mac studio / mac pro m2 ultra

[prince_jr]

hi corona users
i'm still running a maxed out mac pro from 2010 (5.1) at home > 12c, 3.46ghz, 128gb ram.
but as apple released a new version of the mac studio and finally the new mac pro, i'm suggesting to buy a new apple computer. rather a mac studio than a mac pro.

it would be nice to get some real world benchmarks in this thread. does anybody already have the mac studio m2 ultra on his/her desk and can share some infos about working in the viewport, maybe some render times and other important facts?

@corona-team: i guess corona benchmark v2 (optimized for apple silicon) is just around the corner?

thanks for your inputs and thoughts. i'm curious what y'all have to share in the near future. :o)

[BigAl3D]

I'd like to see this as well. One of the tests I like to run since every C4D user already has the scene, is to open the Grapes scene in the Asset Browser and just hit Render to Picture Viewer. Not a Corona test, but a raw CPU power test. I have several render times from over the years and I believe a thread here somewhere where others tried it with newer Macs. The benchmarks are nice, but it's difficult to apply that knowledge to real-world projects.

A smart decision not to get that "new" Mac Pro. Only peeps that have a need for specific PCI cards could possible justify the extra $3,000 USD for that box. Hoping they're fixing the sudden bottleneck with scaling these new chips. Started off so promising.

[Philw]

The native benchmark is up - would be great to see some Ultra scores up there :-)

https://corona-renderer.com/benchmark

I've put my M1Max and Threadripper up there:
https://benchmark.chaos.com/corona/v10/scores/561
https://benchmark.chaos.com/corona/v10/scores/562

[davetwo]

Oh cool. Just for fun I ran it on my M2max MBP (5,761,541). And my old Threadripper 2950x (5,678,105).
Quite suprised that the MBP edged it!

[Philw]

I stupidly updated to 13.4.1 after publishing that benchmark and it dropped me by 1619013 rays/sec. Why Apple?

[masterzone]

happen to me...before 4.000.000 Ray on my MacBook Pro M1 Max and 3.000.000 after the upgrade...I tried tomorrow and results has changed...very strange...

[prince_jr]

cool, thx for the first impressions and benchmarks.
can somebody run the new benchmark with a mac studio ultra m1 and m2, please?

[jojorender]

In the screenshot masterzone posted the 3,8M r/s score for I9-10910 caught my eye.
It’s running macOS 13.4… In comparison my I9-10850K running Mac OS X 10.15.7 comes in consistently in the 5,2M r/s range. Not sure what to make of this, since both CPU’s should be performing in the same range.
It’s been a while since we heard that a OS update boosted hardware performance, but I never heard of an update throttling performance.

@Phil
Is the benchmark running M chips “optimized”? I couldn’t find any Chaos statement about this and I wonder how this actually works. There is only one Mac installer. Does it have code to run “native” Intel AND “native” Apple Silicon?

@ Chaos
There was some discussion about M1 thermal throttling kicking in after 5 min or so.
Why is the benchmark restricted to 1 min tests? Would be great if this would be more flexible (set your own test time) to evaluate performance over time/ thermal stability.

[Philw]

Yep the new Benchmark is a Universal binary so as optimised as the renderer itself, I guess!

[jojorender]

Thanks Phil, forgot about the universal binary... had gpt explain to me how that works ;-)

[masterzone]

@ Chaos
There was some discussion about M1 thermal throttling kicking in after 5 min or so.
Why is the benchmark restricted to 1 min tests? Would be great if this would be more flexible (set your own test time) to evaluate performance over time/ thermal stability.

I agree...as Cinebench do...it has 10 minutes benchmark...

[ASIMO]

Hello

I am super interested in this discussion. Seeing that an Apple update that should boost power dramatically reduces the output, not only on M machines but on Intel Macs as well, is a nightmare !

I did a test with my Apple Intel i9-10910 running on macOS Monterey 12.6.5. It scores 4536424 which seems fine so far.

This means problems seem to occur with macOS 13 only ?

Also regarding the thermo-throttling – This was only true for the MacBooks ? Is there more Info ?

Would be great to see a Mac M2 Ultra performance.

[Philw]

Yep as far as I understand the CPU variability is "cleverness" due to being a laptop. The Studio is ok as far as I know (don't own one).

