Nvidia already dominated in 3D and machine learning market for a long time especially with CUDA. We'll see how it performs once Blender optimized with Apple Silicon with 3.2.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
3D Rendering on Apple Silicon, CPU&GPU
- Thread starter vel0city
- Start date
- Sort by reaction score
Bold statement; possibly true if you think 3D is just gpu rendering, not working with scenes with large amounts of polygons or heavy data or compute workloads
Should probably wait for the review and not the YouTube or gamer reviews before making sweeping statements. We all know what those reviews will be; rendering tiny non production scenes which would fit it the memory of a toaster, because YouTubers need to get the video out asap and can't wait for a benchmark that takes more than 15 min.
Or, to put it in a workflow context, my workstation has an A5000 16GB ($1500) which frequently runs out of memory. On a Mac studio this wouldn't happen. From a productivity view, (not just watching the progress bar as something renders), the Mac Studio would beat the Nvidia card hands down. As a side note it could easily able to sim and render things that the Nvidia card cannot.
In judging whether something is good for 3D, I would argue that responsiveness and startup speed are far more important than waiting for final frames (the last 20% takes 90% of the time).
I'm going to boldly predict that Mac studio will be significantly more responsive and have a better start up time, making it superior for lookdev and artist facing workflows than equivalent Nvidia equipped workstations. I feel this is a safe bet as the M1 Max already does this when doing lookdev in redshift, but let's wait for the Anandtech deep dive and see whose foolish prediction is correct
From the sounds of it 3.2 will be a step forward, but probably just the start of really optimising things. It seems that we're probably a year off decently optimised software, although there are some pretty nice gains already (Houdini was up to 30% faster going from Rosetta to Apple Silicon, and from my testing the Rosetta version was already more responsive than the Intel version - perhaps that speaks to how bad the iMac Pro AMD drivers were more than anything though).
Yes 3.1 was already nice bringing GPU rendering back to Apple machines.Bold statement; possibly true if you think 3D is just gpu rendering, not working with scenes with large amounts of polygons or heavy data or compute workloads
Should probably wait for the review and not the YouTube or gamer reviews before making sweeping statements. We all know what those reviews will be; rendering tiny non production scenes which would fit it the memory of a toaster, because YouTubers need to get the video out asap and can't wait for a benchmark that takes more than 15 min.
Or, to put it in a workflow context, my workstation has an A5000 16GB ($1500) which frequently runs out of memory. On a Mac studio this wouldn't happen. From a productivity view, (not just watching the progress bar as something renders), the Mac Studio would beat the Nvidia card hands down. As a side note it could easily able to sim and render things that the Nvidia card cannot.
In judging whether something is good for 3D, I would argue that responsiveness and startup speed are far more important than waiting for final frames (the last 20% takes 90% of the time).
I'm going to boldly predict that Mac studio will be significantly more responsive and have a better start up time, making it superior for lookdev and artist facing workflows than equivalent Nvidia equipped workstations. I feel this is a safe bet as the M1 Max already does this when doing lookdev in redshift, but let's wait for the Anandtech deep dive and see whose foolish prediction is correct
From the sounds of it 3.2 will be a step forward, but probably just the start of really optimising things. It seems that we're probably a year off decently optimised software, although there are some pretty nice gains already (Houdini was up to 30% faster going from Rosetta to Apple Silicon, and from my testing the Rosetta version was already more responsive than the Intel version - perhaps that speaks to how bad the iMac Pro AMD drivers were more than anything though).
3.2 with the Eevee rewrite and them changing the viewport to Metal should be great.
Yeah, I was primarily referring to the GPU rendering component. You’re right, there may be some other areas it could push ahead, but was hoping the GPU side was going to be looking stronger.
Will still keep my ultra order as it’s not like there’s many other options if you want to be running macos anyway.
I can get pretty reasonable 3D work done on my M1 max MBP even under Rosetta 2 (and poor performance of the app by itself never mind intel or m1 )
If anything your ultra would perform even better.
