3D Rendering on Apple Silicon, CPU&GPU

[blackquartz]

I think the M1 and M2 3D rendering capabilities are very constraining, I do like the experience of using a SoC, which in itself as a mobile solution its outstanding but I want to believe Apple will respect this customer segment enough to present a Desktop oriented solution not considering performance per watt but real unconstrained raw power.

On the other side its great to see the industry doing the effort to adopt AS technology and I do really hope Apple follows trough this effort presenting new GPU oriented hardware.

 

[Romain_H]

Harsh truth is any M1 or M2 is just a SoC made for portable devices, which is impressive in many ways but they wont excel in intense GPU tasks.

Which is not to say I want the Mac to stay this way, I hope we get custom desktop processors, not caring about performance por watt, in mobile I understand that. But for our industry, we deserve raw power.

Indeed. Will be interesting to see what they come up with to address this. Unfortunately, whatever it's gonna be, it's almost certain that it will outprice me by far. The cMP times seem to be over

 

[jujoje]

Some pretty reasonable figures for the M2 studio compared to the M1 on the Apple website, with the it having more than double the performance in Octane (although ffs Apple, do proper charts).

Also nice to see the Moana scene taking full advantage of the memory. Still no raytracing though...

 

[Xiao_Xi]

Why does the performance increase in M2 Ultra vary so much from application to application?

Mac Studio

The new Mac Studio. Supercharged by the M2 Max and M2 Ultra chip. Extensive connectivity. Stunningly compact.

www.apple.com

 

[aytan]

192 GB unified memory is huge. Even base model M2 Ultra 60 GPU cores, it could nearly double the M1 Ultra in rendering+24 cores CPU is great. Hope to see as soon as some benchmarks. Anyway I will upgrade my M1 Ultra to Base M2 Ultra 128 GB 2 TB SSD when it is available for me

 

[aytan]

Why does the performance increase in M2 Ultra vary so much from application to application?

View attachment 2212840

View attachment 2212841

Mac Studio

The new Mac Studio. Supercharged by the M2 Max and M2 Ultra chip. Extensive connectivity. Stunningly compact.

www.apple.com

Apple and Otoy works for a long time for AS chips. Even Otoy is the first brand which uses AS for 3D. Also Apple could invest to this .

 

[aytan]

I expect 3070 ti or 3080 level rendering performance from M2 Ultra, which is just fine for me. I believe anything equal to 3080 for basic 3D rendering is ok especially Mac Studio form factor and such low power usage.

 

[Xiao_Xi]

On Wednesday:

Your guide to Metal ray tracing​

Discover how you can enhance the visual quality of your games and apps with Metal ray tracing. We'll take you through the fundamentals of the Metal ray tracing API. Explore the latest enhancements and techniques that will enable you to create larger and more complex scenes, reduce memory usage and build times, and efficiently render visual content like hair and fur.

On Thursday:

Optimize GPU renderers with Metal​

Discover how to optimize your GPU renderer using the latest Metal features and best practices. We'll show you how to use function specialization and parallel shader compilation to maintain responsive authoring workflows and the fastest rendering speeds, and help you tune your compute shaders for optimal performance.

WWDC24

Join the worldwide developer community online for a week of technology and creativity.

developer.apple.com

 

[aytan]

On Wednesday:

On Thursday:

WWDC24

Join the worldwide developer community online for a week of technology and creativity.

developer.apple.com

That is interesting

 

[jujoje]

Why does the performance increase in M2 Ultra vary so much from application to application?

Maya viewports not the greatest, plus I'm pretty sure it's OpenGL running on Metal rather than anything native. TBH I'm not sure why they'd use it as a benchmark other than for the Autodesk brand name; pretty sure other apps are more representative.