[jojorender]

This means problems seem to occur with macOS 13 only ?

Well, so far there is only one benchmark for this cpu/ os combo and we don’t know if this was run on a idle system…
I ran a bench on I9 /Monterey - score as expected. https://benchmark.chaos.com/corona/v10/scores/891
I don’t have any Ventura nodes to test.

@Phil & masterzone
How long after updating did you run the bench? could some background indexing be the problem?

[Philw]

Possibly - I did consider the housekeeping it does after an update. I'll try it again at some point once I can take a breath :-)

[masterzone]

This means problems seem to occur with macOS 13 only ?

Well, so far there is only one benchmark for this cpu/ os combo and we don’t know if this was run on a idle system…
I ran a bench on I9 /Monterey - score as expected. https://benchmark.chaos.com/corona/v10/scores/891
I don’t have any Ventura nodes to test.

@Phil & masterzone
How long after updating did you run the bench? could some background indexing be the problem?

Intersting question...I hit the corona bench again but I don't know how the first time I got 4 ml/ray and then 3ml/ray....🤦🏻‍♂️

[Philw]

Very changeable!

https://benchmark.chaos.com/corona/v10/scores/905

[Philw]

I noticed that these higher results happened when fans were at 0rpm and high 95°C CPU. Once the CPU temp is lower due to fans at about 1500rpm then the scores aren't as good. So definitely related to some kind of throttling curve.

[prince_jr]

@davetwo
does your m2 max benchmark-result mean that a mac studio m2 ultra will give a score around 11'4M r/s?

[Philw]

I would say its pretty linear so yes 2x

[prince_jr]

@Philw
this would be cool! :oD my maxed out mac pro 12c, 3.46ghz, 128gb (2010) scores 2.3M r/s (osx mojave).
a 5x boost would be insane!!! :o)

[Philw]

Yep I would say it’s pretty decent - especially when my 64 core Threadripper is scoring about 15M with a massive amount of power being used from the wall. Cost of living crisis and all that! (Minus the price of an M2 Ultra, of course 😉)

[jojorender]

my 64 core Threadripper is scoring about 15M

64c rippers usually in the 20M r's range... thermal throttling?

[Philw]

32 core 64 threads I meant

[Philw]

@Philw
this would be cool! :oD my maxed out mac pro 12c, 3.46ghz, 128gb (2010) scores 2.3M r/s (osx mojave).
a 5x boost would be insane!!! :o)

Looking forward to some actual M2 ultra scores though so I don’t have to speculate…

[prince_jr]

Looking forward to some actual M2 ultra scores though so I don’t have to speculate…

me, too! considering one of these configurations...
1. mac studio m2 ultra 24c, 60gb gpu, 192gb, 4tb ssd
2. mac studio m2 ultra 24c, 76gb gpu, 128gb, 4tb ssd

not sure if i really need the 76gb gpu...firstly i do photography (medium format > lightroom/capture one), then 3d-rendering (semi professional) and graphic design. in case of 3d-rendering would there be a difference between these two configurations?

[ASIMO]

Super Super curious on the benchmarks.

Is there no one with a M2 Studio Ultra or knows someone who could run the benchmark ?

[Philw]

I've asked someone I know to run it on theirs but they are away from their office right now - so sometime soon when he returns hopefully.

[Philw]

Here's the score for the Ultra...

[ASIMO]

Here's the score for the Ultra...

Thanks a lot for that update :) Super cool !

So how accurate is this benchmark ? If the benchmark result is twice as good to my current machine it will more or less match also my regular renders ?

[BigAl3D]

So the Corona site shows an example using an i9-13980HX, which is a very recent chip. It came in at 7.5 million so the M2 Ultra scores faster than that chip, but I don't think that's the fastest version. Looks like the i9-13900KF comes in at 14.5 million. Not surprisingly, the Threadripper dominates the benchmark tests. The top chip comes in at a whopping 26 million. You pay for that of course. I believe you can buy a maxed-out Mac Studio for less than just that chip. Then you need the box and a massive cooling system for the Ripper too. Depends on your situation.

Just to show how things have indeed gotten better for Mac users, I ran it on my 2017 iMac Pro 28 core, 128 GB RAM and it came in around 6.4 million. The new chips are much better, but I would have thought that after almost 7 years and new architecture, that difference would be much wider. For me, at least, moving to a new box that's twice as fast would be huge change.