For Blender users it sounds like continued development is being made to lower GPU renderings times on Apple Silicon. The Apple developer on the Blender dev forums claimed they were able to get GPU renderings times cut down to half of what they are now in earlier test. The main hold up seems to be keeping renderings consistent and also making optimizations for Apple silicon without hurting the performance of other GPUs. There was also talk about using the neural engine for de-noising.
So basically the performance we are seeing now with the M1 Max getting absolutely stomped in GPU rendering times compared to the 3070m (which it can actually beat in other GPU task) should change as the year goes on.
So basically the performance we are seeing now with the M1 Max getting absolutely stomped in GPU rendering times compared to the 3070m (which it can actually beat in other GPU task) should change as the year goes on.
It's been in development with Blender and Apple developers for at least 5 months so how much longer does it need to be optimized? Maybe it's the hardware that needs to be optimized with the M1 Ultra getting closer but still 2/3 of mobile Nvidia.
Last edited:
“At least 5 months”. Lol. This says so much.
5 months isn't really enough and even Apple developer admitted that they need to optimize it continuously. Beside, many 3D software were optimized for several years you know.
27 years of Blender development, how dare Apple not have Metal crushing Optix render times, they've had five whole months of development time...!!!
I have some holiday time coming up. Thinking about making a version of photoshop. A week or two should suffice surely?!
It's going to be interesting in the next year or so as 3D Apps finally start moving over to viewports based on Metal and Vulkan; I think Blender will be one of the first. Hopefully this will result in much better performance as a lot of legacy OpenGL stuff being put out to pasture. Also hopefully mean that graphics on MacOS can move past the OpenGL 4.1 feature set (finally).
Nice! I invariably have to use Linux at work; really miss MacOS. Out of curiosity, what are you going to be using the ultra for? Be good to get some idea of it's real world performance outside of the usual benchmark suites.
TBH the main thing holding me back from getting the Ultra is the GPU; it's a bit of an unknown quantity at the moment.
Lol
Speaking of hardware optimisation, with 12.3 coming out, I'd be curious to see how the AMD GPUs compare with the AS ones since they'll both be running on metal now...
5 months already!? God, I don't know how Apple hasn't released their own 3D modelling app by now.
Bold statement; possibly true if you think 3D is just gpu rendering, not working with scenes with large amounts of polygons or heavy data or compute workloads
Should probably wait for the review and not the YouTube or gamer reviews before making sweeping statements. We all know what those reviews will be; rendering tiny non production scenes which would fit it the memory of a toaster, because YouTubers need to get the video out asap and can't wait for a benchmark that takes more than 15 min.
Or, to put it in a workflow context, my workstation has an A5000 16GB ($1500) which frequently runs out of memory. On a Mac studio this wouldn't happen. From a productivity view, (not just watching the progress bar as something renders), the Mac Studio would beat the Nvidia card hands down. As a side note it could easily able to sim and render things that the Nvidia card cannot.
In judging whether something is good for 3D, I would argue that responsiveness and startup speed are far more important than waiting for final frames (the last 20% takes 90% of the time).
I'm going to boldly predict that Mac studio will be significantly more responsive and have a better start up time, making it superior for lookdev and artist facing workflows than equivalent Nvidia equipped workstations. I feel this is a safe bet as the M1 Max already does this when doing lookdev in redshift, but let's wait for the Anandtech deep dive and see whose foolish prediction is correct
From the sounds of it 3.2 will be a step forward, but probably just the start of really optimising things. It seems that we're probably a year off decently optimised software, although there are some pretty nice gains already (Houdini was up to 30% faster going from Rosetta to Apple Silicon, and from my testing the Rosetta version was already more responsive than the Intel version - perhaps that speaks to how bad the iMac Pro AMD drivers were more than anything though).
How did you manage to buy A5000 with only 16GB of RAM?