Speaking of 3D performance and iffy tests, was initially impressed by the 3D simulation being 3.4x faster than a 28 core Xeon (on the Mac Pro page). Then read what they tested for that figure:

Tested with Houdini FX 19.5.534 by Voronoi fracturing a large geometry and simulating the destruction effect using rigid body dynamics. Mac Pro systems tested with an attached 5K display. Performance tests are conducted using specific computer systems and reflect the approximate performance of Mac Pro.

Doesn't sound like the most taxing of tests CPU wise (also what does apple consider large geometry). Still I guess the performance increase is pretty nice.

 

[jujoje]

On Wednesday:

On Thursday:

WWDC24

Join the worldwide developer community online for a week of technology and creativity.

developer.apple.com

Also if Apple are getting behind MaterialX, that's going to be pretty big for 3D Renderers - author one set of materials for Octane / Karma and AR would be a pretty nice workflow:

Explore the USD ecosystem​

Discover the latest updates to Universal Scene Description (USD) on Apple platforms and learn how you can deliver great 3D content for your apps, games, and websites. Get to know USD for visionOS, explore MaterialX shaders and color management, and find out about some of the other improvements to the USD ecosystem.

 

[aytan]

Maya viewports not the greatest, plus I'm pretty sure it's OpenGL running on Metal rather than anything native. TBH I'm not sure why they'd use it as a benchmark other than for the Autodesk brand name; pretty sure other apps are more representative.

Speaking of 3D performance and iffy tests, was initially impressed by the 3D simulation being 3.4x faster than a 28 core Xeon (on the Mac Pro page). Then read what they tested for that figure:



Doesn't sound like the most taxing of tests CPU wise (also what does apple consider large geometry). Still I guess the performance increase is pretty nice.

I agreed with you, even %25 - %30 performance gain is pretty good for M2 Ultra. Honestly I have a hope new MacPro could use last gen AMD GPU's but no chance )). All you can get is 76 GPU cores. There is no mention even AMD 6000 series GPU's. I am not sure about AMD 6000 series GPU's we will see when first items arrived and first videos about MacPro.

 

[innerproduct]

at least we got closure now. We know the eco system and can make informed decisions. M2ultra has the potential to be quite nice even of it doesn’t compare to nvidia offerings. Let’s hope we get benches really soon.

 

[Xiao_Xi]

Has Apple replaced the ray tracing API?

Metal Performance Shaders | Apple Developer Documentation

Optimize graphics and compute performance with kernels that are fine-tuned for the unique characteristics of each Metal GPU family.

developer.apple.com

 

[jmho]

Has Apple replaced the ray tracing API?
View attachment 2213933

Metal Performance Shaders | Apple Developer Documentation

Optimize graphics and compute performance with kernels that are fine-tuned for the unique characteristics of each Metal GPU family.

developer.apple.com

Interesting.

That's only the MPS version, which is not the main way of doing raytracing, and I'd guess the reason they've deprecated it is because they want people to use the Metal API directly.

Maybe they'll say why in the raytracing talk tomorrow.

 

[jujoje]

Things I didn't expect from Apple: a full on node based material X authoring editor in Reality Composer Pro:

Surprisingly fully featured; looking forward to checking it out later this month.

And also the metal raytracing video is now up (haven't watched it yet myself...):

Your guide to Metal ray tracing - WWDC23 - Videos - Apple Developer

Discover how you can enhance the visual quality of your games and apps with Metal ray tracing. We'll take you through the fundamentals of...

developer.apple.com

 

[diamond.g]

Things I didn't expect from Apple: a full on node based material X authoring editor in Reality Composer Pro:
View attachment 2214968

Surprisingly fully featured; looking forward to checking it out later this month.

And also the metal raytracing video is now up (haven't watched it yet myself...):

Your guide to Metal ray tracing - WWDC23 - Videos - Apple Developer

Discover how you can enhance the visual quality of your games and apps with Metal ray tracing. We'll take you through the fundamentals of...

developer.apple.com

Is Apple allowing developers to make their own BLAS/TLAS? That is something the other API's blackbox (as far as I can tell). Wonder if Apple will do/allow dynamic BLAS build for "infinite LOD" (or to drop LOD models I guess).

