Mac: M3 - *Hardware accelerated RT (Part 1)

Cool. It’s good to know if developers tune their apps with Metal in mind then you can see amazing results, at least in rasterization and possibly raytracing.
When BF flipped the switch for the Metal backend all my rasterization performance went up 3 fold on my Mini. It’s amazing how well Eevee is working for me now with a Metal rewrite.

Unfortunately at the moment, I think the Nvidia’s “raytracing code” is too ingrained in the heart of Blender for anyone to be competitive in raytracing against Nvidia no matter how well your GPU performs, even AMD’s latest desktop cards.

With the scores we are starting to see with the new M2 Max (Laptops) popping up we can see that Macs are already well suited for working in Blender, not the fastest by any means but the over all performance makes them a very competent device for Blender professionals. Hopefully the optimizations for Blender continues for more years to come, and we see a complete rewrite that is designed 100% for AS and Metal.

2 Likes

True while Cycles on the GPU was / is nice.

The Metal backend replacing OpenGL helps so much more overall.

The M2 Ultras might even match the AMD 7000 series till they get RT support in Blender.

2 Likes

Not bad!
On my 2060S, Classroom, on CUDA is 1 min 02 secs, on Optix is 33 secs, vs your test at 49 secs!
I believe that Mac Studio will be very very tempting.

I just had to render tonight-today, in segments, - just to have the render going when I was not using the PC - a 1000 frames animation, at a range on 35 to 55 sec per frame, for a total of ~12 hrs.
I almost never do animations, and the bottleneck was the 8GB VRAM, that had me scrap scattered objects, and vegetation (ivy on walls); and editing the scene was a pain.
So, more RAM and more responive scene are the two things that would make me prefer the Mac, over render times.

5 Likes

For those artists interested:
My daily job is being an engineer and i do work in 3d CAD software. Huge models, hours per day, butter smooth experience, no rendering though.
I sold my personal microsoft i7 11gen 32gb ram rtx 3050ti laptop and got an m2 max 30 core.
Initially I got the14 inch pro 12 cpu 16gb ram model but i found that most of the time it was using the ssd because the ram was full. So I sent it back and got the 14inch 32gb max.

While working i love:

The display and how it is consistent in the colours in apps, and while switching. The display of the 14inch pro is awesome compared to microsoft surface and hp zbook, but i think it is also an issue of windows 10/11 that colors are different in different programs… Also what I see in Blender is exactly the same in adobe and safari/chrome browsers. Of course on my iphone everything is more saturated by a few % and on samsung phones is by quite a lot. I love “natural” colors more than saturated.

The Apple silicone is more than 10 times faster in showing eevee viewport. My 32gb i7 rtx3050ti would take even 1 minute to load all textures. Here it takes 3-5 seconds aaand - i can work in “butter smooth” fps. Blender does not show fps, only in animation, but as a guy who likes to model and move objects, draw on them etc, the laptop is responsive to input, and generally faster than the alternative (surface laptop with 3050)

The m2 has more cpu power so yeah is faster in renders, but I did not buy the laptop to render for hours, but to work and get renders in between.

Here is a “big” scene that i work on:


On the Surface RTX 3050ti i would get: 2 minutes and 1 second best time, without preserving data from previous render (if I do that, i have no more Vram to do anything, even the mouse pointer is slow. If I do a second render it will be 1:40 best). The 32gb ram and 4gb vram is just not enough, the laptop has a hard time and while working is possible, is just torture as everything is slowed down.
On the macbook pro 12cpu 19 gpu 16 gb ram
The render is almost the same at 1:25, but consistently getting there, and without video ram error. While rendering animation the rtx laptop would run out of memory or it will jump from 1minute 36 seconds to 2 minutes and 30 seconds max (when it was hot, and cooler at max). This is a huge plus for the mac because I can work and render with best optimisation available in 3.4.1 consistently. The problem I had was that half of the 3D scene was on the ssd because the ram was full, and even so while working, in edit mode, changing textures and more, this macbook was a lot faster and fluid than the 3050ti. It seemed perfect to work but i feared that using the ssd as ram will not be ok. But it was great to work on it, snappy, responsive, fluid.

On the macbook pro 12 cpu 32gb ram and 30 gpu, while working it feels a bit faster that the 19gpu model, it loads the textures even faster and editing using eevee is better but there was no big difference. The previous was good this one is good also. However I don’t put stuff on the ssd anymore so i expand the life of that component. The render is 1 minute long so a lot faster than the 3050ti and it is consistent. If i listen to music it will go up a bit and the laptop get warmer (far from hot), but I can better enjoy waiting the pixels to get in place.

