Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

mi7chy

macrumors G4
Original poster
Oct 24, 2014
10,623
11,295
In theory, when doubling the CPU or GPU cores you'd expect 2x performance scaling in a perfect world but lets look at current real performance scaling to estimate M1 Ultra 20CPU 64GPU performance using Apple's funded and co-developed multithreaded Blender.

Current Blender BMW CPU render times with doubling of performance cores:

MBA M1 4P+4E CPU 5min 51secs
M1 Pro/Max 8P+2E CPU 4min 11secs
1.4x real performance scaling vs 2x theoretical perfect scaling (~2min 56secs)

Current Blender BMW render times with doubling of GPU cores:

M1 Pro 16GPU 1min 18secs
M1 Max 32GPU 43secs
1.81x real performance scaling vs 2x theoretical perfect scaling (39 secs)

Estimating M1 Ultra 20CPU 64GPU performance:

M1 Pro/Max 8P+2E CPU 4min 11secs
M1 Ultra 16P+4E CPU with 1.4x scaling 2min 59secs vs 2x theoretical perfect scaling (2min 6secs)
Comparable to 12600K at 2x and 5600G at 1.4x.
https://openbenchmarking.org/test/pts/blender-3.0.0

M1 Max 32GPU 43secs
M1 Ultra 64GPU with 1.81x scaling 24secs vs 2x theoretical perfect scaling (22 secs)
Closer to 16secs 70W mobile Nvidia 3060.

Update: Reviews and benchmarks are in 3/17/2022 so no more guessing.

M1 Ultra 20CPU 64GPU 128GB

CPU on par with 12600K
GPU half of laptop 70W 3060

The Tech Chap
1647531920294-png.1975123
 
Last edited:
  • Haha
Reactions: MayaUser

ader42

macrumors 6502
Jun 30, 2012
436
390
Stating that 4P+4E is twice 8P+2E is nonsense.

> 1.8 scaling is fine.

happy to wait for multiple real verifiable benchmarks and see real workflows - the viewport performance for even the M1 Max from what I have seen is great.

Blender will likely do further optimisations but I don’t personally care about cheap / free open-source software like Blender. I look forward to comparisons using professional software.

I’d rather have a great experience overall and have slower final render times than the opposite. I get that not everyone can afford Macs so it would be great if you could provide a link to a PC build that equates to the Mac Studio M1 Ultra. Perhaps with a 3970x and an RTX 3080?

Heck even if did use Blender I’d simply use CrowdRender and have a Mac renderfarm to avoid windows - I have multiple Macs sat doing nothing when I use one of them.
 
Last edited:

mi7chy

macrumors G4
Original poster
Oct 24, 2014
10,623
11,295
Blender also runs on Linux and is said to be even faster so what does it have to with Windows? And, multi-GPU support is built into Blender.
 

JimmyjamesEU

Suspended
Jun 28, 2018
397
426
Blender also runs on Linux and is said to be even faster so what does it have to with Windows? And, multi-GPU support is built into Blender.
What a waste of time. Why bother estimating what will be verified within a day or two? Doubly so if you ignore actual fair benchmarks for a cherry picked set of tests universally acknowledged to be under optimized for Apple Silicon.

The real reason for this nonsense is clear to almost everyone here.
 

l0stl0rd

macrumors 6502
Jul 25, 2009
483
420
In theory, when doubling the CPU or GPU cores you'd expect 2x performance scaling in a perfect world but lets look at current real performance scaling to estimate M1 Ultra 20CPU 64GPU performance using Apple's funded and co-developed multithreaded Blender.

Current Blender BMW CPU render times with doubling of performance cores:

MBA M1 4P+4E CPU 5min 51secs
M1 Pro/Max 8P+2E CPU 4min 11secs
1.4x real performance scaling vs 2x theoretical perfect scaling (~2min 56secs)

Current Blender BMW render times with doubling of GPU cores:

M1 Pro 16GPU 1min 18secs
M1 Max 32GPU 43secs
1.81x real performance scaling vs 2x theoretical perfect scaling (39 secs)

Estimating M1 Ultra 20CPU 64GPU performance:

M1 Pro/Max 8P+2E CPU 4min 11secs
M1 Ultra 16P+4E CPU with 1.4x scaling 2min 59secs vs 2x theoretical perfect scaling (2min 6secs)
Comparable to 12600K at 2x and 5600G at 1.4x.
https://openbenchmarking.org/test/pts/blender-3.0.0

M1 Max 32GPU 43secs
M1 Ultra 64GPU with 1.81x scaling 24secs vs 2x theoretical perfect scaling (22 secs)
Closer to 16secs 70W mobile Nvidia 3060.
If they manage 22 sec with the 64 core in blender it is not bad.