 

[aytan]

I m not technical person or game developer, all I understand from the video is Metal rewrites the whole rendering process like Octane. All I can wait from DCC developers adopt these features. Looks like near feature we could have extreme speed ups from rendering engines. Also Metal looks like designed from 0 particularly for AS SOC. Am I wrong ?

 

[Macintosh IIcx]

Why does the performance increase in M2 Ultra vary so much from application to application?

View attachment 2212840

View attachment 2212841

Mac Studio

The new Mac Studio. Supercharged by the M2 Max and M2 Ultra chip. Extensive connectivity. Stunningly compact.

www.apple.com

The Octane X caught My eyes too, and the only reasonable explanation is that the scene is soo big that it uses pretty much all of the M2 Ultra’s 192 GB ram and that the M1 Ultra is Ram starved here.

 

[aytan]

The Octane X caught My eyes too, and the only reasonable explanation is that the scene is soo big that it uses pretty much all of the M2 Ultra’s 192 GB ram and that the M1 Ultra is Ram starved here.

Sure I have been as you described. VRAM is a issue when anyone uses Octane as an render engine. You can check it here.

https://forums.macrumors.com/threads/tracking-3d-software-for-apple-silicon.2385911/

 

[Xiao_Xi]

The Octane X caught My eyes too, and the only reasonable explanation is that the scene is soo big that it uses pretty much all of the M2 Ultra’s 192 GB ram and that the M1 Ultra is Ram starved here.

I find it stranger that a "preproduction Mac Studio systems with Apple M2 Max, 12-core CPU, 38-core GPU, and 96GB of RAM" can beat a "production Mac Studio systems with Apple M1 Ultra, 20-core CPU, 64-core GPU, and 128GB of RAM" in "OTOY Octane X Prime 2022.1.1 tested using a scene with complex materials and 35 million triangles."

Mac Studio

The new Mac Studio. Supercharged by the M2 Max and M2 Ultra chip. Extensive connectivity. Stunningly compact.

www.apple.com

 

[avkills]

I never believe Apple marketing benchmarks.

 

[bcortens]

I find it stranger that a "preproduction Mac Studio systems with Apple M2 Max, 12-core CPU, 38-core GPU, and 96GB of RAM" can beat a "production Mac Studio systems with Apple M1 Ultra, 20-core CPU, 64-core GPU, and 128GB of RAM" in "OTOY Octane X Prime 2022.1.1 tested using a scene with complex materials and 35 million triangles."

Mac Studio

The new Mac Studio. Supercharged by the M2 Max and M2 Ultra chip. Extensive connectivity. Stunningly compact.

www.apple.com

There is something in the way the M1 Max and Ultra GPU scaling works that seems to significantly hold back the GPU on those chips. If you look at blender open data benchmark you can see the M2 Max can match the M1 Ultra in blender rending already and that the M2 Pro -> M2 Max scales more linearly than did M1 Pro -> M1 Max

 

[leman]

Blender benchmark database shows that M2 Pro/Max have substantially improved the rendering performance over the M1 series. M2 Max is same speed or faster than M1 Ultra here.

So something has changed, maybe the way how they schedule work on GPU cores, or maybe how synchronisation is done. There are some recently published patents that describe a new interconnect design for Apple GPU as well as a new work distribution system, we might be seeing the effects of that.

 

[diamond.g]

Blender benchmark database shows that M2 Pro/Max have substantially improved the rendering performance over the M1 series. M2 Max is same speed or faster than M1 Ultra here.

So something has changed, maybe the way how they schedule work on GPU cores, or maybe how synchronisation is done. There are some recently published patents that describe a new interconnect design for Apple GPU as well as a new work distribution system, we might be seeing the effects of that.

Any particular reason they cannot backport any scheduler changes to M1 Ultra? That kind of thing shouldn't be "hardcoded" right?