Edit: Nevermind you are talking about laptops. Doing look development on a laptop? That's interesting.
I'll use it for Software Dev & Visual Design.
So a mix of Xcode, Figma, Blender primarily. Then the odd bits of FCP, Logic, Motion, Photoshop, After Effects.
True that.
No you were right the first time. I misremembered; just checked and it's a Quadro RTX 5000 16GB. I blame Nvidia's stupid naming scheme
In all fairness, the earliest they could have started collaborating with Blender was arguably 18 months ago, with the DTKs. They did start the Apple Silicon transition by contributing to a lot of widely used tools and libraries, it does make some sense to wait until the professional SoCs (M1 Pro & Max) were released to start working on that too.
Interesting use case; that's a pretty mixed range of apps!
With Houdini I get the impression that a fair bit of time was waiting for the underlying tools and libraries to be ported over; once the libraries were in place getting it running on Apple Silicon was relatively quick. I imagine Blender was in the same place as well; 3D apps tend to have a lot of dependancies.
Why not before? In addition to helping it sell hardware, Apple benefits from Blender in other ways. For example, Apple TV+ offers Wolfwalkers, a movie made with Blender.
“2D Isn’t Dead, It Just Became Something Different”: Using Blender For Wolfwalkers — blender.org
How Cartoon Saloon incorporated Blender into their workflow for the Oscar and Golden Globe-nominated Wolfwalkers.
www.blender.org
It’s also made with Blender among many other softwares and usually Blender is not the tool of choice for this stuff.
It is surprising how many here are tirelessly waiting for Blender updates while there are many other tools that works already fast and natively on both CPU and GPU, don’t you have work to do?
It is surprising how many here are tirelessly waiting for Blender updates while there are many other tools that works already fast and natively on both CPU and GPU, don’t you have work to do?
Yep not optimized yet. In other GPU based task and benchmarks, the M1 Max reaches 3070m and in some rare cases even 3080m levels of performance.
GPU rendering has been almost exclusively developed for Nvidia GPUs for the past 5+ years so it’s not surprising that Apple Silicon (and AMD for that matter) are falling behind and need extra development to show their true performance.
If Apple was already getting twice the level of rendering performance before the official Blender 3.1 release then we should be seeing something closer to 3070m performance once the optimizations are built in and especially if the neural network is used for de-noising.
Since version 2.8 onward Blender has quickly moved from a subpar 3D app to one of the top choices for professional 3D users. I personally have spent years paying for the big name 3D modeling applications and last year, I switched to Blender and haven’t looked back. As someone who has been in 3D for 15 years now, I can honestly say Blender has become my preferred software (something I never thought I would say).
Also Blender and Cinema 4D are so far the only 3D software packages that run natively on Apple Silicon. However, for rendering Redshift and Octane are great on Apple Silicon as well (though more optimization needs to be done in those as well).
Update on Blender render with metal (12.3 and Blender 3.1)
M1 Pro GPU+CPU Metal 1:24 (silent of course)
iMac 8-clore i7 (10700K) and 5700 GPU
CPU only: Loud fan 2:50
GPU only: Silent(!) 0:54
GPU+CPU: Load fan 0:50
Looks like the 5700 is rather competitive compared to the M1 pro. I do not think The 10700K and 5700 help each other much. Great 3.1 update. Render time went down from 2:50 to 0:50 and the fan is now silent.
Apple what took you so long???
M1 Pro GPU+CPU Metal 1:24 (silent of course)
iMac 8-clore i7 (10700K) and 5700 GPU
CPU only: Loud fan 2:50
GPU only: Silent(!) 0:54
GPU+CPU: Load fan 0:50
Looks like the 5700 is rather competitive compared to the M1 pro. I do not think The 10700K and 5700 help each other much. Great 3.1 update. Render time went down from 2:50 to 0:50 and the fan is now silent.
Apple what took you so long???