3D Rendering on Apple Silicon, CPU&GPU

[Xiao_Xi]

Dave2D tested M3 Max in Blender 4.0

 

[galad]

Is that on the CPU or the GPU?

 

[jeanlain]

So now add Hardware RT to that 🤔

So, without ray-tracing, the M3 GPU core a 58% faster than the M2's. This seems too good to be true.
I wonder if the RT cores aren't used to a certain extent here. If MetalRT is enabled, could it leverage the dedicated hardware, even under blender 3.6? Is MetalRT used in these tests, or is the custom Blender RT code used instead (which is the default)?

 

[jeanlain]

Dave2D tested M3 Max in Blender 4.0

Hugely disappointing unless something went wrong in his tests.

 

[galad]

Yes, MetalRT in older Blender version can already uses HW Rt. See geekerwarn test posted above.
Performance might be a bit different on more recent versions.

 

[jeanlain]

@leman mentioned this strong showing could be due to the dynamic cache.

Why about the RT hardware? Does cinebench use MetalRT?

 

[galad]

Anyway, please post screenshots that are clear, those random screenshots from random people using random unknown settings add only to the general confusion.

 

[Xiao_Xi]

Is that on the CPU or the GPU?

GPU

Hugely disappointing unless something went wrong in his tests.

It looks like Dave2D used an old open Blender movie for the benchmark. So I'm not sure if that hurts GPU performance.

 

[aytan]

I am waiting for first Redshift Benchmark and e few real life results, also Octane benchmarks could be give an idea what we could have from M3 Max...

 

[diamond.g]

Why about the RT hardware? Does cinebench use MetalRT?

Pretty sure Cinebench 2024 is just straight compute.

 

[MayaUser]

Pretty sure Cinebench 2024 is just straight compute.

it seems cinebench 2024 uses RT

 

[Xiao_Xi]

Pretty sure Cinebench 2024 is just straight compute.

Can anyone confirm if Redshift uses MetalRT or not?

With the incorporation of Redshift, a different rendering engine, larger memory footprint, and more complex scenes, Cinebench 2024 offers a distinctly enhanced and accurate evaluation of modern hardware capabilities.

Maxon Introduces Cinebench 2024

A Leap Forward in Performance Benchmarking for Creatives, Gamers and Hardware Enthusiasts

www.maxon.net

 

[aytan]

This is quite interesting...
4:47
it is in German but does not effects anything

 

[aytan]

Can anyone confirm if Redshift uses MetalRT or not?

Maxon Introduces Cinebench 2024

A Leap Forward in Performance Benchmarking for Creatives, Gamers and Hardware Enthusiasts

www.maxon.net

The only information is this sentence

Redshift GPU performance can be evaluated on systems with compatible Nvidia, AMD and Apple graphics processors.

Base on that it should be, but I am not sure

 

[Spybreakjj]

I've been wanting to post this for a few days, but... embargo lol

I have some Octane numbers. I got a friend with a review unit to render a couple of scenes for me:

Octane Standalone 2023.1 / N64 Render:

Nvidia RTX 4090 (desktop): 42s
Nvidia RTX 3090 (desktop): 1m00
MacBook Pro M3 Max: 1m45s
MacBook Pro M1 Max: 5m20s

About 3x faster from M1 Max to M3 Max. Not sure if Octane is using HW RT, but I wouldn't count on it so there might even be more improvements coming later.

Octane Standalone 2023.1 / OctaneXBench_VolumeDenoiser:

Nvidia RTX 4090 (desktop): 4s
Nvidia RTX 3090 (desktop): 7s
MacBook Pro M3 Max: 15s (this is the same score as an M2 Ultra)
MacBook Pro M1 Max: 30s

 

[aytan]

I've been wanting to post this for a few days, but... embargo lol

I have some Octane numbers. I got a friend with a review unit to render a couple of scenes for me:

Octane Standalone 2023.1 / N64 Render:

Nvidia RTX 4090 (desktop): 42s
Nvidia RTX 3090 (desktop): 1m00
MacBook Pro M3 Max: 1m45s
MacBook Pro M1 Max: 5m20s

About 3x faster from M1 Max to M3 Max

Octane Standalone 2023.1 / OctaneXBench_VolumeDenoiser:

Nvidia RTX 4090 (desktop): 4s
Nvidia RTX 3090 (desktop): 7s
MacBook Pro M3 Max: 15s
MacBook Pro M1 Max: 30s

Thank you very much 👍
I supposed this is impressive for a laptop and 3 times faster than M1
Well this is something new...

 

[jeanlain]

About 3x faster from M1 Max to M3 Max. Not sure if Octane is using HW RT, but I wouldn't count on it so there might even be more improvements coming later.

Unless the nvidia GPUs also run without ray tracing, I don’t se how the M3 Max could achieve this score without RT.
These desktop GPUs consume many times the power.

 

[aytan]

Can anyone confirm if Redshift uses MetalRT or not?

Maxon Introduces Cinebench 2024

A Leap Forward in Performance Benchmarking for Creatives, Gamers and Hardware Enthusiasts

www.maxon.net

Acording to this yes it does

Maxon Redshift and Cinebench Optimized for Mac Lineup with M3 Family…

With the new GPU of M3, features like hardware-accelerated ray tracing offer outstanding performance in Redshift rendering.

www.maxon.net

 

[leman]

Can anyone confirm if Redshift uses MetalRT or not?

It most certainly uses Metal RT

 

[l0stl0rd]

Hugely disappointing unless something went wrong in his tests.

What I thought.

 

[ggCloud]

Hugely disappointing unless something went wrong in his tests.

There's nothing wrong with the test. We just have to wait

 

[aytan]

Guess it is M3 Max 40 GPU cores otherwise it is absurd to reach that level with M3 Max 30 GPU cores

 

[diamond.g]

Guess it is M3 Max 40 GPU cores otherwise it is absurd to reach that level with M3 Max 30 GPU cores

Is there not hardware for the denoising?

 

[Spybreakjj]

Guess it is M3 Max 40 GPU cores otherwise it is absurd to reach that level with M3 Max 30 GPU cores

That's the 40 core one, yes (I provided the score)

 

[aytan]

That's the 40 core one, yes (I provided the score)

Even Octane not yet optimized for AS RT Cores