3D Rendering on Apple Silicon, CPU&GPU

[avkills]

I'm honestly curious to know if you think the majority of Apple's customer base will upgrade their Macs?

How it is "green" if Apple were to expend engineering, manufacturing, shipping, warehouse resources, etc, so that maybe a very small portion of their customer base can enjoy upgrading their Macs?

Everybody I know who has a Mac Pro tower has done some type up upgrade, whether that is memory, storage drives or PCI cards. Hell back in the day I even upgraded my B&W G3 to a G4; but granted things were different back then.

Personally I feel that the iMac was (is) the most wasteful product ever made. But to each their own.

All I know is I would sure like to upgrade my GPUs in my 2019 to 7000 series AMD; doubt that will ever be an option. Maybe Sonoma will have drivers.

 

[MRMSFC]

Everybody I know who has a Mac Pro tower has done some type up upgrade, whether that is memory, storage drives or PCI cards. Hell back in the day I even upgraded my B&W G3 to a G4; but granted things were different back then.

Personally I feel that the iMac was (is) the most wasteful product ever made. But to each their own.

All I know is I would sure like to upgrade my GPUs in my 2019 to 7000 series AMD; doubt that will ever be an option. Maybe Sonoma will have drivers.

I doubt there will be first party drivers for future AMD card releases now that the Intel Mac Pro is dead. Only drivers for the cards shipped with the 7,1.

And yeah, one of my favorite things was tinkering with my 5,1 Mac Pro. I’m probably still gonna get cheap parts for it to mess with it more.

I do think Apple could be more imaginative with the Mac Pro. There’s all sorts of wasted potential for the platform, even outside gpus. But it’s a tough sell, especially considering Apple has more demographic data than we do and they probably have a better understanding of what their customers want than us.

 

[quarkysg]

Everybody I know who has a Mac Pro tower has done some type up upgrade, whether that is memory, storage drives or PCI cards. Hell back in the day I even upgraded my B&W G3 to a G4; but granted things were different back then.

Personally I feel that the iMac was (is) the most wasteful product ever made. But to each their own.

All I know is I would sure like to upgrade my GPUs in my 2019 to 7000 series AMD; doubt that will ever be an option. Maybe Sonoma will have drivers.

All well and good, but do you really think that Apple spending all those resources and thus consuming energy and resources doing so, will offset the upgrades "savings"?

Your aim is to be "green" am I right?

AFAIK, Apple have a very good re-cycling programme for their products.

Perhaps you are using products produced by the wrong company?

 

[avkills]

All well and good, but do you really think that Apple spending all those resources and thus consuming energy and resources doing so, will offset the upgrades "savings"?

Your aim is to be "green" am I right?

AFAIK, Apple have a very good re-cycling programme for their products.

Perhaps you are using products produced by the wrong company?

Actually I don't go out of my way to be overly "green." But I do call out companies claiming to be green who do everything in their power to make people buy all new stuff instead of allowing reasonable upgrades. And by reasonable upgrades I mean RAM and hard drives.

And really Apple can't claim they are green unless even single company that produces parts and assembles their products are also green.

We have derailed this thread enough....let's just stick to talking about 3d rendering. Anyone know if Sonoma is going to have any major updates to the Metal API?

 

[leman]

We have derailed this thread enough....let's just stick to talking about 3d rendering. Anyone know if Sonoma is going to have any major updates to the Metal API?

A few! For rendering most interesting features are probably atomic texture, curve raytracing, as well as some new features in RT acceleration structures.

Then we have bfloat16 support (relevant for ML), more flexible mesh shader setup, and more GPU-driven stuff.

 

[deconstruct60]

Hardly. The latency is already super high. Not that latency matters that much for GPUs, they can hide it fairly well.

You are using Apple's scoped down notion of a GPU there. For computation there is a lower impact. If folks want to interactively see 3D rendering on a viewport then latency does matter. Is the GPU that subsystem out to the video out or just initially filling the frame buffer to hand off to someone else.

The problems kick up when the display controller is all the way 2-3 dies over away . If don't get the complete set of data to it fast enough it is on the next frame. Miss the timing on several frames and users is likely not as happy ( in the it doesn't matter zone). As the refresh rates get higher, the latency constraints tighten.


The GPU designs aimed at the 'high frame rate' market have largely skipped the multiple die set up for substantive reasons. More 'pure compute' ( compute and stuff answers into RAM/Disk) don't have those kinds of output constraints.