[Philw]

Yep - the main reason I sit here using the M1Max Macbook Pro for the majority of my Corona/C4D work is the lack of heat, noise and power draw from the wall. Makes such a huge difference in the current energy mess.

[prince_jr]

Here's the score for the Ultra...

thanks for the benchmark. :o) so, this is it…11.0m r/s for mac studio m2 ultra!?
i’m curious, does ram (128gb or 192gb) have an additional impact on faster render times? because the benchmark test was done with (only) 64gb ram.

[Philw]

I wouldn't say so on the benchmark, no. Only in real life on heavy geometry scenes.

[Philw]

If I was buying it would be a long term investment so I'd be getting 192GB. My M1Max with 64GB is stuggling with heavy scenes.

[burnin]

... and as I gathered, RAM not used by CPU can be used by GPU... which I consider quite a smart thing.

[Philw]

Maybe :-)

Currently trying to work out what's starving my C4D/Corona/M1Max on heavy scenes and I THINK its the VRAM aspect of it. So greedy C4D viewport actually ends up starving RAM for CPU and then boom...

Ongoing investigation for me.

[burnin]

This shouldn't happen.
Likely a mem. leak... (see, if same happens w/ Blender)

[BigAl3D]

Unless I really need to see materials and colors in the viewport on a larger scene, I just turn them off until I need to see them. Makes a big difference for me.

[Philw]

It looks like it MIGHT be a combinaton of issues on C4D vs MacOS 13.x in combination with the :automated out-of-core textures" in render settings.

Still testing.

[wsiew]

Hello,
there is something I still do not quite understand. In the old benchmark, the MacPro 2019 (Xeon W-3275M 2.50GHz) achieves 12,178,400 rays. The MacStudio M1 ultra achieves 6,780,610 rays in the old benchmark. It is only half as fast as the MacPro.
In the new benchmark, the MacStudio M2 ultra achieves 11,029,942 rays (see page 2 in this thread). The MacStudio M1 Max achieves 4,753,992 in the new benchmark. The M1 ultra would be twice as fast (i.e. approx. 9,500,000).
Unfortunately there is no MacPro 2019 result for the new benchmark. How fast would it be. Would it also have 12,178,400 rays? If so, that would be about 10% faster than the M2 ultra.
I am considering buying an M2 ultra. But a MacPro 2019 Xeon W3275 would also be a good alternative. E.g. you could alternatively render on this machine with AMD cards redshift GPU.
Have I classified and estimated the benchmark results correctly?
Many greetings, Wolfgang

[Jef Teerlinck]

Just got my Mac Studio with the M2 ultra. I've uploaded the result of the benchmark to the user-submitted results for all who was waiting for that. The result is very similar to what PhilW posted before.

[frv]

Just benchmarked my Mac Studio M1 128Gb / 2Tb ssd.
9144775

20% slower than the M2 ultra posted here. 20% is also the difference in cores, 24 to 20. So the m1 or the m2 is apart from the core count more or less the same. I think Apple had to raise the cores from 20 to 24 otherwise there would not have been a noticeable difference.

I wonder if there is a difference between the 128 and the 196 model for the benchmark. The same for 60 or 76 GPU.

Looking at the benchmarks the much higher core/thread count of the i9 and the 7950 only makes a 20% difference compared to the m2. The m2 cpu could be much more effective compared to the i9 and 7950 considering the workflow before rendering.

What takes a bit out of the fun working with Apple is that you can buy a 20% faster PC for less than half of the price of a Mac Studio not even considering the much less expensive monitors you would most likely get for a PC. You can get two PC's for the same price as a Mac Studio and be over a 100% faster......or keep working on the other PC while rendering....

Just checked prices:
AMD PC 7950x cpu / 4070ti gpu is 2800€ incl. vat
20% slower rendering for the Mac Studio m2 at 6300€ incl. vat


I just benchmaked my late 2015 4 core (4000MHz) 32Gb iMac. 1533353. 6 times slower than my 20 core m1 Studio. So for the benchmark processor speed per core hasn't changed much over 8 years......I am a bit surprised actually. Apparently core count is all that matters. Hence the scores of the threadrippers.

[Philw]

For me the difference in power draw from the wall with the M1/M2 is a big deal right now.

