOS?

3.2 — blender.org

Blender Foundation and the online developers community are proud to present Blender 3.2!

www.blender.org

I searched the release notes and didn’t find anything specific to Metal or Apple Silicon (or macOS, for that matter) improvements. I’d wager the Apple engineers on the project were preoccupied leading up to WWDC.

I hope they come back with a flurry of commits adding Metal 3 support! Metal 3 looks to have moved more of the rendering pipeline to the GPU, so the bottlenecks we saw with scaling (M1 Pro -> M1 Max -> M1 Ultra) may be mitigated.

Edit: In the Target and optimize GPU binaries with Metal 3 WWDC22 presentation, they talk a bit about Blender development starting at the 10 minute mark.

 

[mi7chy]

Blender has just released its latest version, any improvements for macOS?

3.2 — blender.org

Blender Foundation and the online developers community are proud to present Blender 3.2!

www.blender.org

Biggest change with 3.2 is Linux AMD GPU HIP support. Decided to run it anyway on MacOS and it's gotten 19% slower somewhere since 3.1 alpha. Just in time for M2 to make up for that difference.

2m48.03s - MBA M1 7GPU (GPU Metal Blender 3.1 alpha)
3m19.97s - MBA M1 7GPU (GPU Metal Blender 3.2)

 

[orbitalpunk]

Hi guys, I could use some advice, especially those with Ultras. I'm a Maya/Vray user and currently tried Maya on my M1 Mini and my custom built gaming PC 8900/1080ti. Im looking to upgrade one with about $4k to spend. I could order a Ultra 48gpu core or build a PC. I love working on my Mac more but seeing how Vray in maya uses native RTX GPU rendering seems just so fast. Plus I use Phoenix for simulations. Any other Maya/Vray users here do comparisons between a PC build and a Ultra? How far are they apart in performance? Anyone have a had a similar dilemma? Help.

 

[sirio76]

If you want to use Vray on Mac you only have the CPU option so if you want to use the GPU engine you’ll be forced to use a Windows system, or use another engine that support MacOS (Redshift, Octane etc).
That being said even the Vray CPU engine is plenty fast, even faster that the GPU engine depending on what you are doing.

 

[l0stl0rd]

Hi guys, I could use some advice, especially those with Ultras. I'm a Maya/Vray user and currently tried Maya on my M1 Mini and my custom built gaming PC 8900/1080ti. Im looking to upgrade one with about $4k to spend. I could order a Ultra 48gpu core or build a PC. I love working on my Mac more but seeing how Vray in maya uses native RTX GPU rendering seems just so fast. Plus I use Phoenix for simulations. Any other Maya/Vray users here do comparisons between a PC build and a Ultra? How far are they apart in performance? Anyone have a had a similar dilemma? Help.

Honestly if you work professionally in 3D just build a new Windows machine.

The performance gap is still to big if you ask me. Unless you want to render in the cloud.

 

[terminator-jq]

Hi guys, I could use some advice, especially those with Ultras. I'm a Maya/Vray user and currently tried Maya on my M1 Mini and my custom built gaming PC 8900/1080ti. Im looking to upgrade one with about $4k to spend. I could order a Ultra 48gpu core or build a PC. I love working on my Mac more but seeing how Vray in maya uses native RTX GPU rendering seems just so fast. Plus I use Phoenix for simulations. Any other Maya/Vray users here do comparisons between a PC build and a Ultra? How far are they apart in performance? Anyone have a had a similar dilemma? Help.

Even as a Mac fan, I have to admit the Maya and Vray combo will perform much better on a Windows machine with an Nvidia GPU. We are still early in this transition so maybe the software companies will find ways to bring Apple Silicon closer to Nvidia level performance but if I had to guess, Apple will not be able to compete on that level (on the GPU side) until at least the M3 generation assuming they add hardware ray tracing.

With $4k to spend, it might be worth waiting to see what Apple does for the Mac Pro. But even then, Nvidia is just much better established in the GPU rendering game.

I’d really like to see Apple step it up on those fronts.

 

[ader42]

In some ways it depends how much time you spend rendering versus how much time you spend not-rendering.

I spend most of my daytime in ZBrush on an M1 Max and it's beyond great. If I was doing more animation stuff again in C4D (which would mean much more rendering time) I'd probably just let it render coolly and quietly overnight.

I don't think the Ultra are a good buy unless you want the 20 cpu cores - I wouldn't buy one for the gpu performance.

 

[mi7chy]

If you charge your client by the hour use a Mac otherwise use Windows with Nvidia dGPU if you charge a flat fee.

 

[leman]

If you charge your client by the hour use a Mac otherwise use Windows with Nvidia dGPU if you charge a flat fee.

That was actually funny Didn't know you have it in you

 

[Xiao_Xi]

Hi guys, I could use some advice, especially those with Ultras. I'm a Maya/Vray user and currently tried Maya on my M1 Mini and my custom built gaming PC 8900/1080ti. Im looking to upgrade one with about $4k to spend. I could order a Ultra 48gpu core or build a PC. I love working on my Mac more but seeing how Vray in maya uses native RTX GPU rendering seems just so fast. Plus I use Phoenix for simulations. Any other Maya/Vray users here do comparisons between a PC build and a Ultra? How far are they apart in performance? Anyone have a had a similar dilemma? Help.