 

[leman]

You are using Apple's scoped down notion of a GPU there. For computation there is a lower impact. If folks want to interactively see 3D rendering on a viewport then latency does matter. Is the GPU that subsystem out to the video out or just initially filling the frame buffer to hand off to someone else.

The problems kick up when the display controller is all the way 2-3 dies over away . If don't get the complete set of data to it fast enough it is on the next frame. Miss the timing on several frames and users is likely not as happy ( in the it doesn't matter zone). As the refresh rates get higher, the latency constraints tighten.


The GPU designs aimed at the 'high frame rate' market have largely skipped the multiple die set up for substantive reasons. More 'pure compute' ( compute and stuff answers into RAM/Disk) don't have those kinds of output constraints.

What you are talking about will start being problematic if the latency goes up by dozens of milliseconds. That’s orders of magnitude off what we are discussing here. A few dozen of nanoseconds added data package transfer is not going to break down your rendering.

High-latency graphics output is more common than one might think. Most gaming laptops for example send the frame back over PCIe.

 

[bcortens]

You are using Apple's scoped down notion of a GPU there. For computation there is a lower impact. If folks want to interactively see 3D rendering on a viewport then latency does matter. Is the GPU that subsystem out to the video out or just initially filling the frame buffer to hand off to someone else.

The problems kick up when the display controller is all the way 2-3 dies over away . If don't get the complete set of data to it fast enough it is on the next frame. Miss the timing on several frames and users is likely not as happy ( in the it doesn't matter zone). As the refresh rates get higher, the latency constraints tighten.


The GPU designs aimed at the 'high frame rate' market have largely skipped the multiple die set up for substantive reasons. More 'pure compute' ( compute and stuff answers into RAM/Disk) don't have those kinds of output constraints.

Except that Intel, Nvidia and AMD are all exploring multi-chip designs, there are hard reticle limits and transistor density problems that are starting to crop up that are making it harder and harder to build bigger and bigger GPUs. As time goes on I suspect that over time we will see more and more GPUs adopt a tiled or chip-let based approach.

 

[deconstruct60]

Except that Intel, Nvidia and AMD are all exploring multi-chip designs, there are hard reticle limits and transistor density problems that are starting to crop up that are making it harder and harder to build bigger and bigger GPUs. As time goes on I suspect that over time we will see more and more GPUs adopt a tiled or chip-let based approach.

For data center GPUs ... yes.

AMD reportedly cancelled their very high end mainstream one.

AMD's canceled Radeon RX 8000 "Navi 4C" GPU diagram has been partially leaked - VideoCardz.com

AMD Navi 4C: The GPU that will never be? This would have been the AMD flagship RX 8000 GPU if not for the fact that it was canceled, allegedly. Alleged AMD Navi 4C, Source: Moore’s Law is Dead Moore’s Law is Dead recently shared a diagram of AMD’s unreleased RX 8000 GPU, codenamed Navi 4C. In […]

videocardz.com

[ Likely somewhat in part because there is pragmatically no manufacturing capacity to make hyper large footprint packages. Everyone is chasing the AI/ML boom and that will soak up all capacity. ]


Bigger and bigger and bigger (core counts ) is pretty much highly coupled to more and more and more expensive. Which is smaller and smaller demographics of the buyers.

Some decomposition will creep in because I/O and SRAM/cache is on very different evolutionary paths now. But the maximum bulk core approach doesn't match up with that. And definately not with Apple's 'too chunky', mash-up of monolithic dies posing a 'chiplets'.

 

[BootLoxes]

Zbrush is coming to iPad.

 

[Lone Deranger]

Yep.

https://twitter.com/i/web/status/1708906063133831384

And all it took was another app (NomadSculpt) to start eating their lunch. Yay for competition.

They'd better sort out the GUI though. Can you imagine the mess that it currently is but now on a small 10-11" screen? Endlessly scrolling through an 80K pixel tall Tool palette?

Also, I predict this to be a severely cut down version. I doubt they'll implement all the old legacy features.

 

[ader42]

Yep.

https://twitter.com/i/web/status/1708906063133831384

And all it took was another app (NomadSculpt) to start eating their lunch. Yay for competition.

They'd better sort out the GUI though. Can you imagine the mess that it currently is but now on a small 10-11" screen? Endlessly scrolling through an 80K pixel tall Tool palette?

Also, I predict this to be a severely cut down version. I doubt they'll implement all the old legacy features.

Yep UI will ideally be customised but it’s not too bad as-is.

I currently use my 11.9” iPad Pro /Apple Pencil with my MBP and Studio Display, I had to make a custom / minimised UI set to maximise sculpting area but it works fine.

