What performance can a M2 bring for single core,
will it bring larger memory ?
M1 announcement was impressive and a great value.
In fact I could go on using my M1 for a few years if it would just offer
more memory. For modeling at least. Rendering on the PC.
M1 Pro, and especially Max, disappointed me with the CPU cores but
were OK for their max 64 GB memory.
And problems with core scaling began to start.
And for now, the Studio, especially the Ultra is a scaling nightmare.
If it’s really the small cache that causes it to idle it 50% and clocking
GPU down - M2 could fix it - but will it ?
I have never seen such unrealistic acceleration graph promising
and claims on a Keynote or Apples website before.
(I think similar happened at their hard G4 times but At that time
I was not yet involved)
All the tests I see, even for its home run, Video editing only, are so
underwhelming for the Ultra.
Looks like there is so much more easy and granted acceleration by
just switching from Premiere to Final Cut - or finally to Da Vinci.
If M2 will also rely on Apps to rewrite large amounts of code for
Apple only to support tiled memory, I think this will be very unlikely
to be supported by the Apps I need and use.
(Blender - with Apple’s help - may be an exception …)
And the Apple Tax, for SDD and Memory, reaches again such a
Mac Pro 2019 state, by the higher Studio’s Memory amounts, that
it gets ridiculous.
A grand for 128 GB. What will that mean for a Mac Pro 1 TB upgrade,
16 grand (?)
So for me personally, I see another 12 or 18 months until I finally can
buy a serious Apple device for 3D usage that can still fit to my budget.
Let’s hope for reasonable WWDC announcements …
But is it due to the price or the performance, or price to performance ratio?
Because the way I see it is that when Apple introduced the Mac Mini, for once they got that right. It basically was (and still might be) the best computing bang for the buck for around $600.
Where I see Apple failing is everything else. Once again, imagine a Mac Studio Ultra priced closer to $2500 – I would have bought one on day one. Even if it doesn’t quite perform to the speed of a 3090, I would still be quite happy with the price/performance ratio.
So the Mac Studio Max ends up being at the right price range, but unfortunately a bit too underpowered for someone working in CG, while the Ultra ends up being way out of range for most of us. The end result is that Apple gets less of my money and I stick with Hackintosh. Later this week I’m driving up to the closest MicroCenter to buy a 6900XT for about $1000 which should near double my render speed for a reasonable investment.
It’s ironic how almost exactly 1 year ago, I wrote this article speculating that Hackintoshing was on its way out since Apple would make it (among other things) much more financially appealing to own the real deal. Here we are one year later and somehow Apple has managed to put me firmly right back into the Hackintosh camp for the time being:
For cycles workflows and rendering I can definitely see where you’re coming from with that one. Currently the entire M lineup seems to be stuck in 2018 render performance.
That said, for eevee, and general moving around blender, the M1’s behave more like what you’d expect in 2022, and the silence is not trivial. Even my pc workstation is pretty quiet, but not THIS quiet. (writing this while rendering a 400 frame sequence with eevee on gpu/cpu)
Bcon1 - Big new features
Bcon2 - Small new features can still be added.
Bcon3 - No more new features, branching from master and gaining the beta status.
To me it’s the price to performance ratio. Looking at the test results so far, it seems that Apple’s comparison to the RTX 3090 is exaggerated marketing talk. Perhaps theoretically the Mac Studio M1 Ultra could approach the speed of an RTX 3090 if macOS / Metal support for the M1 Ultra would have fully matured, and if optimized code specifically targeted at the M1 Ultra would exist in renderers right now, but that isn’t the case, and I don’t expect that to be realized until at least one to two years from now.
Then there’s still the OpenGL matter. Apple stopped OpenGL support at OpenGL 4.1, while applications like MagicaCSG require OpenGL 4.6. That’s why the developer stopped developing MagicaVoxel for macOS as well. Right now I believe you can’t even run MagicaCSG on macOS using a WINE solution or even Parallels.
All in all, I really wanted to believe that the Mac had made a real comeback for 3D users, but in reality that is still very much a work in progress, and I expect it won’t be finished until at least the M2 chip generation has matured.
I think Blender would have a bright future on macOS. Afair there are about 6 Devs at Apple who take care for proper support on macOS (Cycles & Eevee). And the optimisations on Cycles just started a few days ago. And the comparison to the RTX 3090 might not be as exaggerated as it looks today.
Just remember how long it took until AMD Cards where really usable with Cycles.
And for me it looks like Metal support for Blender might be there before Vulcan support is present.
I think those who don’t understand why simply have different concerns and workflow logistics. I use Windows 11 more than ever nowadays, and it still feels poorly designed and clunky compared to the MacOS workflow. Having said that, my girlfriend loves working in Windows and it allows her to be considerably more productive than when she’s working in MacOS.
I have no issues whatsoever with the OS anyone chooses to work in – where I do take issue however is when Windows users argue against support for MacOS by saying that there aren’t enough of us and it takes valuable resources away from further Windows development. That is one of the main reasons why I have stopped using Houdini in favor of Blender.
It’s been established that the GPU in these new chips is not fully optimized for rendering in Cycles. The Apple devs have also stated that they will be working to do these optimizations. While the M1 family of GPUs will never reach the speeds of a 3090, I think once they are optimized we should see close to 3060 or 3070 speeds on current hardware (my guess). That’s enough for me right now. As stated above, software dev takes time. I’m just happy to see the work is being done. In the meantime I’ll just keep working away on my new mac studio, and if i need more raw rendering speed I’ll add another RTX gpu to my old pc, or go cloud.
I rather think that Apples slowly get the point, that beating a full power RTX with a tiny M1 is close to impossible. The Mac Studio is twice the scale and only get half the way. Once M1s are equal, the size, energy consumption and noise level will be equal to an RTX card. Thats just physics, all chips build in the same way by ver clever engineering.
That whole battle could have never happened, if Apple simple excepted that Nvidia is no Intel clone, but a very advanced and clever company. Apple will always have to run behind them, no matter if its performance, features or support. And I‘m not sure they keep doing that, or simple they ney to 3d.
In the case of nVidia, it isn’t a good use of physics. Their latest GPUs are horrendously wasteful of power. I hear the latest 4090s will consume up to 600W!
It’s a smart move on Apple’s part and a differentiator, but I doubt that they see nVidia as their main competitor. Intel is in their sights, and they are doing a good job of making them irrelevant.
If Apple do get serious about rendering, you can expect RT cores and at that point nVidia should be worried.
That is an incomplete way to form an opinion. Lets say that 600w value is correct for this exercise. Now lets say thet 600w give you 2 seconds render time. It means you spent 600w in 2 seconds = 1200w . Now lets say Apple M1 spends 100w but needs 30 sec render time: 100w * 30 = 3000w.
Also you can control somewhat the energy the Nvidia is spending playing with power profiles.
In my laptop max power 110-125w gives me 12/13 sec in BMW27
if i put in economical power 40w it gives me 23 sec in BMW27, so this power profile is more efficient in absolute energy terms.
What i mean is that you cannot just look at maximum energy number a card have and form an opinion form just that data point, you need also to look at what output you get from using it. And if there are even other profiles that give you better energy efficiency and noise if that you are after.
I should note that rendering in GPU is considerable less noisy than using an octocore CPU even in max GPU power.
