3D Rendering on Apple Silicon, CPU&GPU

[altaic]

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

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.

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

That first devtalk link is really interesting. 2x performance increase here, 1.4x there… and that’s without Metal 3 improvements. That would bring that 1:37 render down to 35 seconds or so, then w/ Metal 3 I imagine it’ll drop to sub 30 seconds or maybe even sub 20 seconds. It’ll be interesting to see what it takes to get within spitting distance of that 15 second Optix render.

Edit: It’s also quite interesting to see the large amount of effort the Apple software engineers are putting into discrete GPU support for the Metal Cycles renderer. Metal for Cycles requires at least macOS 12.3, so it’s not like this is a legacy thing.

 

[Drifter759]

That first devtalk link is really interesting. 2x performance increase here, 1.4x there… and that’s without Metal 3 improvements. That would bring that 1:37 render down to 35 seconds or so, then w/ Metal 3 I imagine it’ll drop to sub 30 seconds or maybe even sub 20 seconds. It’ll be interesting to see what it takes to get within spitting distance of that 15 second Optix render.

Edit: It’s also quite interesting to see the large amount of effort the Apple software engineers are putting into discrete GPU support for the Metal Cycles renderer. Metal for Cycles requires at least macOS 12.3, so it’s not like this is a legacy thing.

I don't think there's any way they get even close to Optix without hardware RT acceleration. Like others here, I'm pinning my hopes on the M3 generation. Having it on the higher-end M2s would be a nice surprise, but I'm not holding my breath.

I'll definitely buy one once they include it, though. Having to use any Windows laptop capable of doing 3D work away from a power outlet is pure misery - even if you're just editing documents or sending emails. And being able to do HDR stuff on a laptop would be cool. Not to mention color management being crappy on Windows and all the usual stuff.

 

[leman]

That first devtalk link is really interesting. 2x performance increase here, 1.4x there… and that’s without Metal 3 improvements. That would bring that 1:37 render down to 35 seconds or so, then w/ Metal 3 I imagine it’ll drop to sub 30 seconds or maybe even sub 20 seconds. It’ll be interesting to see what it takes to get within spitting distance of that 15 second Optix render.

Edit: It’s also quite interesting to see the large amount of effort the Apple software engineers are putting into discrete GPU support for the Metal Cycles renderer. Metal for Cycles requires at least macOS 12.3, so it’s not like this is a legacy thing.

I doubt that Metal 3 will do much for production raytracing performance. The big advantage Nvidia has is RT acceleration. Once Apple adds that, they should easily outperform Nvidia for renders as that's where Apple's advantage in cache sizes will become decisive.

 

[altaic]

I doubt that Metal 3 will do much for production raytracing performance. The big advantage Nvidia has is RT acceleration. Once Apple adds that, they should easily outperform Nvidia for renders as that's where Apple's advantage in cache sizes will become decisive.

Culling has been moved into the GPU. Prior, ray tracing required a lot of CPU breaks that essentially pause the GPU. From the WWDC presentations, I think Metal 3 will give a minimum 20% boost to ray tracing. The rest of the pipeline is much easier pickings, so I expect a lot of gains for Blender or any other renderer, given optimization effort. Blender should be considered a case study for ASi.

 

[leman]

Culling has been moved into the GPU.

Ah, did they? I missed that. Looking forward to seeing the results!

 

[altaic]

Just came across an interesting dev discussion about MoltenVK taking advantage of Metal 3. Looks promising.

 

[leman]

Just came across an interesting dev discussion about MoltenVK taking advantage of Metal 3. Looks promising.

For production rendering, the relevant bits is the improved RT pipeline (faster and better acceleration structure construction) - and maybe, not sure - the fact that some primitive data can be encoded in the acceleration structure directly. No idea if that can be taken advantage of in production renderers however.

 

[altaic]

For production rendering, the relevant bits is the improved RT pipeline (faster and better acceleration structure construction) - and maybe, not sure - the fact that some primitive data can be encoded in the acceleration structure directly. No idea if that can be taken advantage of in production renderers however.

My main takeaway was that Metal 3 more easily/efficiently maps to/from Vulkan (e.g. argument buffers and mesh shaders). Not too sure about RT at this stage.

 

[leman]

My main takeaway was that Metal 3 more easily/efficiently maps to/from Vulkan (e.g. argument buffers and mesh shaders). Not too sure about RT at this stage.

