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

Had some time this afternoon to test out the new 3.1 alpha on some demo scenes. (very excited, thank you Apple and Blender devs). I didn’t compare to any previous GPUs (laptop or desktop) from OpenData etc, as they don’t really compare with Cuda and Optic etc. So all results are from Cycles CPU 3.0 and 3.1 only, as that’s what we had before.

Specs:
16in MBP - M1 Max - 32core GPU, 64gb RAM
High Power Mode = ON

——
Blender 2.92 Splash Screen (Dinosaur)
CPU - 15m 32s - (Blender 3.0)

CPU - 15m 95s - (6% slower than Blender 3.0)
GPU - 6m 43.02s - (147% faster than CPU)
GPU+CPU - 6m 50.33s - (2% slower than GPU only)

——
Nishita Sky Demo
CPU - 58.17s - (Blender 3.0)

CPU - 58.75s - (1% slower than Blender 3.0)
GPU - 19.76s - (197% faster than CPU)
GPU+CPU - 18.35s - (8% faster than GPU only)

——
Classroom
CPU - 7m 24.85s - (Blender 3.0)

CPU - 7m 52.97s - (6% slower than Blender 3.0)
GPU - 1m 48.10s - (338% faster than CPU)
GPU+CPU - 1m 42.09s - (6% faster than GPU only)

——
Barcelona Pavilion - CPU & GPU Benchmark scenes
CPU - 59.44s - (Blender 3.0)

CPU - 57.62s - (3% faster than Blender 3.0)
GPU - 17.51s - (229% faster than CPU)
GPU+CPU - 15.53s - (13% faster than GPU only)

——
Monster Under The Bed
CPU - 11m 17.65s - (Blender 3.0)

CPU - 12m 24.60s - (9% slower than Blender 3.0)
GPU - 3m 08.77s - (294% faster than CPU)
GPU+CPU - 2m 56.80s - (7% faster than GPU only)

——
BMW - CPU & GPU Benchmark scenes
CPU - 3m 14.77s - (Blender 3.0)

CPU - 3m 13.52s - (1% faster than Blender 3.0)
GPU - 43.89s - (341% faster than CPU)
GPU+CPU - 41.41s - (6% faster than GPU only)

Observations:
I had the MBP plugged in all day, and during these render tests, it appeared to use 20% battery whilst being plugged in - almost like it was using the battery faster than it could charge it. Never seen that before, might be because of High Power Mode, but need to test further.

Edit: Now with more accurate percentages ((old-new)/new*100%)

6 Likes

Thanks for taking the time to do these. Awesome times! :metal:t2:

3 Likes

my math might be failing me right now, but generally your results seem to place the gpu roughly 3-4x faster than the cpu, whereas you’re translating that to ~100% faster, which would suggest only 2x faster?

3 Likes

Just an fyi, the formula for calculating how much faster something is percentage wise is (old-new)/new * 100% (speed is proportional to the reciprocal of time)

So in the monster under the bed scene, the GPU is 296% faster than CPU
(744s-188s)/188s * 100%

6 Likes

Question that is very non blender sorry guys.

Can you verify something for me please:

Is the 14" only about 8,9 or 9 cm there ?

Yup, just measured. Around 8.9 cm.

1 Like

Great thanks for confirming, I think I will return the 16" then, find it uncomfortable :frowning:

That about the same as the air which I preferred for typing.
This is 9,1 cm is anyone is curious.

1 Like

A bit off-topic, sorry, but too remarkable to not mention here: Maxon acquires Pixologic. :open_mouth:

I wonder if this will accelerate ZBrush being optimized for Apple Silicon, as Maxon is known for focusing on Mac users. I’m afraid ZBrush will become subscription-only though, like Redshift. :slightly_frowning_face:

Join the discussion here if you like:

2 Likes

2 x GTX 1070 Ti Cuda
30 seconds

1 x GTX 1070 Ti
50 seconds

M1 MacMini GPU
110 seconds

I am not that mad about the macMini GPU anymore …

3 Likes

Ooh not are they not going to update my perpetual license anymore?

On a side note I am sending back my 16” 🥲

1 Like

I went in to the apple store to see the 16” and couldn’t imagine getting away with it in cafes…. And then the typing experience on my forearms was not good.

