3D Rendering on Apple Silicon, CPU&GPU

[aytan]

I will give a try, I have my render times for previous versions and could figured out improvment.

On December 28th White Lands MetalRT GPU 2:50:38 / Metal GPU 2:34:50
Now White Lands MetalRT GPU 2:40:63 / Metal GPU 2:26:06, there is a slightly Cycles improvement but as developers says this update has effects mainly Evee and UI.
There is a quite big improvement about UI overall, everything works better and updates faster.
That's all I can observe.

 

[Xiao_Xi]

this update has effects mainly Evee and UI.

If you're curious about how your hardware performs compared to PC GPUs, TechGage has run a couple of tests to compare desktop PC GPU performance in EEVEE and viewport.

Blender 3.4 Performance Deep-dive: Cycles, Eevee & Viewport

Blender 3.4 released just a few weeks ago, and as usual, we couldn't wait long before jumping in and exploring its performance with current GPUs from AMD, Intel, and NVIDIA. With 22 graphics cards and six Cycles and Eevee projects in-hand for rendering and viewport testing, let's find out which...

techgage.com

 

[Pressure]

Metal works way different than OpenGL; so other than stubborn developers who refuse to update to use Metal; things seem to be looking good. Here's hoping Apple releases a monitor PCI-E based GPU with the new Mac Pro that also works in the older 2019 model as well. (I know, doubtful...lol)

Hopefully it would work as an eGPU in a Thunderbolt enclosure as well. It would also be great if someone made a TB enclosure with PCIe tunneling, so you could connect it over two TB4 cables for double the bandwidth.

 

[aytan]

If you're curious about how your hardware performs compared to PC GPUs, TechGage has run a couple of tests to compare desktop PC GPU performance in EEVEE and viewport.

Blender 3.4 Performance Deep-dive: Cycles, Eevee & Viewport

Blender 3.4 released just a few weeks ago, and as usual, we couldn't wait long before jumping in and exploring its performance with current GPUs from AMD, Intel, and NVIDIA. With 22 graphics cards and six Cycles and Eevee projects in-hand for rendering and viewport testing, let's find out which...

techgage.com

Honestly I do not curious about PC GPU's. I have a PC which has got RTX 3070 side by side with MS Ultra. In real world workloads these benchmarks do not mean anything. Some scenes 3070 delivers 3 times faster, but sometimes it did not ( also there is VRAM issue even if you have 24 GB GPU Memory with 64 GB RAM, for complex simulations even 24 GB GPU memory is not working well ). Here as an example which I had yesterday. On my PC C4D stops working. There is no particular reason, it was fine last time I rendered out a scene last week. But now it won't open, start up screen loads and noting happens further. I wrote to Maxon and send them crash reports yesterday, uninstall and reinstall did not work also. There is no response from Maxon untill now and I can not use PC anymore. I loose 1 day which is critical while there is a deadline. I will try to fix C4D issue on my PC later today but I m not optimistic at all. Long story short, I am not sure about faster PC is better overall. A 3070 PC on Blender final render times sometimes 3x faster, for some complex scenes not too much shorter and also it is not reliable, it varies even frame by frame also crashes more than usual.

 

[vladi]

Metal works way different than OpenGL; so other than stubborn developers who refuse to update to use Metal; things seem to be looking good. Here's hoping Apple releases a monitor PCI-E based GPU with the new Mac Pro that also works in the older 2019 model as well. (I know, doubtful...lol)

At the end of the line it's all about market share really. Apple has been neglecting parametric/direct modeling and rendering for so long.

 

[leman]

At the end of the line it's all about market share really. Apple has been neglecting parametric/direct modeling and rendering for so long.

If the value proposition is there, market share can change. The question is whether Apple can deliver technology that will be disruptive enough. If their ray-tracing tech is advanced enough so that their laptops can achieve competitive rendering speeds agains mid-range Nvidia desktop, there could be a lot of interest from the professional community.

 

[aytan]

If you're curious about how your hardware performs compared to PC GPUs, TechGage has run a couple of tests to compare desktop PC GPU performance in EEVEE and viewport.

Blender 3.4 Performance Deep-dive: Cycles, Eevee & Viewport

Blender 3.4 released just a few weeks ago, and as usual, we couldn't wait long before jumping in and exploring its performance with current GPUs from AMD, Intel, and NVIDIA. With 22 graphics cards and six Cycles and Eevee projects in-hand for rendering and viewport testing, let's find out which...

techgage.com

I tried with White Lands Scene, results are here
3070 PC Blender 3.4.1 Optix 01:07:51
3070 PC Blender 3.4.1 Cuda 01:21:09
MS Ultra Blender 3.4.0 2:50:27
MS Ultra Blender 3.5.0 2:25:25

 

[aytan]

ZBrush 2023 | ZBrush Docs

Zbrush is finally native!

I compare MS Ultra and 3070 PC ZBrush Redshift render on Zbrush,
3070 PC way beyond than MS Ultra base on Redshift render. Simply Nvidia GPU reacts 10 times faster on Zbrush. But its still workable on M1. I guess this difference should be expected.

 

[Xiao_Xi]

It's amazing how fast Apple is moving. Blender optimizations for the latest M2 SoCs have already started.

Cycles: Occupancy tuning for new higher end M2 machines · 08b3426df9

This patch adds occupancy tuning for the newly announced high-end M2 machines, giving 10-15% render speedup over a pre-tuned build. Reviewed By: brecht Differential Revision: https://developer.blender.org/D17037

developer.blender.org

 

[diamond.g]

It's amazing how fast Apple is moving. Blender optimizations for the latest M2 SoCs have already started.