It’s nice to have the Studio Display to look at periodically but I do mostly look at the iPad.

I do want a bigger iPad but that‘s more about wanting more sculpting area than anything.

I think the biggest issue might be managing files and the amount of RAM…

 

[Pressure]

Yep UI will ideally be customised but it’s not too bad as-is.

I currently use my 11.9” iPad Pro /Apple Pencil with my MBP and Studio Display, I had to make a custom / minimised UI set to maximise sculpting area but it works fine.

It’s nice to have the Studio Display to look at periodically but I do mostly look at the iPad.

I do want a bigger iPad but that‘s more about wanting more sculpting area than anything.

I think the biggest issue might be managing files and the amount of RAM…

Would be nice if memory wasn't tied to storage on iPads 😅

 

[BootLoxes]

My body is ready

 

[aytan]

My body is ready

View attachment 2288023

I fell like I could be required at least 32 GB RAM and their aim is for M3 iPad version. Recent M2 iPadPro could be 16 GB RAM max for 1 TB iPadPro models. 2018 iPadPro models have 4/6 GB RAM, M1 iPadPro has 8/?? GB RAM and M2 has 8/16 GB RAM versions. I have been worked/play with ZBrush since it's 1st release. RAM amount could be an issue for a number amount of ZBrush operations. We have been tested similar issue with 2018 PadPro and M1 iPadPro on DaVinci Resolve, sometimes ( depends on your project details such as length/number or data clips ext. ) it can not even run some projects. I fell like it is not logical to by an iPadPro for ZBrush or any 3D sculpting app, but if anyone recently have one it could be useful . and other issue is screen size, personally I can work with an 12.9 inch screen for only digital painting for the sake of Apple Pencil Maybe there will be an 15 inch or larger iPadPro version for M3.. who knows ?

 

[ssgbryan]

I'm honestly curious to know if you think the majority of Apple's customer base will upgrade their Macs?

How it is "green" if Apple were to expend engineering, manufacturing, shipping, warehouse resources, etc, so that maybe a very small portion of their customer base can enjoy upgrading their Macs?

Not anymore.

People like me left. I want to get things done (3d Art) and apple is no longer in the space I work in.

 

[ader42]

Not anymore.

People like me left. I want to get things done (3d Art) and apple is no longer in the space I work in.

I will be upgrading to an M3 Ultra Mac Studio when it comes out, and if the biggest iPad they release with M3.

Life is too short to run windows now lol - (I did that as a programmer for decades when I ran 2 machines, one Mac and one PC).

I’m liking Redshift inside ZBrush, hopefully this will be in the iPad version of ZBrush too (but likely an in-app subscription which I’d be fine with).

 

[LymeChips]

Excited to see the new GPU features in Octane and Cycles soon 👀 I haven’t heard of dedicated mesh tracing or whatever it’s called - I wonder if that’s more for real time or if offline rendering would benefit?

 

[diamond.g]

Excited to see the new GPU features in Octane and Cycles soon 👀 I haven’t heard of dedicated mesh tracing or whatever it’s called - I wonder if that’s more for real time or if offline rendering would benefit?

I don't think Blender supports mesh shaders.

 

[ader42]

Mesh shading is more for games.

The dynamic caching and hardware accelerated raytracing will be of more interest to this thread; Apple showed Cinema4D so hopefully Maxon are in the loop and will have it integrated asap.

 

[Upgrader]

They also mentioned Redshift working well with it.

 

[l0stl0rd]

Blender sure is getting ready already

Cycles: Apple M3 tuning including hardware raytracing · f2bb4c617f

This PR adds tunings for the [newly announced](https://www.youtube.com/watch?v=ctkW3V0Mh-k) M3 family of chips. In particular, MetalRT will be enabled as the automatic default for intersection testing on M3 and beyond to take advantage of hardware raytracing. This will result in significant path...

projects.blender.org

 

[Romain_H]

What do the experts here think - how is Unreal Engine gonna be affected?

 

[diamond.g]

What do the experts here think - how is Unreal Enginegonna be affected?

Assuming the current build use Metal RT, you should see a performance improvement (when doing legacy RT rendering). Nanite performance isn't reliant on Mesh Shaders (they help though). Lumen SW mode should see speed increase but it isn't dependent on HWRT.

 

[dugbug]

While dynamic caching is kind of fuzzy at the moment I think this article attempts to explain it pretty well

Dynamic Caching: an attempt to explain M3's biggest feature | Digital Trends

Apple's Dynamic Caching is one of the most technically dense features the company has ever revealed. Here's our best understanding of it.

www.digitaltrends.com