I bet they could easily get it under 20 with some optimization, which is still open according to the dev task.

I am curious how much the double memory bandwidth will affect things.
 

mi7chy

macrumors G4
Original poster
Oct 24, 2014
10,623
11,295
I am curious how much the double memory bandwidth will affect things.

Memory latency is also a consideration. Don't seem to recall the memory bandwidth increase from 16GB to 32GB to 64GB having much of an impact so my guess would be little to 128GB. However, the wildcard in performance scaling is the interconnect bus between the two M1 Max SoCs. Don't foresee much impact on the GPU side but not so sure about CPU. It'll be interesting to see on the 18th.
 
Last edited:
  • Like
Reactions: l0stl0rd

MayaUser

macrumors 68040
Nov 22, 2021
3,178
7,199
I am curious how much the double memory bandwidth will affect things.
i suspect that you refer between the M1 max and M1 ultra bandwidth
yes, i suspect to be a leap since m1 max has 400 and m1 ultra 800
 
Last edited:

MayaUser

macrumors 68040
Nov 22, 2021
3,178
7,199
is there a difference in memory bandwidth between M1 max 32gb and M1 max 64 ??
NVM she edited her post
 
Last edited:

l0stl0rd

macrumors 6502
Jul 25, 2009
483
420
i suspect that you refer between the M1 max and M1 ultra bandwidth
yes, i suspect to be a leap since m1 max has 400 and m1 ultra 800
Yes that is what I mean, the connection between both max chips should not matter as the bandwidth of that is 2,5 TB/s.

But the Ultra having 800 GB/s that is 3080-3090 level memory bandwidth.

The 400 of the Max was more like 3060 level.

I guess we will se in a few days.
 

sunny5

macrumors 68000
Jun 11, 2021
1,838
1,706
Here's the problem:

There aren't many software to prove that Apple Silicon is powerful
. Video editing? They heavily benefits from media engines and ProRes encoder/decoder. Benchmarking? Doesn't really provide actual performance in real life.

At this point, Apple Silicon cant really prove its performance well especially with GPU unless many high end software supports and optimize well on Apple Silicon.
 
  • Like
Reactions: Sopel

l0stl0rd

macrumors 6502
Jul 25, 2009
483
420
Here's the problem:

There aren't many software to prove that Apple Silicon is powerful
. Video editing? They heavily benefits from media engines and ProRes encoder/decoder. Benchmarking? Doesn't really provide actual performance in real life.

At this point, Apple Silicon cant really prove its performance well especially with GPU unless many high end software supports and optimize well on Apple Silicon.
Yes sadly true, not many 3d apps that are well optimized and well games are kinda non existent and most not optimized for M1 either.
 

JimmyjamesEU

Suspended
Jun 28, 2018
397
426
Here's an interesting tidbit about performance and Apple Silicon. It's from the arstechnica forum (with the original post from a Redshift page on facebook, and also from the Maxon forum).

also

A very high end render scene 'Moana'.

The M1 Max completes the scene faster than lol but a 3090. Faster than a 3080, faster than 2x2080ti.
It's not video editing and it can be (with optimisation) very fast indeed.

edited to correct the mistake that the M1 Max is faster than a 3090. It should have said “faster than all but a 3090”.
 
Last edited:

Homy

macrumors 68030
Jan 14, 2006
2,507
2,459
Sweden
Here's an interesting tidbit about performance and Apple Silicon. It's from the arstechnica forum (with the original post from a Redshift page on facebook, and also from the Maxon forum).

also

A very high end render scene 'Moana'.

The M1 Max completes the scene faster than a 3090. Faster than a 3080, faster than 2x2080ti.
It's not video editing and it can be (with optimisation) very fast indeed.
Very interesting! Regarding 3090 it's faster than M1 Max according to those posts but Ultra will outperform 3090 as Apple stated. :)

2x 2080ti = 34m:17s
M1 Max = 28m:27s
Single 3090 = 21m:45s
2x 3090 = 12m:44s

My guess for M1 Ultra 48c is 18m:58s and for 64c 14m:13s.
 
