I’m having power pc flashbacks with comments like this coming forth. ?
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
Because Apple does NOT have such thing. Even FCPX was acquired and created from original Premier Pro developers. Also, if Max's Tech is right, then optimizing would be very difficult especially for 3D software which is already too friendly to Nvidia.
Optimizing wouldn't be difficult for Apple's own engineers.
Plus the difference in architecture is the main reason why Apple wouldn't want to purchase (or fork) existing solutions, and would instead want to start from scratch.
Xcode itself contains an entire 3D editor that almost nobody knows exists, so developer time is not really a problem for Apple.
I mean we're talking about a company that has just literally made its own GPU. In comparison making a simple renderer is child's play.
Plus the difference in architecture is the main reason why Apple wouldn't want to purchase (or fork) existing solutions, and would instead want to start from scratch.
Xcode itself contains an entire 3D editor that almost nobody knows exists, so developer time is not really a problem for Apple.
I mean we're talking about a company that has just literally made its own GPU. In comparison making a simple renderer is child's play.
Thing is, a SoC has many TLBs, at the very least each core has its own TLB. In addition there are TLBs in the fabric, wherever some address translations via page tables happen.
You are right, TLBs are not anywhere close to 32MByte in size. Assuming 3072 entries and 8Byte per entry, we are talking about 24kByte per TLB - which are already huge TLBs
Thing is, a SoC has many TLBs, at the very least each core has its own TLB. In addition there are TLBs in the fabric, wherever some address translations via page tables happen.
You are right, TLBs are not anywhere close to 32MByte in size. Assuming 3072 entries and 8Byte per entry, we are talking about 24kByte per TLB - which are already huge TLBs
Thanks, it is indeed a gap in my education, the last time I directly worked with TLBs was on a 80386 and some time passed since that ? Modern cache hierarchies are something else entirely and I urgently need to brush up my knowledge on how these things work. Any good, readable sources on caches, tagging and all that fun stuff you could suggest?
Why would it take Apple longer to optimize a renderer like Cycles than to create one from scratch?
Would a renderer that could not be used in cloud render farms be popular?
Because Apple can't just completely rewrite the internals of Cycles.
If you look at the code for Apple's cycles patch, it's almost more of a translation than an "Apple renderer". They don't have the freedom to start tweaking things massively because it's not their codebase.
They are also bound to use Blender's cross platform code specifically so that a cycles render done on a Mac looks identical to a cycles render done on an nVidia or AMD card. (It would suck if some people did a collaborative animation and all the frames rendered by the "Mac guy" looked different)
An Apple render engine. They could call it 'MetalRay'. 😅
...I'll get my coat.
...I'll get my coat.
I fully agree that Apple should develop and provide a render engine, especially given how everybody seems to be latching onto this area as the only negative about the new hardware. The cost to Apple should be completely insignificant given their resources.
For my own part it looks like i’ll be sticking to the MBP 16” M1 Max until I see what 2024 brings — final render times are not my biggest concern to be honest, I spend far more time not-rendering.
For my own part it looks like i’ll be sticking to the MBP 16” M1 Max until I see what 2024 brings — final render times are not my biggest concern to be honest, I spend far more time not-rendering.
That was my hope for some kind of reference renderer like Pro Render at WWDC. I feel there is reasonably good place where MacOS in general would benefit from having a renderer which is … Preview.
Preview currently uses a metal version of Pixar's Storm for Quicklook and opening USD files which essentially amounts to a fairly low feature OpenGL display. Being able to Preview 3D files in a basic path traced renderer without opening an app would be really good. Also being a Hydra delegate it could be used up in other apps.
Either way it would be good to have a reference render to dispel all these debate over the hardware / software capabilities.
A colleague of mine knocked up a basic ray trace renderer in XSI using the ICE nodes in an afternoon. How hard can it be Apple ?
This is a current issue with Houdini; the Apple Silicon version returns different random values, so whenever you use a randomly generated value to drive, say noise, you get a different result between the Intel and AS versions.
Funnily enough remember having this issue back in the day with Athlon 64 and Intel; fractal patterns would render differently which was a fun thing to discover when just before delivery half the frames were coming back from the farm different. Good times.
Gets my vote!
Chromed-out Corvette Stingray for a turntable demo...! ;^p
Maybe Apple has done it, but it doesn't work the way they expect it to.
What features should Apple's reference renderer have?
Just wait for the first round of optimizations 40% faster
developer.blender.org
Blender Archive - developer.blender.org
My M1 Max 16” MacBook Pro runs circles around everything intel in my house… and I have a lot of 12 and 24 cores intel hardware with GPU acceleration… in a really low TDP design… just the M1 Max 32 Cores GPU is a powerhouse compared to everything I have at home (including a SLI of nvidia GTX 1070 with an i7 5820k)
And my cinebench score on my m1 max was over 12500 last time I ran it.
Just ran Cinebench…
And my cinebench score on my m1 max was over 12500 last time I ran it.