If there is a mac to recommend, the M2 is the first one. Best value would be the 12 cpu 32 gb ram 19gpu and 512gb ssd. This is a 500€ save. The extra gpu power is useful perhaps in the future when and if Blender gets better in this regard. I did not make a custom laptop because is harder to find a buyer in a few years. People usually buy standard configs and not personalised .

Any questions? Ask!

10 Likes

Pablo always said in Blender today that Vulkan switch itself will not
provide any noticeable speed improvements.
As the graphic demanding parts are already pretty optimized and
Vulkan switch in large parts will relate to GUI things and such.
It is mainly for getting future proof, make administration easier and makes
it easier to access/dock other APIs like Cuda, …, Metal, … to Blender.

So if Vulkan indirectly helps acceleration then maybe by Blender
adding some newer advanced graphic calls in its code or that
3rd parties like Metal can find more docks which they can further
“metalize”.

Not sure how well that will work for Mac Os sure hope it does not make things worse again as there is no Vulkan in Mac Os.

All there is and not sure how complete or how well it works is a MoltenVK, which is a “non official / non Apple” translation layer.

Unless Mac OS 14 gets native Vulkan support but I doubt it. The could codename it Mac OS Vulkan the. :wink:

I think this is pretty unlikely.
But I expect Apple to just prepare Blender to get everything relevant
done by Metal, for macOS.
I don’t think Apple came only make Blender native ARM and Eevee
and Cycles Metal accelerated, to stop there and watch potential
lags bottlenecks in Object handling or GUI, negatively impacting
all previous work.

In the end it is 1000 times better for us that Apple makes Blender Metal
ready, than some communities would try to make a final Vulkan switched
Blender work via MoltenVK.

1 Like

Oh, everything LTT OK as always …

Forget about Vulkan and MoltenVK. It make no sense at all for macOS. The topic of Vulkan/Molten came up before Apple started working on Blender Metal.
It is not only about “API”. It is also about Apple hardware <–> software optimisations.

1 Like

Yes there was a time not to long ago were GPU renders on Mac Os was totally gone (well cycles).

They sure turned it around nicely.

Beside the VRAM limit which is serious limitation in Windows until Nvidia feels threatened you bought a Microsoft Surface. Those laptops are not made for performance with 40w normal 50w limit GPU power while the 3050 card can go to 90w…
But your case reinforces that strategically the great advantage of Apple is the unified memory. Until Nvidia splashes 16-96GB VRAM in all cards…

It might be a ‘nice to have’ on MacOS. If a software is not Metal savvy, eg. Luxcorerender, and the only way to use it is with translation (Molten VK) or any other, I don’t mind having a Vulkan backend.

1 Like

I really hope that Apple doesn’t bump up the M2 Mac Studio prices (or if it does that it increases the base model drive and RAM). At the moment I’m looking at what I think would work for me in the M1 Studio line and it comes out to $2800…expensive but doable. I’m not going to buy it…but if in the Fall (or ideally earlier) they come up with an M2 version and the price for that configuration runs out the same, I will probably buy one.

1 Like

Next Nvidia card definitely needs 24 to 32 GB. 16 GB is good for most, but not enough, but so most high end cards have not more.
Anyway the price for memory Apple wants for the jump from 32 to 64 is insane.
Ok, at least I know that next time a Mac becomes interesting it must have at least 128 GPU cores in the base model.

Or we get hardware RT acceleration and a 3nm process.

2 Likes

Will FOSS soon be coming to the iPad, this means Blender, Krita, Godot, ect…?
US Commerce Dept demands Apple, Google permit more stores on platforms (gamedeveloper.com)

In short, the US Government is now expressing an interest in tearing down Apple’s walled garden and making iOS an open platform. Now of course there will definitely be legal challenges as this could be construed as attempted government micromanagement of private enterprise (like what China and Cuba does), and thereby overstepping their constitutional bounds.

1 Like

That would be great !

Somehow Apple managed the we will not die from malware
with the more “open” macOS so far too so far …

But why just target Apple with this initiative? If we are now in the business of breaking walled gardens, then shouldn’t Xbox, Playstation, and Switch be included as well (so as to allow the use of any game engine as long as there is proper compatibility and as long as the creator has the official dev. box)? It would help Godot users mainly (as it is one of the few big 3D engines outside of the enforced Unity/Unreal duopoly for indies).

2 Likes

OK, here are my Mini M2 Pro results (19 core GPU)
I only have 1920x1080 monitor here so Tree Creature a bit faster than 4K as requested.

Tree creature - 15-16 fps

Monster under the bed 2m05s

Party Tug
Shader compile 11s
First go 13.28s
Second go: 6.27s

Classroom 1m34s

Lone Monk 12m02s

4 Likes

They should …

I am just not experienced with those or having priority for me …