It kinda is. You don’t want people dropping 3 grand on a Mac after reading through this thread, assuming that the M1’s are these absolute beast machines without peer or compare, only to end up being slightly disappointed.
You always want to manage expectations. Keeps you from getting death threats.
yep I think its about calibrating expectations and appreciating relative improvements.
on my 2018 MacBook Pro, bmw takes nearly 6 min to render on cpu, even on blender 3.0. the 2021 machine already improves that to just over 3 min on cpu, and 43s on the max gpu. so yeah maybe I’m 4x slower than a beefy pc setup, but I’m also almost 9x faster than where I was stuck before, and I get to keep the os and form factor I like.
plus as others have pointed out, these laptops are even more competitive in other areas like video work, which I also do.
finally, I may well have switched to a pc in the past year if hardware had been actually available. all the performance in the world is meaningless if you can’t buy it outside dodgy overpriced scalpers.
Yeah, I believe you manage expectations by posting render times from benchmarks, (like I have been doing 
) .
I don’t think people will be confused about the difference between a desktop and these beast M1 laptops. If you want decent mobile render times on the go, your only option is Apple…
This sure is interesting from devtalk on an M1 Max:
I rendered a 2000 frame @ 2k resolution animation at 500 samples per frame and noticed the following:
average and approximate:
frames 1-500: 2 minutes per frame
frames 501-1000: 1:40 per frame
frames 1000-1500: 1:25 per frame
frames 1500-2000 0:53 per frame
The animation is a seamless loop, doesn’t change much frame to frame in terms of load - there was a noticeable performance gain on the longer render, persistent data enabled.
No clue what his scene looks like. None the less the speed up is not what one would expect on a seamless loop.
I am afraid you are almost 4x slower than a 1100 euro laptop with a 3060 not some beefy PC. With a consumption of 110-125w for GPU plus about 10-15w for semi-dormant CPU for render time of 12 seconds, maybe in this case is not only the speed but also energy efficiency not in Apple side. Do have a way to check your energy consumption for those 43 seconds?
your laptop is the stuff of legends, we get it 
I have a work issued pc with a 2080 in it sitting next to me, I’ll do some formal tests later, but I’m pretty sure its in the range I was comparing.
We speak of it in whispers. A machine that could render as the wind…
Probably low power draw likely considering the fans didn’t turn on, but I don’t know the figures. To compare power draw, post your times of the BMW render, GPU only, on battery.
Thanks for pushing me to it.
Tried again with my Admin Account and was able to install the Pro Driver now.
BMW:
RX 6800 without any XT – 0:27.77
(3950X CPU : 01:37.01)
Which means 5.34 times faster than my M1 Mini with Metal GPU
BTW,
Taskmanager in Win11 does not show any GPU Load while Rendering BMW …
I don’t expect one 2080 to be much better than a 3060. I tested in a friend system he has a desktop with a dual 2080 system one but in CUDA(he did not installed +370 drivers so he did not had Optix available) and it was 1 second slower than 3060 in Optix. If you want install the Studio drivers https://www.nvidia.com/Download/driverResults.aspx/184241/en-us
These Nvidia drivers are tested for Blender.
Just did that, i verified and i am in Windows default balanced power plan and did not enforced any max power option or changed anything.
In battery (Balanced power plan) : BMW 25 sec first try, 23 seconds second try. Maximum consumption 40 watts by the Nvidia so a big reduction from 110-125w… No Fan working, max GPU temp was 46ºC. It was a short test, i suspect that continuous work will drive the temp above 50ºC and moderate fan will start. Source of all this data is HWiNFO.
Good to see the M1 Max is in the same ballpark as a 3060 at this early alpha stage.
In what way?
Because it represents the floor for Apple Silicon with Blender. In engineering, it’s not uncommon to get speedups that are multiples of previous amounts. When you go to optimize, you try to find optimizations that are an entire order of magnitude or more faster than before. Then you try to find ones that are 3x-4x or something. Then you try to double performance. Then you try 50%. etc etc
Not sure where this is currently, but optimization hasn’t been the focus, so there should still be room for noticeable improvements.
Better ANE usage for denoising seems like the most obvious place to me, which would occur in the OpenImage repo and not here, but I’m sure there are other places to go for perf gains.
Already it’s more than half the performance of a 3060, which have specialized ray tracing cores. Previously you would be looking at 12 to 15 times slower.
Right now it is - M1 Max data from posted above- is 43sec(? watts) vs 12 sec(110-125watts) or 23sec(40watts). I don’t see how is this in same ballpark.
Same order of magnitude. And it’s early days…
I just played with my RX 6800 / Ryzen 3950X on PC.
BMW with GPU is 3.5 times faster than CPU.
But when I activated GPU + CPU in Blender,
this will be more than 2 times slower than GPU-only too.
So I think it is more a current Cycles(X) general problem,
that GPU+CPU is slower than GPU only,
than a special problem with Apple ARM SoCs.
Can you download HWiNFO from here and check how much power you drawing from the 2080ti ?
https://www.hwinfo.com there is a portable version that do not need instal.
I think Metal needs RT and denoising.
