M1 Mac Mini, Mac Studio - Universal Binary - Peformance?

Guys i am urged to buy a new computer due to moving and acute space issues. I initially wanted to wait for an upgrade of the M1 Mac Mini, but it seems to be taking more time than anticipated (rumours are quoting 2023 for an upgrade) and the new Mac Studio is just a bit too big for my new table and a bit pricy, still considering though.

Does anybody have a M1 Mac Mini and has tested if the now available Universal Binary brought any performance improvement and makes working maybe including TestMetal possible without having to look back in vain? i am running a 2014 macbook pro, so i am not very spoiled but would like at least a noticeable performance improvement regarding rendering.

@dan @nathanletwory anything you guys have noticed? did you manage to test the entry level mac mini?

thanks in advance.

1 Like

Hi -

As far as I understand, the performance gains will come from improvements in the Metal display pipeline; not from native support for Apple Silicon.
Also see:

-wim

not so fast wim :slight_smile: a fully adapted version should improve the performance a bit. yes Metal is the main gain i know that much, @stevebaer has commented on that either. Dan’s post is not entirely clear to me, if launching faster and running normal could not also mean a little bump in general :man_shrugging:

so it leaves me wondering if a M1 Mac mini is a good idea at all, or if i should just sit it out or take a mac studio and squeeze it somehow under the armpits of my monitor which then will give me a ducks neck. the mac mini can at least be placed on a vertical holder not so sure about that funny cubus with fans sucking in from below.

It’s a pretty sound, little desktop computer, but not meant for serious 3D work.
There are a couple of videos on YouTube of people trying it with Rhino, but I wouldn’t expect any wonders, especially when it comes to rendering.
It also only comes with 8 or 16 GB of memory, which gets shared between the CPU and GPU, which is something to keep in mind!

The only significant performance differences that I have noticed in moving from Rosetta2 to Apple Silicon native, when running on Apple Silicon, are related to launch times. Please note: this is not true of the transition to Metal (which is still underway). The transition to Metal should show significant display performance increases in the viewport.

When it comes to Cycles, (Rhino Rendering out-of-the-box) we are not yet a point where we can, with any real confidence, say what the performance gains will be. I have high hopes, but much work remains to be done.

It could be worthwhile to take a peek at the Cycles-specific source code from Blender 3.1, since they now advertise to have fully implemented Metal GPU support for Apple Silicon and even AMD chips under macOS Monterey 12.3.. :wink:

I’m sure @nathanletwory has done much more than just peak :laughing: He’s organizing quite a bit of his work-life around it.

When I get to a stage I am happy with I’ll publish the new repository PowerCycles.

1 Like

i have seen that video you linked above, i think its important to point out that he used 8gb instead of the possible 16gb which drastically would improve graphics performance.

also this little sound box left quite a few not that old computers behind in many regards, computers which are actually much younger than mine from 2014, you think hoping for it to be a little more sufficient for Rhino than my old computer is really that much off? i am not in need for a monster machine though i wish for navigational speed and windows to pop up, commands to load and complete swiftly, that all depends on how the software development is proceeding, well yes rendering could be a little faster.

so… anybody out there having experience?

Yes, but as I pointed out above, you probably have to split that in half, which in case of the 16GB configuration, would mean 8GB for the CPU - which isn’t great for any CG work, even 2D - and 8GB for the GPU, which also is only mediocre at best.
My 2016 MacBook Pro has 16 GB of system memory and 4GB GPU RAM, which when summed up is more (i.e. 20GB).
It should be considered though that the CPU and GPU will probably only rarely both tap hardcore into the memory pool at the same time. Also the Apple Silicon memory seems to be widely faster than any non-soldered, less integrated, traditional memory.