To me the 16” looks comically oversized vs the sleek composition of the 14”. I can entirely understand the many reasons one would choose the 16”, however. Not trying to discredit.

Thankfully my eyesight is great so turning scaling to more space is just fine for me :slight_smile:

2 Likes

I have slight astigmatism, so I’ll probably go with the 16".

3 Likes

Yes so many thing I thought it would be fine, but find it uncomfortable, heavy and not liking my choice to go with 16 Gb of RAM bot sure what I was thinking.

As for the screen I can live with it the air was ok even with my slight corrective glasses.

So I am taking the occasion at the same time to bump it up to 32 GB, that will stop me from freaking out leaving some other apps open while working in blender.

Not sure how I feel about increasing the GPU count however.
I am very tempted to take the 24 core, as I will go for 32 GB that not a big jump anymore.

But slow slightly over budget as I wanted to stay at 3kish like Renzatic.

1 Like

Yes…in many ways the 16 such a better value considering the larger higher resolution screen, bigger battery, better cooling, better speakers (better mic too in my comparison)…its just a better overall buy for the money. Its just too big which is a shame because I want to have all that other stuff :slight_smile:

3 Likes

Ah, I might have got my formula wrong. Thanks for this, shall update my post.

2 Likes

well said I agree.

Except the screen but I cant say as I have not used a 14” yet.

Not that noticeable in Blender I thought but very much so in Diablo 3.

Even running at 120 fps it feels slow or not very reactive. I would not be surprised if that the very long response time of the 16” (91 ms black to white and 42ms grey to grey you sure can feel that).

Not that I play much on these but still.
14” is 40 ms black to white and 58ms grey to grey so sure am curious how different that feels.

Would love to try both in different scenarios next to each other but that is not really an option :frowning:

1 Like

Thanks for doing the tests. Myself I get confused also about what’s the best way to convey speed in percentages… if it is a multiplier or an offset from the base:

  • Car A goes at 200 Km/h, car B at 100 Km/h: A goes at 2x the speed (200% or +100%?) of B, or in times Car A takes 0.5x (50%) the time of B to travel the same distance.
  • Machine A takes 3mins, B 12mins to render: A takes 0.25x (25%) the time of B. Or in speeds, the inverse, A goes at 4x the speed of B (400% or +300%?)

Taxes are always an offset though, price + 15% for example. Discounts too (30% off the price = 0.7x the price).

That in paper is a long time, didn’t know about any of this except “it’s a bit slower” description. 91ms at 60fps is 5.5 frames late or 11 frames late at 120fps. In games ~6frames can feel long but I’m assuming it’s the response of the brightness right? The image is already there it’s just that the leds are ramping up or down to towards the new content?

Note: Are the 14” numbers flipped? Grey to grey is faster than black to white?

Regarding the 14” cores, it seems to be 24-core GPU the sweet spot for the wattage and thermal capacity. 32-core for the money offers very little extra and not that negligible extra battery drains. Also at 32GB, the price difference is small plus the double everything else (screens, bandwidths, encoders/decoders, etc) helps the deal.

No the numbers are like notebook check had them in their review, unless they had them flipped.

As for the other thing I have been thinking, the 16 core might actually be the sweet spot.

I wish there were apple stores in this country… third party official resellers had non in the shelves.

I still don’t like the idea of a Max in the 14” even if it is nice that Apple offers the same config in both form factors.

I think I might stick with the Pro.

1 Like

No worries. Yeh I was similar, but figured best to show render times as base values, then percentage difference. Thought 2x 4x 6x values were too vague.

1 Like

Hmmh,
well, my PC has no GPU Rendering on Blender as I have a standard RX 6800 AMD GPU
(No AMD Pro Card with HID driver)
so I ran the BMW scene on the PC by CPU only (Ryzen 3950X)

PC BMW CPU : 01:37.01
vs
M1 BMW CPU : 5:16.30
M1 BMW GPU : 2:28.26

So the 3950X is still quite faster than my M1 Mini by GPU and Metal Support.
(Or nearly 4 times faster than M1 CPU)

I would be so curios what my RX 6800 16 GB would score - if it would be supported.