ONEofTECH has compared a 64-core M1 Ultra Mac Studio to a custom-built PC with an Nvidia RTX 3090.

Among other things, he compared them in 3D rendering using Blender 3.1.

 

[jujoje]

Hi guys, I could use some advice, especially those with Ultras. I'm a Maya/Vray user and currently tried Maya on my M1 Mini and my custom built gaming PC 8900/1080ti. Im looking to upgrade one with about $4k to spend. I could order a Ultra 48gpu core or build a PC. I love working on my Mac more but seeing how Vray in maya uses native RTX GPU rendering seems just so fast. Plus I use Phoenix for simulations. Any other Maya/Vray users here do comparisons between a PC build and a Ultra? How far are they apart in performance? Anyone have a had a similar dilemma? Help.

Much as everyone else is saying, for Maya and Vray a pc is the way to go, particularly for GPU rendering.

I don’t think Autodesk have officially announced Apple Silicon support (last I checked they were noncommittal), so you’d be loosing around 10-30% performance (based off native Houdini vs Rosetta Houdini; so not apples to apples (heh), but probably a reasonable assumption).

That said, having a pretty much totally silent machine is awesome. My iMac Pro sounds like a rocket taking off under any load but the Mac studio was so silent I forgot it was rendering.

ONEofTECH has compared a 64-core M1 Ultra Mac Studio to a custom-built PC with an Nvidia RTX 3090.

Among other things, he compared them in 3D rendering using Blender 3.1.
View attachment 2019177

Haven’t watched the video, but gotta ask, wtf is it with custom pcs and rgb technicolor vomit?

 

[jujoje]

Oh and fwiw; i did have a base mac studio but ended up returning it due to gpu slowness.

Was more likely a software bug than a hardware limitation, but for that amount of money I’m going with what it works for now rather than the hypothetical future when everything has been optimised.

 

[terminator-jq]

ONEofTECH has compared a 64-core M1 Ultra Mac Studio to a custom-built PC with an Nvidia RTX 3090.

Among other things, he compared them in 3D rendering using Blender 3.1.
View attachment 2019177

Blender is an interesting case. It’s still pretty early in it’s Apple Silicon development but it sounds like they have already found a lot of room for optimization. In fact an Apple engineer said they were getting render times half as long in a test build.

It will be very interesting to see how the GPU performance increases as things are more optimized for A.S

 

[altaic]

In fact an Apple engineer said they were getting render times half as long in a test build.

Link please? I wonder if that test build is on macOS 13 and incorporates Metal 3 improvements.

 

[orbitalpunk]

ONEofTECH has compared a 64-core M1 Ultra Mac Studio to a custom-built PC with an Nvidia RTX 3090.

Among other things, he compared them in 3D rendering using Blender 3.1.

Great video. Thanks for sharing. It really shows the advantage in my case going with a PC when you take his Blender example, the Studio CPU render was like 4+ minutes vs the PC GPU being 15 seconds since Maya/Vray only has CPU options for Mac.

I do feel Autodesk/Vray will support Apple Silicon in a big way especially after the Mac Pro is released in the future. The reason why I feel this way is because 2 weeks ago Vray released Phoenix for the Mac which they never did before so I thought that showed a huge commitment to the Mac. But still, CUDA is still so much more mature over Metal so it will be a while. Perhaps next upgrade will go to the Mac. Fingers crossed.

Can anyone recommend a good PC case with Apple asthetics in ITX format? I'd been using a Ncase M1 but it can't handle many of the larger videos cards so need to change it.

 

[jujoje]

In fact an Apple engineer said they were getting render times half as long in a test build.

If I remember correctly I think the engineer was saying that they were getting half as long render times, but the images weren't matching the reference renders.

Not to say that there isn't still a lot of improvement, but I don't think that render times are going to jump that much in the next build.

 

[Xiao_Xi]

Link please?

In March, one of the Apple engineers working on the metal backend for Blender's Cycles wrote:

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.

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

Blender Archive - developer.blender.org

developer.blender.org

Eevee, Blender's other renderer, will also have a Metal backend, most likely in Blender 3.4.

We expect that the Metal back-end could slip to another release

2022-06-13 Eevee/Viewport Module Meeting

Attendees: Clement Foucault Brecht van Lommel Omar Emara Jeroen Bakker Dalai Felinto Release 3.2 [T98708] should be added to a corrective release. It fixes an issue on platforms That supports OpenGL 4.3, but not the conservative depth extension. [T98620] is fixed in master, but was to late...

devtalk.blender.org

Metal Viewport

**Status:** `?` Initial stages of development --- **Description** Bringing Metal support to the Blender viewport for optimal performance on macOS. This will be implemented as a new GPU Backend within Blender's GPU Module. **GHOST** * - [x] Encapsulate use of OpenGL within GHOST_ContextCGL * -...

developer.blender.org

 

[mi7chy]

Link please? I wonder if that test build is on macOS 13 and incorporates Metal 3 improvements.

Half as long sounds like marketing speak for this slide which shows rendering at a lower resolution then using upscaling with MetalFX which is similar in idea to Nvidia DLSS, AMD FSR/RSR, etc.

https://developer.apple.com/videos/play/wwdc2022/10103

Native resolution will always be best since there's no upscaling artifact but for certain scenarios better performance is worth the quality trade like Unreal Engine 5 that has poor frame rate on my 3060 laptop without it.