I can’t really tell you, since I don’t have one, but judging from Trevor’s video and others, I’d say that it’s probably ok-ish for small and medium Rhino projects. He doesn’t show a more elaborate one though.
He also mentions from time to time that the things that don’t work as expected in Rhino, are the same on his Windows machine, however the specs of the latter remain unknown, which makes it kind of hard to compare.
People always tend to lean towards preaching that Rhino works better on Windows, but that’s simply not true. It largely depends on your hardware and its performance.

Exactly, it’s now up to the developers to squeeze every bit of performance out of the Apple Silicon stuff that they can and up to Apple to render that feasible and not too hard.
Blender 3.1 has done a huge leap, with allowing GPU rendering for both older AMD Macs and the M1 chips, which previously was not the case (except for the thrid-party AMD Renderer add-on or using an NVIDIA eGPU).
Rhino also seems to have made good progress so far and better things are even on the horizon.

It’s “only” about $1000 so why not get one and test it for a month.
If it doesn’t please you, you can resell it for at least $700-800, which doesn’t seem so bad.

that is what i was thinking but i am considering to get it out of my way and take a mac studio if i am not fully convinced that the mini will do. i dont have much time for experiments right now.

well the truth is a bit more complex than that and different to what we are used to. i am not a computer scientist so what i say has to be taken with a grain of salt. it is claimed that RAM now simply has less relevancy, specifically since the data ping pong became so tightly and readily integrated.

the m1 mac mini handles high resolution video far beyond to what mine can do, so why not also handling a few meshes better than before, i now have a 2 gb video card, that is even considering the transfer rate might have been/still is with faster video memory if that is true at all still quite a difference.

so having all that information it makes me believe the mac mini could be actually plenty in these regards if i dont want to make real time raytracing matrix style.

1 Like

Yes, but the film industry and YouTube influencer crowd mean big business for Apple, and they always make sure that video and photo are amongst the things that works best. They have a whole panoply of hardware video encoders and decoders in the new M1 chips (hardware-accelerated H.264, HEVC, ProRes, and ProRes RAW). Intel and AMD chips have similar features too, but often to a lesser degree, or only in their upper tier. The M1 is even in iPads nowadays.

When it comes to 3D, things are wildly different. It’s a far more niche thing to do stuff in 3D with a much steeper learning curve than for instance casual Lightroom/Photoshop editing or cutting family videos in iMovie, which effectively means that there are far less people that you can sell your wares to.
It’s only in the last few years that Apple seems to have realized that 3D is even a thing, probably because of iOS and iPadOS game development, something that they weren’t really ever interested in on macOS. But then again, damn these iPhones sell!
Furthermore, most 3D softwares still use OpenGL in their display pipeline, which Apple depreciated a few years ago now in favour of their Metal framework and that brought a whole lot of issues and nay-sayers (even though there’s DirectX on Windows and NVIDIA proprietary everything).

As in photorealistic product renderings and/or arch viz?

fwiw our renderer definitely prefers native execution, ran a quick test to give you concrete numbers:

rosetta: benchmark 1975, time 39s
native:  benchmark 3400, time 22s

this is running on cpu only, and I found the difference to be enough to justify spending the time to figure how to get our external gui to start up native, even when being kicked off from a rosetta-hosted rhino instance, and to detect rosetta execution and put up a big red banner warning to switch to native, if anyone should somehow get it started up under rosetta (say starting it from a script).

6 Likes

quoted from apples page.

Metal provides a platform-optimized, low-overhead API for developing the latest 3D pro applications and amazing games

metal has been around for quite a while now. i dont think that they treated this facet of computer technology peripherally, yes it brings issues but to be frank not due to its technological aspect much rather due to the developers having to play ball integrating it.

yes, mainly for architectural renderings nothing out of the ordinary but currently i am still testing it, though it seems pretty capable.

Metal was first introduced with iOS 8 in 2014, but nobody really cared except for the folks in iOS game development. I feel like it didn’t gain much traction on macOS until the hole shift to ARM about two years ago now, which is factually totally unrelated.