[frv]

I wonder how much difference is that. I think you have a few years to go before you've made up for 3000€.

[Philw]

Ha could soon get to 3000 with a few Corona and Redshift renders :-)

[prince_jr]

Just got my Mac Studio with the M2 ultra. I've uploaded the result of the benchmark to the user-submitted results for all who was waiting for that. The result is very similar to what PhilW posted before.

meanwhile i ordered a mac studio m2 ultra (24c, 60 cores gpu, 192gb ram, 4tb ssd). should arrive in about 3 weeks. will post the benchmark when the new mac is ready to use.

[wsiew]

@prince_jr: You ordered a M2 ultra with 60 core GPU but not with 76 core. I asked me if there is a difference between these configurations. Do you know if the 76 core could accelerate the viewport or anything else in C4D/Corona?

[prince_jr]

@wsiew can't answer your question finally. but corona basically is cpu related. others here can provide better information than me.

to me, it was important that i have a new mac studio that i can use for a long time. so, rather more ram than gpu cores. my main work is photography with heavy mediumformat raw files (semi self-employed), cad and lastly 3d-rendering. but rendering not for a living.

btw...i don't work in the video segment, so gpu is not prior to me. and the price of mac studio m2 ultra is enormous, too. i can't afford a maxed out mac studio m2 ultra. that's why chose between 60cores/192gb ram or 76cores/128gb

maybe there are some benchmarks on the web/youtube, which focus on 60 vs 76 cores. but i guess it's not a giant leap in performance. and the m2 ultra is a hell of a machine anyway.

[prince_jr]

hi folks
here's my result running the benchmark > see attachment

it's the same speed as the other ranked mac studio m2 ultra in the benchmark results.

[BigAl3D]

This discussion makes me think about previous threads where I lamented the meaning of these benchmark numbers. Basically, it's difficult to relate them to the real-world. I couldn't find that thread, but in the past over at core4d.com, I would ask people to open the Grapes scene from the Asset Browser and just hit Render to Picture Viewer. At least it's an actual scene one might need to render and we all have access to that scene. It is not Corona, but will test the CPU and give you a render time.

For example, I just found a couple of older screen grabs of the Grapes test. 2013 Trashcan 6-core: 10 min, 3 sec. 2017 iMac Pro 18-core:  3 min, 19 sec. For my brain, I can figure out that's about a 223% increase in render speed. When I upgrade, that's what I want to see, not a meager 20%. I can use this info to guess how long a job will take to render on the new machine. That's just me.

I don't remember how the Corona benchmark works. Does it just render for a specific time and then just see how many rays it used?

[Philw]

Yep I think time must be the constant factor.

[prince_jr]

some may remember this thread here:
https://forum.corona-renderer.com/index.php?PHPSESSID=53f27ps93iqd7jar9dp9fg74nv&topic=33540.0;all

there we have a comparison of an architectural scene:
https://blog.corona-renderer.com/corona-c4d-a1-1/

i share my render now (3840x2160px, noise level 3.0%) > see attachment
i'm happy with the render time result, as 3d rendering is more of a part time job than a full time job for me.

@BigAl3D: i hope you now have an idea of ​​render times for the mac studio m2 ultra.
i wasn't able to render the grapes scene, because the scene loaded weird in the newest c4d-version. that's why i decided to use the apartment scene.

[prince_jr]

compared to my apartment rendering (mac pro 2010, 12core, 3.46 ghz, 128gb) with 152mins in the other thread, 39mins with my new mac studio is fast :o) > 3.9x faster exactly.

[ASIMO]

Hello Community and contributors to this thread

I wanted to invite you to look into a critical performance issue I am having with my Apple Mac Studio M2. See link below.
_

Maybe you are affected as well, having your M-chip performing up to 5 times slower in some scenes and not noticing it.

If you have the chance to render a a recent scene on an Intel Mac in comparison with an Apple M System I would be interested in the speed difference in relation to the Corona Benchmark results or Cinebench. Are the values/differences accordingly ?

Maybe you can also generate a basic scene (with objects and textures) that renders around 10-15 min on your Mac Studio so I can test it on my iMac i9 to have some results ? That would be great.
_

See the other post :

https://forum.corona-renderer.com/index.php?topic=41180.0

Thank you

Best – A