Corona for Apple Silicon M1?

[BigAl3D]

Were we supposed to use Denoise for these tests?

[jojorender]

I would say “no” on the denoiser.
BigAl3D, that iMac has a lot of life left in her. Pretty ok result for a 5y old.

I’m not sure what my expectation for the ultra was but that score is in line with Philw’s Max best 4:25 min score. Ultra doubling Max power.
I think the important number here is the rays/s to make things comparable.
I was sort of hoping the Ultra to be in the 12M rays/s range, only because all my I9 10 cores hover around 6M r/s. Wanted to see at least double the power.
This Ultra comes in at 9.7M r/s, that’s according to the benchmark list the middle ground for Ryzen 9 5950X 16 core but also some older 32 cores rippers
If the coming Mac Pro Extreme doubles the Ultra to 20M r/s that would still barley touch a 64 core ripper. They mostly cruise around 25-30M r/s

@ Philw’s any updates on the render power weirdness between PV and VFB?
I don’t know the first thing about coding, but maybe corona M1 code optimization will be able to squeeze out more power in the future?

@piredude Did you run just 1 test or multiple with the same result?
Any difference when rendering only to PV or to PV+VFB? Just for comparison, can you also run the corona benchmark? I’m curious to see the rays/s there 

[BigAl3D]

For giggles, I ran a 4k test with Nose Level set to 3 as suggested above. I think this also has IntelDenoise ON, but that goes pretty fast on this machine.

This test might be a bit much for a benchmark. An hour is perfectly fine for stills, but a no-go for an animation unless you're Pixar or Disney.

[piredude]

So the Ultra is faster than the fully loaded iMac Pro at the default resolution. Here are my results at 4K width. (png is too large, so attaching a 100% jpg.)

[BigAl3D]

I was hoping that Ultra time would be faster to be honest. I know this iMac Pro was expensive when we got it, but that loaded Studio isn't cheaper either. The question remains if indeed there are any additional software updates needed to take full advantage of the new dual chip systems. Supposedly, the way Apple set these up, the system is going to see it as one big CPU instead of a dual system.

I watched the Max Tech first review on these Macs, and he addressed this by noticing Apple's presentation speed claims were using an un-released update to Final Cut Pro. That would indicate an update might be necessary. Who knows.

Not sure if that would be an update to the OS so all M1-optimized apps take full advantage, or each piece of software. Apple are so focused using less power over just making it as fast as they can and also try to save some power, that it's hurting the product. Max Tech also noticed the GPUs are being throttled in certain tasks. Maybe that's a future "update" and they can release that power that is already there.

So many questions.

[Philw]

I'll put the Threadripper on it as soon as it has stopped heating the house with a 1000 frame Redshift job :-)

[prince_jr]

hi corona-users

i think y'all should use the corona benchmark for comparison. there's only one result yet:
https://corona-renderer.com/benchmark/results/cpu/ultra

would be nice to see more users testing their mac studio like this.

for comparison:
my mac pro (5.1, 2010) 2x6 core, 3.46 ghz, 128gb ram needs approx. 2 minutes for the benchmark scene.
> i seems the new mac studio 20-core is "only" almost twice as fast...1m15s

what do y'all say about that?

[Philw]

@prince_jr We are avoiding it because its not M1 native yet.

[prince_jr]

@Philw ah okay. didn't know that. this brings me to additional question.
so...one day if corona renderer is native m1, will the renderer benefit more from the processor and be faster than now?

[Philw]

Corona v8 IS native M1 - just the benchmark isn't - so we were just trying to standardise on a scene for now. So performance now is pretty much just number of available CPU cores as per Intel - but without the overhead of the Rosetta conversion you would get in V7.

[BigAl3D]

I'm assuming I don't need to do anything for this benchmark since it starts on its own. To reinforce the statement that it is not M1 native, my 2017 iMac Pro beats out the M1 score.

Corona 1.3 Benchmark Finished
BTR Scene 16 passes
18-Core Intel(R) Xeon(R) W-2191B CPU @ 2.30GHz
Render Time: 0:00:59, Rays/sec: 8,164,390

[jojorender]

Yo Phil, is the ripper still keeping your place nice and cozy?…asking for a friend.
Keep that beast running, we don’t want you to freeze.
I just checked the benchmark list and it turns out the currently 3 fastest “Macs” (16-18M r/s) are all Threadripper 3970X 32-Core - LOL, sorry Apple.
The living room scene we picked for M1 testing gives me (Intel Mac) a similar rays/s number as the official benchmark scene.  Same for BigAl3D’s iMac Pro 8.1M r/s in both.
That makes it easy to see where the “M1 optimized” rays/s score plugs into that benchmark list.
Ultra lives in the 10M r/s region, not 6.5M r/s as the official benchmark suggests
I believe macOs is already fine-tuned to support all the power Ultra has to offer. It comes down to the individual app to add “M1 optimized” support for the task at hand.
The scores where Ultras shine, all use hardware acceleration like the Media Engine.
Render engines probably only tap into the actual cpu cores and nothing else.
How much initial M1 software support can be further optimized is anybody’s guess.
I believe the Corona M1 support is a one-and-done thing. No use in hoping for a further optimization miracle.

At this point, I don’t think the ultra makes a great “render box”, but no doubt an amazing machine to work in a scene. The insane direct connected memory speed is probably the real game changer here . To take advantage of that you have to max out internal SSD and have all your textures and other assets live in that tiny box… not sure if I’m comfortable with that.
External storage interface would still be a huge bottleneck. Going 10GbE network to feed that memory is another big expense.
I think Phil hit the sweet spot, Max to work in and ripper to crunch those rays.

[Philw]

4K - Threadripper 32/64 3% Noise limit

[prince_jr]

here's what my machine rendered...see attachment (mac pro 5.1 2010, 2 x 6-core 3.46ghz, 128ram, c4d r25)

impressive render time from philw's threadripper...amazing! but i must say, piredude's mac studio ultra is fast, too. could be a venerable successor for my 12-year-old "cheese grater".

[frv]

Scene parsing on the Threadripper is 3s and the Mac ultra 2 sec looking at the post here. May mean nothing for this test but my scene parsings on much bigger files take minutes on a late 2015 iMac and are the biggest bottleneck for me to get work done. If I am not mistaken render to first pixel is basically scene parsing.

Since most designers need fast feedback the Mac Ultra might just be 50% faster than a Threadripper. That would be huge. To set up a Mac Ultra with 128Gb ram/4TB SSD and have all my textures on the internal SSD is totally fine with me. I see a lot of testing done looking at final render times at SL +100. But for me as a designer only scene testing and finetuning materials is important. It would be a lot more interesting to me to see how interactive rendering is performing and the time to first pixel. I have not found any reviews or tests on that. Once a scene is fully tested and ready for a final image,  the work is done,  and render times less than a few hours a non-issue these days.