Last edited:

sunny5

macrumors 68000
Jun 11, 2021
1,838
1,706
Here's an interesting tidbit about performance and Apple Silicon. It's from the arstechnica forum (with the original post from a Redshift page on facebook, and also from the Maxon forum).

also

A very high end render scene 'Moana'.

The M1 Max completes the scene faster than a 3090. Faster than a 3080, faster than 2x2080ti.
It's not video editing and it can be (with optimisation) very fast indeed.
I only see words instead of actual test results. Any idea how they tested?
 

tomO2013

macrumors member
Feb 11, 2020
67
102
Canada
Hi Mi7chy,

There is a common theme to this post (and other posts that you have made) when the M1 Pro/Max were announced.

I tried to help you at the time during the launch of the M1 Pro/Max my post got lost in the sea of responses that you previously received that you didn’t get around to / forgot to respond to me so I’ll ask again here.

Blender:
Would you mind sharing some samples of your professional/hobbyist 3d work that is dependent on Blender and perhaps share with us regarding how flexible your workflow is?


I ask for a number of reasons…. there are some ’small’ details omitted from your original post suggesting that you may/may not be aware that blender on M1 may not be an ideal platform for you…

1. Apple only started to fund development of a Metal Apple Silicon port in October 2021. This is a very recent thing relatively speaking, even against the life span of the M1 chip itself. It’s a very very early days optimization build. An expectation for performance parity even at this stage is IMHO quite unrealistic especially given how different the architecture is between RTX and M1 at a rendering level (immediate rendering vs tile based, dedicated ray trace light ray intersection hardware VS no dedicated hardware etc…).
2. Based on the release notes for blender 3.1 and comments from Brecht Van Lommel, the cycles metal backend still has a ways to go in terms of bug fixes and performance optimizations. I.e. Blender 3.1 is somewhat of a performance baseline but already showing more than 3X improvement over the initial M1 CPU only rendering. Will it ever close the gap with rtx dedicated ray trace hardware - unlikely in my opinion on current generation silicon, not until IMG CXT ray tracing IP makes it’s way into future M1/M2/Mx apple silicon chips. Similarly I don’t see x86 Intel/Amd and Nvidia RTX catching up on the video side of things on a Red Raw 8k footage handling, h.265, pro-res export until dedicated hardware is provided.
3. Many of the 3d benchmarks used to convey the superiority of x86 in Blender are actually evidencing the NVidia proprietary CUDA or Optix code paths.

If you could share some samples of your professional (or even better) your hobbyist work the community here may be able to help you find a faster, better optimized alternative for your needs that is available today VS blender :)

Stockfish:
I‘d sincerely never heard of this chess engine before the forum around here. Clearly this is also very important to your workflow. I’ve never seen any of the mainstream review sites utilize it as a metric for general mainstream compute performance in the same way that H.265 export, ‘can it run crysis’, etc.. are utilized.
Nevertheless, whether mainstream review sites use it or not is really irrelevant if your workflow is heavily depended on this chess engine and stock fish will be used to determine your purchasing decisions.
From what I have seen and comments online, I cannot help you here. Stockfish looks to be heavily dependent on handtuned x86 SIMD. I can see from this thread that there have been huge performance gains already since the original complaints from Leifi and yourself : https://forums.macrumors.com/thread...-performance-on-m1-by-up-to-80.2326552/page-6
Given the very niche nature of such an engine however, I think that you would be better served staying with x86 or at the very least owning both x86 and Apple silicon and use both for the software where each can show their legs.

General commentary in the interests of assisting you:
I have observed a common theme to these threads (quite repetitive) around here especially when there are many many many many other more mainstream utilities applications that we could be discussing that are actually used in a production capacity by professionals in the target demographic for a MBP or Mac Studio. :)
I’m getting an impression that you are very committed to finding a solution/justification for the purchase of an M1 for your specific use cases and keep finding it very difficult to justify the cost/performance/benefit over your AMD/Nvidia laptop as it stands immediately today. That is cool - different platforms suit different use cases. I really don’t think the Apple platform is for you. Personally I have an Intel Xeon workstation with Quadro graphics and an M1 max in the home office. I use each for tasks where they are both best suited. The majority of the time (from a performance perspective) I’m on the Mac but I do still have a need to spend time on the Xeon/Quadro for certain niche use cases where the time savings more than justify the cost of having both.

