Mac Mini 2018 for Rhino 6?


#1

I am longing to replace my now soon 10 years old mid 2009 MacBook Pro with something affordable that still can work me through some stuff. I am aware that the internal GPU of the Mac Mini is pretty weak, but testing on a 3 year old MacBook Pro 13 inch from time to time without any specific GPU also seems to work “fine”.

The performance of my GeForce 9600M GT 512 MB from my old MacBook Pro would still perform well, IF you guys would not decide to drop El Capitan which is the last version I can use unfortunately. Not all the blame on you of course but also on Apple I guess, it is after all also a pretty old System which I have.

Anyway lets get to the point (questions)
• does it make any sense whatsoever to get a Mac Mini
for the usage of Rhino without intention of buying an eGPU?
• does cycles run or lets say perform anywhere usable?
@nathanletwory did you try cycles on such a system already?

Yes I know its all beta still, but is there anything you guys will or can do at all to reach out towards the standard hardware to make it perform as well as it can get from your side?


(Nathan 'jesterKing' Letwory) #2

@encephalon, I have not tested such a machine, but Raytraced will use in the very least the CPU and will work just fine - I test it daily on my Macpewter

kuva

AMD GPU - recent enough anyway - will be able to be used with OpenCL. It should work pretty well.


#3

@nathanletwory the GPU of the mac mini is actually an Intel UHD Graphics 630 or what did you mean? it is unfortunately by far the weak spot of that entire little Macpewter… which makes me very concerned if i should not wait still. connecting a bulky eGPU defies the sense of a compact-ewter. not so much about not affording it (my personal view at least)

anyway here @dan wrote that you have 2 Mac Minis available. I take it from what I read that you are from finnland and probably are not around testing one of these? if anybody of you guys could fire it up for a 3 and a half minutes to see if working on it with a decent amount of object in shaded and wire viewport seems acceptably smooth, maybe a quick raytraced test render.

not super urgent, but sure also interesting for others, since the Mac Mini is definitely a hot contester in affordability and performance.


(Nathan 'jesterKing' Letwory) #4

@dan wrote:[

quote=“dan, post:3, topic:75113”]
The Raytraced display mode seems slow, but does display. This is likely due to the Intel UHD Graphics 630.
[/quote]

The slowness is because there is not a supported GPU to use - so Raytraced runs on the CPU. It should work just fine. For lookdev I’d set the advanced setting RhinoCycles.DpiScale to a value that results in a bit blockier in the viewport, but ViewCaptureToFile and ViewCaptureToClipboard would still do full resolution - the non-blocky version.

Perhaps @dan could post some Raytraced images done on the mentioned mac minis for added effect.


#5

my main concern is working modeling in viewport. i am just assuming it should work ok, still some definite testing would be rather reassuring, if it turns out to be worse than what i have now it does not make any sense of course.

regarding rendering, yes i personally could live with renderings running on cpu, i am not too spoiled with real time renderings anyway. but would that mean that it would not make sense at all to combine cpu and gpu on an onboard graphics for cycles? since both utilize the same memory maybe not much boost?

anyway thanks for checking, commenting


(Nathan 'jesterKing' Letwory) #6

Not sure I understood that, but it boils down to this: I have not been able to get CUDA working on the NVidia GPU Mac machine we have at the office, so not able to even test CUDA rendering with Cycles, thus CUDA support in Cycles is not enabled. OpenCL works fine for supported cards (AMD GCN2 and newer), so that should help for such machines.

This is definitely something for @dan to comment on.


(Dan Belcher) #8

Hi @encephalon-

We have a Macmini8,1 that we use as a build machine. It’s a bit difficult to run a rigorous test, but let me see if I can provide something to get started…

This is a screen recording of the MacMini taken over Screen Sharing (remote desktop), running the latest RhinoWIP with a couple arbitrary models calling TestMaxSpeed:

I apologize for the video-tearing which is due to the screen-sharing connection (that wouldn’t happen locally).

I don’t know if this test satisfies your curiosity or not. Also, we still have more work to do on the RhinoWIP, of course.

-Dan


#10

thanks @dan the regular shaded mode seems to run fast enough, but i just see that the viewport is pretty small, did you hook it up to a 15" screen? that of course skews the result and impression, not sure myself now how helpful that is. on my 32 inch on my dinosaur macbook pro i get an average of 1.15 seconds.