Sure, but those things are hardly of interest to production renderers which are usually optimised with specific hardware in mind. And I don't think there is a meaningful intersection of mesh shaders and RT in this domain. If you are generating geometry on the fly, it means you can't ray trace it efficiently using the standard pipeline — it is probably better to use custom intersection shaders for this purpose.

Fun fact: with Metal 3 Apple now provides the most comprehensive and consistent implementation of mesh shaders. Running Ventura and have a Mac that was made in the last 4-5 years? Then you can use mesh shaders without restrictions. Contrast it with Vulkan where mesh shaders are available on a whopping 4% of devices. Not sure how things are with DX12 since I couldn't find any info.

And they are fast too! I did some experiments with generating geometry and I can easily push around 130 million triangles per second. Mind, this is pure geometry generation, not culling, with a single mesh shader invocation producing 254 triangles. Will still need to compare this to pre-generated geometry to measure the overhead, but so far I really like mesh shaders. It's a very GPU way of doing things and feels like exposing how the hardware worked all along.

 

[Xiao_Xi]

Blender is improving its support on macOS and a metal backend for the Blender viewport is on the way. What about other programs?

Metal Backend

• This week the [command buffer] and [memory manager] for Metal has been reviewed and landed in master. There were some issues with missing license headers, but that will be taken care of.
• Next target will be to add Vertex Buffers, Uniform Buffers and Index Buffers. These are fairly straight forward.
• After that Batch and Shader precompilation will be added. This would eventually be the first time we should see pixels drawn by the Metal backend.

2022-07-04 Eevee/Viewport Module Meeting

Practical Info This is a weekly video chat meeting for planning and discussion of Blender Eevee/viewport module development. Any contributor (developer, UI/UX designer, writer, …) working on Eevee/viewport in Blender is welcome to join. For users and other interested parties, we ask to read...

devtalk.blender.org

Checking the progress of the Metal backend, I found this.

A number of features available in Blender require API functionality which is either not directly available in the Metal API, such as geometry shaders, or requires a higher degree of explicitness from the high-level renderer.

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

Has Metal 3 included those missing funcionalities?

 

[l0stl0rd]

Blender is improving its support on macOS and a metal backend for the Blender viewport is on the way. What about other programs?

2022-07-04 Eevee/Viewport Module Meeting

Practical Info This is a weekly video chat meeting for planning and discussion of Blender Eevee/viewport module development. Any contributor (developer, UI/UX designer, writer, …) working on Eevee/viewport in Blender is welcome to join. For users and other interested parties, we ask to read...

devtalk.blender.org

Checking the progress of the Metal backend, I found this.

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

Has Metal 3 included those missing funcionalities?

Metal 3 has meshshaders which should work I guess.

Transform your geometry with Metal mesh shaders - WWDC22 - Videos - Apple Developer

Meet Metal mesh shaders — a modern and flexible pipeline in Metal for GPU-driven geometry creation and processing. We'll explore how this...

developer.apple.com

 

[leman]

Has Metal 3 included those missing funcionalities?

I think mesh shaders can be used to implement much of this, but of course, it depends on the details. At any rate, not without significant engineering investment and who knows what the performance impact will be.

 

[BootLoxes]

Not sure how much longer I can keep waiting for my Zbrush to be native.

 

[sirio76]

Why? I’ve read several reviews and it looks plenty fast

 

[BootLoxes]

Why? I’ve read several reviews and it looks plenty fast

Battery life and efficiency

 

[sirio76]

Very weird reasons, once native it will be like 20% faster (that will not translate in 20% faster workflow) and surely will not consume less energy.
Seems just an excuse to not use new hardware/software, but that’s what people do here, tirelessly waiting for beta software to became mature or magical native version that will be 300%faster and consume no energy at all

 

[BootLoxes]

Very weird reasons, once native it will be like 20% faster (that will not translate in 20% faster workflow) and surely will not consume less energy.
Seems just an excuse to not use new hardware/software, but that’s what people do here, tirelessly waiting for beta software to became mature or magical native version that will be 300%faster and consume no energy at all

I am either getting a Mac Studio this year or building a new PC. Access to native versions of Zbrush and other 3D software play a role on which device I am prioritizing.

 

[vinegarshots]

I am either getting a Mac Studio this year or building a new PC. Access to native versions of Zbrush and other 3D software play a role on which device I am prioritizing.

If you work in 3D, no Mac will outperform a PC. If you care about performance, then it’s a no-brainer

 

[l0stl0rd]

If you work in 3D, no Mac will outperform a PC. If you care about performance, then it’s a no-brainer

Exactly and it is not even more efficient than a 3090 like some showed, especially at rendering (or at least at the moment).

Personally I thing even AMD GPUs are not an option.

 

[mi7chy]

With GPU price crashing it's putting a lot of pricing pressure on Mac options. If you're OS agnostic Nvidia is best bang for the buck with NIB 3080ti dipping below $1K. If you want portable solution mobile Nvidia is faster and more power efficient than Mac Studio. If you have a budget and/or want to build a hackintosh which is limited to AMD it's going for ~$630 NIB for 6800xt and still faster than $5K+ Mac Studio 64GPU.

GPU Blender BMW (lower time is better)
7.23s - Nvidia 3080ti (GPU OptiX Blender Linux Mint 20)
8.21s - Nvidia 3080ti (GPU OptiX Blender Windows 10)
13.13s - Nvidia 3080 mobile (GPU OptiX Blender)
16.39s - Nvidia 3060 70W mobile (GPU OptiX Blender 3.0)
18.84s - AMD 6900xt (GPU HIP Blender 3.2)
24.04s - AMD 6800 (GPU HIP Blender 3.2)
29s - 2070 Super (GPU OptiX)
34s - M1 Ultra 20CPU 64GPU (GPU Metal Blender 3.1)
37s - M1 Ultra 20CPU 48GPU (GPU Metal Blender 3.1)
42.79s - M1 Max 32GPU (GPU Metal Blender 3.1 alpha)
48s - M1 Max 24GPU (GPU Metal Blender 3.1 alpha + patch)
51s - Nvidia 2070 Super (GPU CUDA)
1m18.34s - M1 Pro 16GPU (GPU Metal Blender 3.1 alpha + patch)
1m43s - M1 Ultra 20CPU 64GPU (CPU Blender 3.1)
1m50s - M1 Ultra 20CPU 48GPU (CPU Blender 3.1)
2m0.04s - Mac Mini M1 (GPU Metal Blender 3.1 alpha + patch)
2m48.03s - MBA M1 7GPU (GPU Metal Blender 3.1 alpha)

 

[BootLoxes]

If you work in 3D, no Mac will outperform a PC. If you care about performance, then it’s a no-brainer

I use the device for more than just 3D.

 

[ader42]

If you don’t spend all your time rendering then you don’t need a PC.
If you care about more than rendering speed then stick with Mac it’s a no-brainer.
The current rosetta ZBrush works great for me with tens of millions of polygons…
Even rendering in ZBrush is nice for me, an A2 300DPI poster I recently made (nice Giclee print) of a scene with 40 million polygons only took several minutes to render on maximum quality settings… (I rendered at 8192x8192)

 

[eric237]

Keyshot 11.2 comes with a beta version of Apple Silicon support

KeyShot 11.2 Now Available - Includes Apple Silicon Support, Workflow Features, and Improvements

For KeyShot 11.2, we introduce updates to the latest features, along with new support for Mac with Apple silicon, keyframe animation improvements, and additional workflow updates.

blog.keyshot.com

 

[BootLoxes]

If you don’t spend all your time rendering then you don’t need a PC.
If you care about more than rendering speed then stick with Mac it’s a no-brainer.
The current rosetta ZBrush works great for me with tens of millions of polygons…
Even rendering in ZBrush is nice for me, an A2 300DPI poster I recently made (nice Giclee print) of a scene with 40 million polygons only took several minutes to render on maximum quality settings… (I rendered at 8192x8192)

Yeah, for my needs the M1 max would be suffice. I plan to wait until the M3 variants before getting a studio.

Though the new ryzen cpu is just around the corner and that is tempting.

 

[Xiao_Xi]

Apple seems to be getting more involved in Blender. There is a Metal back-end for MNEE on the way.

Cycles Manifold Next Event Estimation

Short term: - [x] Initial implementation [D13533: Adding Manifold Next Event Estimation Sampling Technique](https://archive.blender.org/developer/D13533) - [x] Solve potential issues with motion blur - [x] Walk without relying on UV maps for consistent directions - [x] Metal GPU support - [ ] Sep...

developer.blender.org