However if you really are interested in staying on the Mac OS ecosystem, then I think you really need to:
1. get involved over at blender and stockfish (if you have the coding competencies) and contribute to a Mac OS optimized build.
This is a nice introduction to ray trace acceleration on metal https://developer.apple.com/documen...ceshaders/metal_for_accelerating_ray_tracing/
2. Could you make use of any native Use Mac OS M1 optimized apps where it leaves alderlake/Ryzen + NVidis in the dust e.g. video editing 8k h265 content in FCP / Resolve, Red Raw footage timeline handling, music production in Logic Logic, Affinity Photo for photo editing/graphic design, Compilation speed e.g. use x86 where it excels - gaming, stock fish, blender. The last two are two of my personal big drivers for utilizing M1 over x86.
3. Consider a different workflow (different renderer to blender cycles if that is an option for you) - hence the ask to see samples of your work?

If this still is not helpful to you… it might be best to take a break from the Apple Silicon forum because I think your points have been made, heard and debated ad-nauseum by the community with increasing regularity.
This could be perceived by some as gaslighting however giving the benefit of the doubt that actually this is not intended as a gaslighting post, then maybe check in again on the forum downstream in 6 months to a year and see if the state of play has changed on stockfish and blender?
At this juncture based on what you have shared with us here, I don’t think that Apple Silicon and the software ecosystem is for you… and possibly also not this forum, but that is ultimately for you to decide (or a mod :) ).

Best Regards and stay healthy and happy.

Tom.
 

l0stl0rd

macrumors 6502
Jul 25, 2009
483
420
459DFB43-4F0E-451A-A948-42A163403367.jpeg
I expect the ultra to do really well in redshift.

6800 XT
B51AB5BE-1F17-4398-A20B-29B2536133DE.jpeg


If it scales as expected it will beat the 6800 XT.
 

Homy

macrumors 68030
Jan 14, 2006
2,507
2,459
Sweden
Hi Mi7chy,

There is a common theme to this post (and other posts that you have made) when the M1 Pro/Max were announced.

I tried to help you at the time during the launch of the M1 Pro/Max my post got lost in the sea of responses that you previously received that you didn’t get around to / forgot to respond to me so I’ll ask again here.

All I have to say is don't waste your time. This been going on since 2020 and the launch of M1, not only M1 Pro/Max/Ultra. There are a group of people who do this with every new Mac. The OP had a M1 before in their signature which they tried to sell and now seem to have sold but even after that they hang out here and go on and mention Stockfish for example as a benchmark tool. There have been huge threads about Stockfish but as you can see these people don't care about counter-arguments and logic. The OP did the same about gaming on the original M1 where they used Windows games benchmarks not even available for Mac as a sign of M1's poor GPU and the superiority of AMD Vega 8 iGPU. Don't waste your time!
 
Last edited:

JimmyjamesEU

Suspended
Jun 28, 2018
397
426
Thx but I dont see the result that M1 Max is faster than 3090. I only see Apple M1 Max 64gb = 28m:27s which is slower than 3090 but still impressive.
Apologies. My post should have read “faster than all but the 3090”. I will correct it now. It was faster than a 3080 and 2x2080ti.
 
  • Like
Reactions: Argoduck

sunny5

macrumors 68000
Jun 11, 2021
1,838
1,706
Very interesting! Regarding 3090 it's faster than M1 Max according to those posts but Ultra will outperform 3090 as Apple stated. :)

2x 2080ti = 34m:17s
M1 Max = 28m:27s
Single 3090 = 21m:45s
2x 3090 = 12m:44s

My guess for M1 Ultra 48c is 18m:58s and for 64c 14m:13s.
Well, if you do 3D stuff, then they def gonna use multiple GPU so in reality, Apple Silicon wont be able to compete with them. PC can put 4x RTX 3090 while Mac Studio has only one GPU. I hope Mac Pro allow multiple graphic cards with 128 GPU cores tho.
 
  • Haha
Reactions: iBug2

JimmyjamesEU

Suspended
Jun 28, 2018
397
426
Well, if you do 3D stuff, then they def gonna use multiple GPU so in reality, Apple Silicon wont be able to compete with them. PC can put 4x RTX 3090 while Mac Studio has only one GPU. I hope Mac Pro allow multiple graphic cards with 128 GPU cores tho.
So now we’re in a situation where it’s no longer good enough to beat the best gpu, it has to beat an infinite amount. Seems reasonable.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.