Mac: M1, M2 - Metal 3 <---

A post was split to a new topic: Issue with sculpting

The part II of the Barefeats tests:

I’ve noticed an interesting psychological effect. The more M1 MAX tests I watch, the more my beloved MacBook M1 slows down in my daily work.


If you want 32gb of ram, it makes sense to upgrade the to the M1 Max becuase that CPU/GPU upgrade includes the 32gb of ram upgrade…

1 Like

but they are still charging for it…

I wouldn’t do that for 2 cores. It’s not that much gain, and the binned chip is actually cheaper. 300 dollar for 2 cpu and 2 gpu cores? Is that worth it?

Could someone with the new Pro/Max machines do LuxCoreRender GPU (+CPU) benching please?
0. (attention: links initiate download except last one)

  1. Download and install Blender 2.93.5 INTEL
  2. Download and install latest LuxBlend Add-on 2.6alpha
  3. Download Benchmark scene, use 256 samples, SOBOL progressive sampling and no denoiser because of Rosetta penalty.
  4. Run Path CPU, Path GPU and Path GPU+CPU tests. If you could provide wall power metering in Watt during the main render that would be awesome!

My MacBook Air vs Win PC.

In the end the power efficiency is not that different.
My MacBook Air renders at 1Wh CPU+GPU compared to 1.67Wh on the PC.

Does Lux support Metal? I don’t think it does, so you can’t test the GPU. Unless maybe there is some OpenCL rendering going on. Not sure if M1 support OpenCL anymore. ?

1 Watt Hour? Sign me up!!! The 16 MBP has a 100wh battery.
I think you may have made an error and mixing up terms. 1Wh would mean you can get 100hr worth of rendering on one charge. :nerd_face: :stuck_out_tongue_winking_eye:

Yes, LuxCore GPU renders with OpenCL on M1, CPU via Rosetta.
While deprecated since 2018 still works like a charm!
The main rendering of said scene took 2 mins @ 29W, that’s 2/60h x 29w ≈ 0.001 kWh.
PC was 15s x 400W ≈ 0.00167 kWh for the same render job (w/o initial scene setup).
Talking energy used per render job here (disregarding time).

1 Like

My computer consumes 1.3 jiggabits per squared cubic yottabyte of ramules.


mine consumes 1.21 gigawatts and travels back in time.

1 Like



I knew the API was still in previous MacOS, wasn’t sure if it made it over to Monterey.
Good to see someone is still using OpenCL. I know it goes away in Blender 3.0 and I think, don’t quote me, Octane has already removed it. If not removed I remember Jules saying OpenCL is dead.

1 Like

While the mighty Apple has their engineers helping out the big render guys like Maxon, Otoy and now Blender Foundation, lil indie developers like the LuxCore Team don’t have the manpower for a Metal dev branch, let alone the funding :frowning_face:

1 Like

kinda sad. i remember back in 2009 when snow leopard released and apple made a huge deal of open cl. the idea of gpu rendering was mind bending. then of course cuda caught on as the de-facto standard, and apple ditched nvidia and consequently ended up being the platform largely left out of the gpu rendering party. hope it goes better this time around. :sob:


openCL development was really slow and Apple was also fed up with it.

NVIDIA we know wants to own the world - And Intel has had issues with producing the CPUs Apple wanted

thus their move makes sense.

but yeah that does not make it better for project like Lux - I really like the engine

Ok, take this with a huge grain of salt, but Pablo’s (Blender Today) previous subtle hints about Blender on Macs have all come true.
They’re usually proposed as questions and ultra quick, and he then quickly moves to the next subject. i.e. the Metal hint he dropped a few months ago while talking to Brecht on the stream.

Today he dropped another quick hint posed in the form of a question, as always.
While talking about font storage, he posed the question of, “fonts on iOS?”.

If we have fun with the rumor/hint (and remember to have fun with it) does this mean Blender for iPads is being worked on?


Hey guys,

Just wanted to quickly jump here and mention that you can get considerably better performance in your scenes with the M1Pro/Max by rendering with 256 or 512 blocks instead of the default 128. Please note that the Redshift Benchmark always renders with a 128 block size.

The larger block sizes help reduce some CPU-GPU communications which appear to suffer on the Metal API. This is why you see some timing be larger on macOS compared to Windows or Linux (close to 0.1ms on the former versus 0.01ms on the latter). We recently added some optimizations that cut down these communications (and which also helped performance on NVidia) and do have plans on adding even more optimizations to help with this issue.

Hope this helps! :slight_smile:


Redshift forum.


OpenCl still seems to work with the Radeon Pro render under Rosetta2.

I did try a GPU only render and it worked with 100 load on the GPU cores :slight_smile:


It’d be nice having a pared down version of Blender aimed primarily at sculpting and/or texture painting, but…I dunno. Even with Blender getting some of Apple’s love and attention, the chances of us seeing an iPad rev are pretty slim.

We’ll always have Sidecar.


So you think just the RAM up tp 32GB and leave the chip or if I upgrade the chip go big?