(Dan Belcher) #11

You are correct, the viewport is small…artificially small in fact.

No, I did not hook it up to any screen, but rather used Remote Desktop to connect (this machine is in a server rack and difficult to access). The “display” resolution is 1280 × 1024. I would expect the performance to degrade when you attach a Retina display.

I agree this test of limited use, but it was quick to run.

Perhaps someone out there has a Macmini8,1 who can report back with a more “real world” scenario.


#12

I’ve got various era Mac Mini’s kicking around. When I get some copious free time I’ll give it a whirl.

Getting Cuda to play nice with non EFI GTX cards can be really challenging. I’ve got both 4.1 and 5.1 era Mac pro towers that I’ve been running 2 GTX-970’s per box in for a while now. OS upgrades require specific builds of the drivers to work, and it gets even more complicated if you’ve flashed the firmware to move a 4.1 box to a 5.1. I recently bumped my 5.1 to 10.13.x up from 10.11.x in order to run other software that required it, and it was a really miserable process

The challenge seems to be in getting cuda (and the web non cuda drivers) working in the pref panels once. This may require hacking the driver install to match the build number to what the installer is looking for, which if it succeeds will allow you to install it, but then come up reporting that the driver version is out of date, at which point it will pull down the correct build for the Hardware and OSX built combination, at which point after that install tends to work.

I’ve had times when manually pulling down the exact release the auto updater gets and installing it from package did NOT work (comes up disabled due to a some perceived version mismatch).

I have seen anecdotal stuff on MacRumors where some folks have gone in and hacked the KEXTs to get the drivers to run, but this is just too time consuming for me to mess with.

It’d certainly be nice if Apple decided to actually support this stuff, but I’ll credit NVidia for at least getting it mostly there so we’re not locked down to the few “official” GPU choices out there.


(Nathan 'jesterKing' Letwory) #13

Indeed. I have failed to get it working on the on NVidia MBP we have in the Finland office - therefor no CUDA support in Raytraced on the Mac, unfortunately.


#14

Well I do have it working on both the 5.1 and 4.1 I have here. Cuda working fine in Blender (cycles).

What I ended up doing on the 5.1 box to get it running under 10.13.x was to do a clean OS install of 10.11, install the 10.11.x build specific drivers and get that working.

Once that worked, I updated to the last point release of 10.11, and upon reboot the pref panels (Nvidia and the Cuda one) both reported updated drivers were available, and I did the updates from the pref panels.

That worked.

Then I updated to 10.13, and repeated the driver update process to bring the Nvidia stuff current.

Box is currently on 10.13.4 running 387.10.10.10.30.106 Nvidia Driver, and 410.130 for Cuda.

One trick that has helped is once you get ANY version running is to go into the pref pane and change it to Default MacOS Driver before doing any updates, as once it boots to any version that doesn’t match the driver, it won’t load, but it doesn’t discover that until almost completely through he boot cycle, and as a result you won’t see the gray startup screen until the desktop pops up. Not so much an issue on my towers as both have GT120 cards in addition to the 2 GTX-970’s, and being as they’re EFI’d for MacOS from the get go it, it’ll display regardless of driver loaded.

There’s a very long thread on MacRumors about this whole business going back for several years that was invaluable in getting this stuff to work.

This is the one that covered a lot of it:

https://forums.macrumors.com/threads/frequently-asked-questions-about-nvidia-pc-non-efi-graphics-cards.1440150/

And this is the current one:

https://forums.macrumors.com/threads/faq-nvidia-web-drivers.2170497/#post-27138260

Key point to note: They don’t work at all currently in 10.14, hence the reason I haven’t updated to 10.14, nor will I until Apple gets it’s act together in this regard.


(Nathan 'jesterKing' Letwory) #15

Talk about unwieldingly hard… (:

Still would be interested to hear if Rhino 6 in general works OK on your boxes…


(Graham) #16

Fancy providing official support for that method? :wink:


#17

as long as it is not unyieldingly hard…
dont tell me you have or had to give up on that completely?

if the newest series is upon them i would be glad to hear about it. its not very urgent but the sooner the better, my computer is still holding, but the urge for new software (versions) not supported by my system anymore, is becoming unstoppable :blush:


(Nathan 'jesterKing' Letwory) #18

On the backburner until I get my hands on a machine that is configured for me. Instead of wasting time trying to get this configured I instead work on stuff that isn’t a timesink - just today I completed: