3D Rendering on Apple Silicon, CPU&GPU

[mi7chy]

So this guy uploaded a video comparing the m1 max in octane to a 1080ti desktop gpu, 2080ti desktop gpu, and 3080ti desktop gpu. His results show the m1 max performing in line with the 1080ti.
So in his video and in reply to many of his comments he keeps saying apple lied in their marketing and its nowhere near as powerful as they said. But the 1080ti is equivalent in performance to a 3080 laptop gpu...so his results back up apples claims directly, unfortunately he doesnt even understand this and blamed me in the comments for "spreading false information"..
smh

What tests have you done or what data are you going by that shows a desktop 1080ti equal to mobile 155W 3080 (MSI GE76 Raider 11UH-053 used in Apple's presentation)? Just taking a quick look at what's available on YouTube a mobile 155W 3080 is about 70% of a desktop 320W 3080 while a desktop 1080ti is about 50% of a desktop 3080. He seems more experienced and credible.

Just Google searched his name. This guy is no joke compared to a lot of the YouTube videos where they're fumbling around not knowing what they're doing in Blender, etc. and think they're testing dGPU but actually testing CPU.

https://leonbaisden.com

 

[ader42]

What tests have you done or what data are you going by that shows a desktop 1080ti equal to mobile 155W 3080 (MSI GE76 Raider 11UH-053 used in Apple's presentation)? Just taking a quick look at what's available on YouTube a mobile 155W 3080 is about 70% of a desktop 320W 3080 while a desktop 1080ti is about 50% of a desktop 3080. He seems more experienced and credible.

You need to compare to a lower wattage PC laptop e.g. 100 watt, like the Tally Ho comparisons:

 

[crazy dave]

You need to compare to a lower wattage PC laptop e.g. 100 watt, like the Tally Ho comparisons:

Apple compared the Max to the 155W 3080 mobile GPU, which in specific circumstances it can reach, but that makes it fair game to compare the two in other circumstances too. More germane is that a lot of renderers are still a work in progress for M1. So the guy in the video posted by @hefeglass saying Apple lied is being a bit overly dramatic.

 

[mi7chy]

You need to compare to a lower wattage PC laptop e.g. 100 watt, like the Tally Ho comparisons:

Apple is clearly highlighting the 155W 3080 MSI GE76 Raider 11UH-053. Look at the fine print.

 

[Boil]

But if you step a bit back and look at things, I’m actually impressed that the Apple GPU performs like a desktop 1080ti in regards to compute performance in Octane. Raw compute performance isn’t the top strength of the Apple GPU architecture in my book and I have no idea why some people think it would be.

Yes...! I believe Apple silicon is one of those edge cases where "the whole is greater than the sum of its parts"... And I would think this should already be apparent to anyone using the new M1 Pro / Max SoCs for video work, and will become more apparent to those doing 3D / DCC as the various software packages are optimized for Apple silicon...?!?

 

[vel0city]

My M1 Pro/Max/64Gb arrived a few days early. Not had a chance to run any official benchmarks (tbh I find these a bit pointless anyway as they don't reflect how I work) but I have opened and worked with some of my existing C4D/Redshift projects.

In terms of the day-to-day usage, lookdev and creating images experience, this machine is absolutely sublime compared to my Mac Pro/64 GB/1080ti.

Projects with softbody dynamics, millions of particles, displacement and high density meshes are completely fluid and responsive in the viewport. There is just no choking or bottlenecking even with millions of mograph clones or instances. The kind of scenes that would have bought my Mac Pro to its knees and locked into an endless beachball are like butter on the Max. It's a pleasure to scrub the timeline and see realtime updates and not experiencing any lag or stickiness.

I was really blown away to run a softbody simulation animation and to see it running in realtime in the viewport and rendering simultaneously in the Redshift IPR without dropping frames. Previously I would be stop/starting the animation and IPR, but now they can run together giving real-time updates.

The Redshift IPR starts a LOT faster than my Mac Pro with 1080ti. When Redshift is processing textures and calculating tessellation and subdivision/displacement I got used to a long wait on my Mac Pro, on the Max the time to calculate the image in progressive mode is hugely reduced. Not instant, but so fast that it makes experimentation far less time consuming and laborious.

This is all with the Max completely silent and the fans not picking up at all. I activated High Power mode to see if I could rev the fans up but it just stays silent and running cool. Any of these project files would have turned my Mac Pro into a leafblower.

I'd also advise caution with a lot of the people running benchmarks on Youtube. I've seen Youtubers using C4D files from TFMStyle to use as demos but his scenes are quite unoptimised and not professionally configured. In general the majority (not all) of Redshift/C4D/M1 Pro/Max demos I've watched are not by people who use the software daily and are just looking at numbers.

 

[chouki]

My M1 Pro/Max/64Gb arrived a few days early. Not had a chance to run any official benchmarks (tbh I find these a bit pointless anyway as they don't reflect how I work) but I have opened and worked with some of my existing C4D/Redshift projects.

In terms of the day-to-day usage, lookdev and creating images experience, this machine is absolutely sublime compared to my Mac Pro/64 GB/1080ti.

Projects with softbody dynamics, millions of particles, displacement and high density meshes are completely fluid and responsive in the viewport. There is just no choking or bottlenecking even with millions of mograph clones or instances. The kind of scenes that would have bought my Mac Pro to its knees and locked into an endless beachball are like butter on the Max. It's a pleasure to scrub the timeline and see realtime updates and not experiencing any lag or stickiness.

I was really blown away to run a softbody simulation animation and to see it running in realtime in the viewport and rendering simultaneously in the Redshift IPR without dropping frames. Previously I would be stop/starting the animation and IPR, but now they can run together giving real-time updates.

The Redshift IPR starts a LOT faster than my Mac Pro with 1080ti. When Redshift is processing textures and calculating tessellation and subdivision/displacement I got used to a long wait on my Mac Pro, on the Max the time to calculate the image in progressive mode is hugely reduced. Not instant, but so fast that it makes experimentation far less time consuming and laborious.

This is all with the Max completely silent and the fans not picking up at all. I activated High Power mode to see if I could rev the fans up but it just stays silent and running cool. Any of these project files would have turned my Mac Pro into a leafblower.

I'd also advise caution with a lot of the people running benchmarks on Youtube. I've seen Youtubers using C4D files from TFMStyle to use as demos but his scenes are quite unoptimised and not professionally configured. In general the majority (not all) of Redshift/C4D/M1 Pro/Max demos I've watched are not by people who use the software daily and are just looking at numbers.

Thanks for this little "report"
Wow that's pretty exciting, and convincing if I was still a bit hesitating to get one.

How is the experience with After Effects, if you happen to use it?

 

[vel0city]

Thanks for this little "report"
Wow that's pretty exciting, and convincing if I was still a bit hesitating to get one.

How is the experience with After Effects, if you happen to use it?

I'm not an AE user so don't really have anything to compare the Max to, but I'll install the new beta if you'd like me to try anything, it's native now.

I will say that Photoshop is absolutely flying on this Max though. Compared to my M1 Mini 16GB which was already impressive, this is another experience entirely. Working with 4k .PSB images with smart objects, heavy layers, the Max just eats through everything.

 

[l0stl0rd]

5-6 months? What the ?? Will ANY render programs support Metal and fully take advantage of the GPU before next spring?

Well actually I was about to ask….

Can someone with an M1 Pro or Max get the blender 2.93.5 the intel / x86 version and install Radeon Pro Render?

You can render on the GPU using Metal on there.

Yes it uses Rosetta2 but still would be interesting to see.

 

[chouki]

I'm not an AE user so don't really have anything to compare the Max to, but I'll install the new beta if you'd like me to try anything, it's native now.

I will say that Photoshop is absolutely flying on this Max though. Compared to my M1 Mini 16GB which was already impressive, this is another experience entirely. Working with 4k .PSB images with smart objects, heavy layers, the Max just eats through everything.

Yes definitely interested in AE performance with M1 Max, mainly regarding the interface, wich was awfully sluggish on the Mac Pro 7,1 (render was fine)
If ever you have an old AE project with many layers to open on the Max I'd be very interested

 

[vel0city]

Yes definitely interested in AE performance with M1 Max, mainly regarding the interface, wich was awfully sluggish on the Mac Pro 7,1 (render was fine)
If ever you have an old AE project with many layers to open on the Max I'd be very interested

Might be what you're looking for, AE tests start at 10.00

 

[jujoje]

Well actually I was about to ask….

Can someone with an M1 Pro or Max get the blender 2.93.5 the intel / x86 version and install Radeon Pro Render?

You can render on the GPU using Metal on there.

Yes it uses Rosetta2 but still would be interesting to see.

I don't think RPR is a particularly good benchmark; only tried it in Houdini but was finding it a bit slow and lacking features compared to Karma (18.5) possibly the blender plugin is in a better place? While it ran on mental it didn't feel like it was making much use of it.


On the redshift forums some interesting numbers (for the Disney Moana dataset):

Posting here for reference some test results. All the below are without tessellation

My GPU tests
2x 2080ti = 34m:17s
Single 3090 = 21m:45s
2x 3090 = 12m:44s
Apple M1 Max 64gb = 28m:27s

Not bad for a laptop I'm guessing the unified memory really helps given the size of the scene. The general impression I get is that, while you won't get full fat 3080 rendering speed, the time to first pixel and interactivity is pretty outstanding.

I'd also advise caution with a lot of the people running benchmarks on Youtube. I've seen Youtubers using C4D files from TFMStyle to use as demos but his scenes are quite unoptimised and not professionally configured. In general the majority (not all) of Redshift/C4D/M1 Pro/Max demos I've watched are not by people who use the software daily and are just looking at numbers.

Thanks for the info and nice to be confirmed in my prejudice there It certainly seems that a lot of the 'benchmarks' for C4D and Blender are done by people who don't use the programs professionally and their comments are based of sketchy knowledge of what the numbers relate to (kudos for them trying some 3D benchmarks - most reviewiers are of expertise is 2D or video and it's great to have some idea). Wish there was a more accurate benchmark (I had been hoping to roll my own, but unfortunately a bit on the busy side atm).

 

[jujoje]

Just Google searched his name. This guy is no joke compared to a lot of the YouTube videos where they're fumbling around not knowing what they're doing in Blender, etc. and think they're testing dGPU but actually testing CPU.

Without having downloaded the files (not having C4D), non of the test scenes he showed in his benchmark / review videos looked particularly heavy. I wouldn't expect to see much of an advantage from the unified memory or the benefits of having 64Gb available. Which I guess is what you see in the Moana dataset, being a decent sized production scene. As always depends on what you're trying to render.

Also, as a side note, I would imagine that the next version of octane will the optimised for the newer hardware. Octane X was released a while back and doesn't seem to have had significant updates since then particularly compared to the main branch (Octane development, naming and plugins are all over the shop, so might be wrong).

 

[chouki]

Might be what you're looking for, AE tests start at 10.00

thx

 

[Macintosh IIcx]

My M1 Pro/Max/64Gb arrived a few days early. Not had a chance to run any official benchmarks (tbh I find these a bit pointless anyway as they don't reflect how I work) but I have opened and worked with some of my existing C4D/Redshift projects.

In terms of the day-to-day usage, lookdev and creating images experience, this machine is absolutely sublime compared to my Mac Pro/64 GB/1080ti.

Projects with softbody dynamics, millions of particles, displacement and high density meshes are completely fluid and responsive in the viewport. There is just no choking or bottlenecking even with millions of mograph clones or instances. The kind of scenes that would have bought my Mac Pro to its knees and locked into an endless beachball are like butter on the Max. It's a pleasure to scrub the timeline and see realtime updates and not experiencing any lag or stickiness.

I was really blown away to run a softbody simulation animation and to see it running in realtime in the viewport and rendering simultaneously in the Redshift IPR without dropping frames. Previously I would be stop/starting the animation and IPR, but now they can run together giving real-time updates.

The Redshift IPR starts a LOT faster than my Mac Pro with 1080ti. When Redshift is processing textures and calculating tessellation and subdivision/displacement I got used to a long wait on my Mac Pro, on the Max the time to calculate the image in progressive mode is hugely reduced. Not instant, but so fast that it makes experimentation far less time consuming and laborious.

This is all with the Max completely silent and the fans not picking up at all. I activated High Power mode to see if I could rev the fans up but it just stays silent and running cool. Any of these project files would have turned my Mac Pro into a leafblower.

I'd also advise caution with a lot of the people running benchmarks on Youtube. I've seen Youtubers using C4D files from TFMStyle to use as demos but his scenes are quite unoptimised and not professionally configured. In general the majority (not all) of Redshift/C4D/M1 Pro/Max demos I've watched are not by people who use the software daily and are just looking at numbers.

Thanks for the report, very encouraging.

 

[vel0city]

Redshift warning message

 

[vel0city]

Just dropping this in here in case anyone is interested - 3D painting directly onto meshes now available in Procreate:

Looks pretty great, easy export to C4D/redshift.

 

[mi7chy]

Redshift warning message

Sounds similar to Blender tile size performance difference.

https://docs.blender.org/manual/en/latest/render/cycles/render_settings/performance.html

"Depending on what device you are using for rendering, different tile sizes can give faster renders. For CPU rendering smaller tiles sizes (like 32 × 32) tend to be faster, while for GPU rendering larger tile sizes give a better performance (like 256 × 256)."

dGPU finishes Blender Classroom demo render in 2:04:20 (min:sec:subsec) with default 32 x 32 tile size but cuts almost in half with 1:10:28 when using 256 x 256 tile size.

 

[jujoje]

Just dropping this in here in case anyone is interested - 3D painting directly onto meshes now available in Procreate:

This is pretty amazing. The brush engine in procreate is great; gave it a go and worked really well. Still need's some additional bits and bobs (bump/displacement, texture projection and would love to see smart materials) but feels great and is pretty elegant ui wise.

Actually been impressed with 3D apps on the iPad of late; Nomad is pretty impressive for sculpting and now Procreate is doing a good job of the painting side of things. Just wish there was a low poly modeller in there as well.

 

[thenewperson]

Actually been impressed with 3D apps on the iPad of late; Nomad is pretty impressive for sculpting and now Procreate is doing a good job of the painting side of things. Just wish there was a low poly modeller in there as well.

They've been slowly improving with the iPads as people keep arguing over "content consumption" vs "content creation".

 

[sirio76]

Procreate is amazing, I use it regularly for painting high res texture and the performance is outstanding compared to other desktop tools. The time laps feature is very nice too, here I’ve recorded a short video:

Can’t wait to test it for 3D painting.

 

[JMacHack]

On the topic of compute benchmarks, I’ve always found the relationship between compute and gpu performance weird. Example: AMDs GCN architecture was baaaad for graphics. Often barely competing with NVidia’s 1000 series. Yet they blew it out of the water in compute. Iirc Radeon 7 wasn’t topped for compute until the 30 series (?). Which makes the GCN cards popular with miners.

I’d wager that ASi leverages heavily on graphics efficiency similarly to Pascal.

 

[vel0city]

This is pretty amazing. The brush engine in procreate is great; gave it a go and worked really well. Still need's some additional bits and bobs (bump/displacement, texture projection and would love to see smart materials) but feels great and is pretty elegant ui wise.

Actually been impressed with 3D apps on the iPad of late; Nomad is pretty impressive for sculpting and now Procreate is doing a good job of the painting side of things. Just wish there was a low poly modeller in there as well.

Something like ZModeler from Zbrush would be a good fit for the iPad, if they could refine the interface and focus all the important functionality around the Apple Pencil it could be an app in itself. Paired with the 3D painting from Procreate and sculpting in Nomad, that is a lot of 3D bases covered.

I did have a fever dream that Apple Pencil would be compatible with the M1 Max & Pro trackpad. That would be a wonderful addition.

Great video @sirio76 thanks for sharing.

 

[l0stl0rd]

Curious to see what will hapen to Forger now that Maxon bought them.

 

[vel0city]

Curious to see what will hapen to Forger now that Maxon bought them.

Prolly $100/month subscription and no new features knowing Maxon of late.