Is there any indication on how much performance boost can be expected though Metal-integration optimisations? I know no one can make promises, but getting an idea about what to expect would be really nice. And really help me with my upgrading decisions.
While Mac Studio ist already tempting… I suspect the Mac Pro will be a beast… but as we all see… pure power doesn’t do much if the software is not optimised for it… So either of those won’t be worth the money as long as the software (in this case of cause blender) can’t harness its potential.
From the blender metal benchmark tests I have seen, the M1 Max and M1 Ultra are only about 1/4th as fast as their processing power would allow… Watt usage also indicates this. They are simply not being used to their full potential.
If the render time would scale accordingly (I know there is a LOT more to it then computing power), the Macs would suddenly not be an expensive machine anymore… seeing the results the M1 chips deliver in Mac optimised software, this should also be possible for cycles, given a real full native metal integration.
Can anyone make a founded estimation on how this is going to develop?
Devs have already hinted at optimization gains. Compared to the 3.1 stable release, they already have over 2x speed in the dev build that they are working on, with projections for even more. I’d say that 3x is a hopeful but still optimistic achievement, with 2.5x being a safe bet.
If you compare the graphical TFLOPS power of a Mac Studio with M1 Max/32 core GPU, being 3x faster would make it comparable to the TFLOPS power of an nVidia card. The 32 core M1 Max is supposedly around 10.6 TFLOPS, the RTX 2080 (not super or TI) is 10.07.
So in this video, someone compares a 3070 mobile to an M1 Max:
3070 Mobile has 15.6 TFLOPS, and does the classroom scene in 22 seconds. The M1 Max currently does the classroom scene in 105 seconds. If we get 3x speed, thats 35 seconds. If we get more speed than that, we inch to the M1 Max having the graphical power of a card that is 50% more powerful (in purely abstract, semi-useless benchmark stats.) Remember that OptiX and ray tracing processing on the RTX cards give it a big advantage beyond raw TFLOPS numbers.
So while the $2,000 Mac Studio won’t be the best graphical powerhouse ever, the projected gains (from what little we know) should be good enough to allow for Mac users to do good work with M1 chips.
Hello guys. I noticed that on Blender 3.1 when I close it while youtube (video, or vimeo or apple tv)) is still running in the background, the Macbook pro m1pro gets a weird sound on the speaker. It only happens when i close Blender. I tried other apps (from Adobe, open source, microsoft etc) and the only one that when closed makes a sound is Blender (while a video is running in the background). Does anybody else heard this issue?
Possibly unrelated to the subject of this forum… but yes, I’ve had that issue too. When closing blender on my intel MacBook, it causes a weird scratching/glitching effect in the speakers.
I am asking because the laptop is new and I wish to know if it is a bug in 3.1 Apple version of Blender or is the Mac and I should return it while I can. Sadly I cannot test Blender in the Apple Store
Submitting this as a bug report will bring the issue to the attention of the developers.
If you make use of the ‘Help’ menu ➔ ‘Report a bug’ option from Blender, the necessary system info and Blender version will be automatically entered for you. Alternatively, you can go to this page and manually enter your info.
I’m wondering when the next revision of Apple Silicon/Metal support will go out. Jason hinted at performance gains as large as 2X the current implementation a little while ago in their “early R&D”.
How soon can we some of those materialise?
Can anyone say when the metal improvements are slated to be released? Will 3.2 or 3.3 include metal performance improvements or are they later than this? I don’t recall seeing it mentioned in the 3.2 notes.
We’re Apple fans at work but we’re about to change to PC workstations because of our view port and render time perf just not being where the team need it, which is sad but I’d like to see if this is about to change. A 2x improvement would fix this for now
Thanks
As such, it’s possible this patch could be implemented soon. However, based on these rendering meeting notes (link) it seems like some adjustments are expected to be made before the feature gets merged into Cycles. And I don’t know how long it will take to make those changes and so I can’t tell you when this patch will be accepted and merged with Blender.
I’m not completely sure about this but I believe that the 1.1x improvement already landed. Previously around 03:11 in Classroom on my M1 Pro. Now in 3.3 alpha it’s 02:53. Very small change but I think it’s worth mentioning.