Before, Apple in my opinion didn’t care about game development and 3D. It took the success of the iPhone and thus iOS and mobile gaming for them to realise that they now had an opening into a market that was previously heavily dominated by Microsoft and DirectX, even so brutally that macs still are thought of as not good for gaming.
With the huge success of iOS came Unity, the Unreal Engine, and other related CG applications that also must have exerted pressure for things to move forward.

So you’ve gotten an M1? If so, that was quick. :slight_smile:

I don’t know about you, but on my current home setup, Rhino is utterly useless for arch viz or really any rendering with Cycles. It’s so slow that it’s simply not worth even starting the UV-warping and texturing process.
I use Blender, which is much better in terms of viewport rendering and live updating, but render times are too long. A medium-sized scene with a couple of trees, some vegetation, the building, and a couple of textures can easily take up to an hour or longer to put out a full HD rendering (1000 samples), which I find painful.

I might invest in a Mac Studio in the future. We’ll see.

che right, no i meant i am testing the raytracer :smiley:

well it is an accurate render and delivers good quality, i was/am a bit spoiled with Cinema 4d though i stayed clear from any fancy add on render, Cycles for Rhino is just convenient since i do most of my actual modelling there and have started experimenting with it more and more. the speed very much also depends on a smart setup of light, fancy low light indirect light conditions are practically neck breaking, whilst a lot of direct light renderers faster, i assume because the quality does not depend on the bounces forever swirling around trying to fill up the scene that much then.

indirect or partially indirect is where cycles gets really slow, there i would wish a speed bump, but @nathanletwory is cooking up something great and i am sure that will be way more fun then.

till then i still need a new computer which can handle that at least sufficiently.

the ultra version is just too pricey and i am also a bit upset how awkward it looks in general and how big it is. i assume the specs could have fit into the mac mini case. it kind of looks like they wanted to make an artificial size to performance analogy also because the size of the m1 mac mini was kind of oversized, but right that is just heavy speculation now.

here a speed modelling test example i did for my old flat which will accommodate my mother in law, basically just for fun and to keep in shape, we bought some eames replica (i made a very crude version of it so dont kill me). it has a more indirect light setup, it rendered for an hour i think i snipped that shot somewhere at 700 bounces in a small window, grainy but the quality of the light looks promising and therefore motivating, i quite like it (not the grain of course)

Yeah, but its single and multi-core performance scores from the leaked Geekbench benchmark far exceed those of the 2019 Mac Pro’s fastest 28-core Xeon W-3275M processor, which costs close to $13 000.
I’ll probably get the less costly Max though, if I get one.

Look at x-ray rendering on the website, it grants a look inside.

The cooling system is huge. It takes up about two thirds of the enclosure. The Ultra is even said to feature a costly and heavy copper heat sink, whereas the Max has a cheaper and lighter aluminium one.

I wouldn’t even have noticed. :wink:

For an hour that’s way to noisy for my standards. Even with denoising turned off, Blender does a better job in way less time. It renders the viewport in about 2 to 5 minutes to quite okay image quality.

1 Like

i am thinking about it either, it should do for a while

it really is because the light has to bounce around in the room heavily and takes a huge amount of time. i also have a 2014 computer actually it is the late 2013 model 2,3 GHz Quad-Core Intel Core i7, so it is not the fastest :slight_smile: maybe blender would be faster, but i really cant stand the brain numbing interface and navigation, which i tried to hack for hours and still does not fully comply…

Yes, but it’s not really different in Blender. It’s even the same render engine, at least by name.

It surely would, but it’s a little less convenient because you need to export the entire Rhino scene into Blender as meshes.

In terms of UI, I find it less cluttered than for instance Rhino. It seems more homogenous over all.
The controls are like in any other CG application, if you set them to industry standard. It’s really Rhino that’s different, but it’s also more of a CAD-CG-chimera. I feel like with stuff like SubD modelling it’s really the Rhino controls that come back and bite you in the ass.