Just ran Cinebench…
Attachments
Last edited:
Just wait for the first round of optimizations 40% faster
Blender Archive - developer.blender.org
Well spotted!
That's more like it.
I'm really loving this scenario where we have actual Apple developers working directly on Blender code.
Probably better to keep this together under the same roof, no? Unless a mod disagrees.
Anyway, my non-developer/3D artist take on it is this: Wether the TLB claim proves to be accurate and a contributing factor or not remains to be seen. We haven't gotten to the bottom of this yet. I don't like this automatic talk of click-bait just because it comes off of YouTube.
It's good to see people are digging into the disappointing GPU numbers of the Ultra, just as it is good to see others refuting it so long as it is done so on facts.
It's a fascinating story that is developing, and certainly not one I thought we'd be having just a short 2-3 years ago when the Mac (3D/CG) landscape looked dismal. Look how far Apple has come and they haven't even gotten to the Mac Pro yet.
It probably wouldn't be a terrible idea to support the same range of features as usdz as that's the format they're targeting for AR and quicklook already supports the basic geometry types (mesh, curves, points, cameras).
Other than that all you need it a principled / Disney shader and lights and you'd have something that would work with the output of most of the main DCC apps. Add a fast path tracer and you should be good to go.
For future proofing MaterialX support would be good as that is becoming the standard and there are other things that would be nice (volumes being the main one, but also more complex instancing etc), and also more complex edge case scenarios but these days I think polygons + lights + principledshader would cover what 90% of people need from a renderer.
Would be fascinating to see what they could come up with, if they didn't have to deal with legacy code and heavy optimisation for CUDA or Optix.
TLDR: More speed! more power!
It's a fascinating story that is developing, and certainly not one I thought we'd be having just a short 2-3 years ago when the Mac (3D/CG) landscape looked dismal. Look how far Apple has come and they haven't even gotten to the Mac Pro yet.
I think a lot of this can be traced back to the creation of a dedicated Pro Apps team, I think around 2019 (from what Gruber said). From what I've heard Apple hired a fair few people from the industry (both developers and artists) to focus on workflows and hardware, as well as provide support for developers in optimising for Apple hardware.
What we're seeing now is really the initial fruits from that. I get the impression that before then there was no real dedicated team for 3D so any development in that area was really sporadic and uncoordinated. So far feel it's been pretty successful in getting Apple taken more seriously for 3D, although it still has a fair way to go. Things are definitely looking promising on the 3D front.
I wonder if the Pro Apps team is the Special Projects Group that Brad Peebler (formally associated with Newtek/Lightwave and Luxology/Modo) heads up for Apple...?
I would love to see a comprehensive 3D/DCC software suite from Apple...!
Hmm not sure.
If I'd have to guess the Special Projects group would be for the AR and in-house 3d tools (hence Brad) and Pro Apps more for the DCC and industry relations. Total speculation though. Or they could totally be the same thing.
Would totally be on board with Apple Sherlocking Modo though
Still after good modelling software...
Can confirm. Several ex-colleagues and friends from places like Framestore and Weta D. ended up getting hired there.
Now that I remember, someone from Apple Special Projects did get in touch with me as well, but that was back in 2015. Though that was likely about something different.
I also know someone who used to be on the Maya beta-test program with me who I think is now in the Pro Apps team. He got hired by Apple back in 2014.
Holy crap! Brad's at Apple now? Awesome! And what a steal for Apple. I was under the impression he had left the industry after leaving Foundry. Always loved what he did at NewTek and then later with Luxology.
I hope that he does indeed have some sway in the overall 3D future strategy decision making process.
Would totally be on board with Apple Sherlocking Modo though
Ah yeah... Modo. Bought a license way back at the start. v101 I think it was. It sure had a lot of promise back then. I just couldn't get on board with their decision to go with a stack approach for their shading system, instead of going with a node-graph. That was an odd choice. So by v502 or so I was out. I heard that as they kept adding features it started to suffer from a lot of slowdown/bottleneck issues, around the time Foundry picked it up. And with Foundry seemingly not investing all that much R&D into it, I wonder if it has gotten any better.
Apple should crack open that fat wallet, buy up a bunch of 3D software companies, and start work on a 3D/DCC suite codenamed "Facts of Life"...
"You take the good, you take the bad..."
;^p
"You take the good, you take the bad..."
;^p
Modo went downhill fast after Brad left. I have respect for what Shane is doing for the Modo team, he seems like a good guy but Foundry is not giving Modo the support it deserves. I paid for it from 10.2 - 15.0 and I just couldn’t justify it any longer. Blender has come way too far to ignore and even putting $150 worth of plugins into Blender puts it in a very competitive spot.
Apple making their own 3D software would be amazing! Just seeing what they have done with Logic and FCP shows what they can do when the software is specifically made for the hardware. Realistically Apple probably won’t go that route but maybe there’s hope for a company like Affinity.
Apple making their own 3D software would be amazing! Just seeing what they have done with Logic and FCP shows what they can do when the software is specifically made for the hardware. Realistically Apple probably won’t go that route but maybe there’s hope for a company like Affinity.