Apple Silicon For GPU Rendering and 3D

[singhs.apps]

Damn 3 Titan RTXs! Yeah pretty sure it's not going to beat that Be curious to hear how it fairs though - without having delved into it too much I think the metal version of redshift is pretty decent, roughly comparable to CUDA so interested as to how it stacks up.

Also, if you happen to have Houdini kicking about, would be curious as to how high res a Pyro OpenCL / Minimal Solve you can fit in that amount of memory.



That would be awesome; Karma's been coming along really nicely and seems to be a pretty solid production renderer (after a bit of a rocky start). I think they're both pretty Optix based on the GPU front though. Speaking of which really hoping the new Mac Pro gets hardware based ray trace acceleration...

I remember when they announced the trash can Mac Pro demoing Maria painting Pixar assets from Monsters University. That didn't age well

I have Houdini Indie. If you have a sample scene, I could try whenever the MBP arrives. Sim is not my forte though.

With regards to memory, in Redshift tests, it seems Apple is limiting the amount of RAM available for the GPU.

Redshift Benchmark Results (Updated) - CG Director

Extensive Redshift Benchmark Results List with all modern GPUs, Operating Systems, CPUs and Multi-GPU-Setups. Find the best performing GPU for Redshift.

www.cgdirector.com

Last two under redshift Apple tests, the m1 and m1 (Mac mini) show different vram amounts ( 5 and 10 respectively). I am assuming the first one is an 8GB system and the other a 16 GB one.

Also Geekbench Opencl : Look under the Device-Memory section (https://browser.geekbench.com/v5/compute/3551790)
Memory used = 42.7 GB (out of 64 GB ram I assume)

This seems in line with someone’s comments in these forums viz the redshift Maxon developers mentioning ‘vram’ availability for the renderer as slightly above 40GB.

All this points to 60 - 62.5% ram available for the GPU, consistently whether ram is 8 GB or 64 GB

Thus if we get a ‘quad’ m1 max for the new Mac Pro, we might have around 170 GB available out of the total 256 GB.This would be underwhelming, leaving all the rest of ram out of the equation.

Someone can throw a light on this ?

 

[Macintosh IIcx]

I remember when they announced the trash can Mac Pro demoing Maria painting Pixar assets from Monsters University. That didn't age well

Mari running Metal, that looked so promising. Sigh. And now I think that the Foundry even killed development of Mari on Mac earlier this year. Idiots.

 

[Boil]

With regards to memory, in Redshift tests, it seems Apple is limiting the amount of RAM available for the GPU.

Redshift Benchmark Results (Updated) - CG Director

Extensive Redshift Benchmark Results List with all modern GPUs, Operating Systems, CPUs and Multi-GPU-Setups. Find the best performing GPU for Redshift.

www.cgdirector.com

Last two under redshift Apple tests, the m1 and m1 (Mac mini) show different vram amounts ( 5 and 10 respectively). I am assuming the first one is an 8GB system and the other a 16 GB one.

Also Geekbench Opencl : Look under the Device-Memory section (https://browser.geekbench.com/v5/compute/3551790)
Memory used = 42.7 GB (out of 64 GB ram I assume)

This seems in line with someone’s comments in these forums viz the redshift Maxon developers mentioning ‘vram’ availability for the renderer as slightly above 40GB.

All this points to 60 - 62.5% ram available for the GPU, consistently whether ram is 8 GB or 64 GB

Thus if we get a ‘quad’ m1 max for the new Mac Pro, we might have around 170 GB available out of the total 256 GB.This would be underwhelming, leaving all the rest of ram out of the equation.

Someone can throw a light on this ?

I would think, especially with the single SoC units, Apple reserves that 37.5% to 40% of RAM to ensure the OS & apps have enough RAM to work with; for the dual & quad SoC units, I would expect the Reserved RAM Ratio to change, opening up more RAM to the GPU cores to allow larger data sets without swapping to SSD...?

 

[singhs.apps]

Perhaps.
Well in the current 64GB M1 max Soc 40 GB is pretty good for dedicated ‘vram’… but yes I hope apple will probably fix this ‘vram’ limitation, esp for large ram capacity variants.

On a side note, I am also hoping for dedicated MPX GPU modules in the new Mac Pro…will be great for those who need serious GPU compute, perhaps even dedicated raytracing compute units

 

[jujoje]

I would think, especially with the single SoC units, Apple reserves that 37.5% to 40% of RAM to ensure the OS & apps have enough RAM to work with; for the dual & quad SoC units, I would expect the Reserved RAM Ratio to change, opening up more RAM to the GPU cores to allow larger data sets without swapping to SSD...?

This would make sense. I guess running out of system and gpu memory at the same time might be kinda bad Hopefully they increase the % allocated with the amount of memory and that the buffer between running out of allocated ram and the hard RAM limit would allow things to fail a bit more gracefully.

I have Houdini Indie. If you have a sample scene, I could try whenever the MBP arrives. Sim is not my forte though.

I'll try and put something together, possibly this weekend. Probably be something based off the shelf tools so nothing fancy but should helpfully get some representative results. I get the feeling the my iMac Pro will not come out of this looking good...

 

[jujoje]

I think that the Foundry even killed development of Mari on Mac earlier this year. Idiots.

I think they'd effectively stopped updating Mari for macOS a couple of years before that (there was a beta but it was two or three releases behind the Win/Linux versions). Also wondered what happened to the Modo Metal viewport they demoed; suspect we're not going to be seeing that one any time soon.

 

[singhs.apps]

Also wondered what happened to the Modo Metal viewport they demoed; suspect we're not going to be seeing that one any time soon.

Let Foundry get Modo working properly under Windows first before bellyaching over open gl deprecation in macOS.

 

[Macintosh IIcx]

Also wondered what happened to the Modo Metal viewport they demoed; suspect we're not going to be seeing that one any time soon.

They stated very recently that porting Modo to ARM/Metal is not on the current roadmap (or something to that effect). To be honest, I think that porting the OpenGL viewport to Metal is an effort beyond what they are willing to put resources into.

Ironically, a viewport running in unified memory (Metal) would probably smoke everything else in the market, as updating OpenGL in the Vram constantly back and forth between CPU and GPU is slow and clunky by todays standards. Oh well... ?

 

[hefeglass]

ran some blender tests..here is a video
granted this is only testing CPU vs CPU..no point in testing GPU vs CPU imo.

I hope a metal branch drops soon!

 

[Boil]

I hope a metal branch drops soon!

If it's Metal, wouldn't it be a vein...?!? ;^p