Cycles: Occupancy tuning for new higher end M2 machines · 08b3426df9

This patch adds occupancy tuning for the newly announced high-end M2 machines, giving 10-15% render speedup over a pre-tuned build. Reviewed By: brecht Differential Revision: https://developer.blender.org/D17037

developer.blender.org

I wonder if Apple will continue optimizing for M1 SoCs since the M2 is out. Will M2 optimizations work on M1 units?

 

[l0stl0rd]

I wonder if Apple will continue optimizing for M1 SoCs since the M2 is out. Will M2 optimizations work on M1 units?

I think this might have to to with the bigger TLB as some people like maxtech think. So it would not do anything for M1 machines.

As some speculate that the GPU misc area is bigger, and the 20th core is not just disabled but there is other stuff there.

 

[Xiao_Xi]

I think this might have to to with the bigger TLB as some people like maxtech think. So it would not do anything for M1 machines.

I find it odd that Apple has different parameters for M2 Pro/Max, but not for M1 Pro/Max/Ultra.

intern/cycles/device/metal/kernel.mm

 

[l0stl0rd]

I find it odd that Apple has different parameters for M2 Pro/Max, but not for M1 Pro/Max/Ultra.

intern/cycles/device/metal/kernel.mm
View attachment 2146103

Not necessarily the M1 chips might all have the same limitations GPU wise.

Even if M2 Pro and Max did not get RT there might be some changes we don’t know about.

Perhaps we will get a confirmation about some changes once some got their units.

 

[sirio76]

I think this might have to to with the bigger TLB as some people like maxtech think

Don't trust anything from maxtech, his main purpose is not to be informative but to generate the more traffic/engagement/likes/subscrtiptions as possible in order to get more income.

 

[innerproduct]

M2max blender benchmark is beating the m1 ultra! Quite unexpected but very welcome! https://opendata.blender.org/benchm...AL&blender_version=3.4.0&group_by=device_name

 

[Xiao_Xi]

M2max blender benchmark is beating the m1 ultra! Quite unexpected but very welcome! https://opendata.blender.org/benchm...AL&blender_version=3.4.0&group_by=device_name

It seems that M2 GPU scales better in Blender.

SoC family​	Number of GPU cores​	Blender Score​
M2​	38​	1515​
M1​	64​	1483​
M1​	48​	1307​
M1​	32​	863​
M2​	19​	766​
M1​	24​	722​
M1​	16​	435​
M1​	14​	408​
M2​	10​	334​
M2​	8​	293​
M1​	8​	242​

Do any other programs show a similar trend?

 

[aytan]

It seems that M2 GPU scales better in Blender.

View attachment 2146685

SoC family​	Number of GPU cores​	Blender Score​
M2​	38​	1515​
M1​	64​	1483​
M1​	48​	1307​
M1​	32​	863​
M2​	19​	766​
M1​	24​	722​
M1​	16​	435​
M1​	14​	408​
M2​	10​	334​
M2​	8​	293​
M1​	8​	242​

Do any other programs show a similar trend?

Thanks for this data, its look like next generation Ultra/MacPro could be really powerful if any software build up on right terms. I can say C4D/Redshift implementation not bad at all right now on AS M1. Also VRay works well with AS M1.

 

[Xiao_Xi]

Has anyone tested M2 Pro/Max with Cycles and Eevee from Blender 3.5? How are the results?

 

[aytan]

Has anyone tested M2 Pro/Max with Cycles and Eevee from Blender 3.5? How are the results?

I really look for this also

 

[leman]

It seems that M2 GPU scales better in Blender.

View attachment 2146685

SoC family​	Number of GPU cores​	Blender Score​
M2​	38​	1515​
M1​	64​	1483​
M1​	48​	1307​
M1​	32​	863​
M2​	19​	766​
M1​	24​	722​
M1​	16​	435​
M1​	14​	408​
M2​	10​	334​
M2​	8​	293​
M1​	8​	242​

Do any other programs show a similar trend?

That's fairly interesting stuff. Few observations:

- scaling in M1 is pretty much linear until 48 cores, then we have a quick drop
- M2 Pro/Max appress to behave very differently to the base M2. The slope from 8-core to 10-core M2 is very similar to M1, then the slope instantly rises by 2x

 

[Xiao_Xi]

That's fairly interesting stuff. Few observations:

- scaling in M1 is pretty much linear until 48 cores, then we have a quick drop
- M2 Pro/Max appress to behave very differently to the base M2. The slope from 8-core to 10-core M2 is very similar to M1, then the slope instantly rises by 2x

I would not go into too much detail because some of these points for the M2 SoC family represent only a benchmark, so a larger sample could shift the points and show a different graph.

On top of that, there is a patch in the next version of Blender that will increase performance in M2 Pro/Max.

 

[aytan]

I would not go into too much detail because some of these points for the M2 SoC family represent only a benchmark, so a larger sample could shift the points and show a different graph.

On top of that, there is a patch in the next version of Blender that will increase performance in M2 Pro/Max.

Looks like this x2 performance gain only at Blender right now. In Redshift M2 GPU cores does not behave like Blender. Maybe there will be an upgrade and Redshift will adapt to M2 better in the future.

 

[leman]

Looks like this x2 performance gain only at Blender right now. In Redshift M2 GPU cores does not behave like Blender. Maybe there will be an upgrade and Redshift will adapt to M2 better in the future.

Do you have some info on Redshift performance for various M2 systems?

 

[aytan]

Do you have some info on Redshift performance for various M2 systems?

There is another topic

M2 Max 30-core Benchmarks​

you can check it, also I saw it on Redshift forum, M2 Max 30 Core gpu 09:37 / M1 Ultra 48 core gpu 07:17

 

[jujoje]

Some Apple related eye candy, courtesy of ManvsMachine:

Eye Candy

ManvsMachine — Design & Direction Co.

mvsm.com

(website takes forever to load though)