Attention: Metal, Big Sur and Apple Silicon
- Thread starter Martin
- Start date
MonkeyT
0
But that also increases its odds of it hanging around as long as Rosetta2 does.
Last edited:
I was wondering why is it that Affinity and other software can update for OS 10 thru M1 the latest Mac OS and it works fine some other companies too. But Cheetah can not? I know nothing of programing but was curious. I keep waiting to be abandoned by Affinity but it hasn't happen as of yet.
Thanks an answer is not necessary as I'm curious about a number of things that have no answers.... smiling.
Thanks an answer is not necessary as I'm curious about a number of things that have no answers.... smiling.
Affinity started out relatively recently with a much more modern starting point in terms of the libraries used, while Cheetah has roots way back to the Carbon days. It is simply more complicated to move on that basis alone.
Also Affinity is a 2D application and can tap into the 2D Apple frameworks which are very functional. 3D libraries from Apple are significantly more lacking so there is a lot more work to be done for a 3D port.
Correct me if I am wrong, but the two renderers in Cheetah are based on open source renderers that don't support Metal at all, so also there he is pretty much on his own.
Just my 2 cents.
Sure, but they are still using OpenGL even if rendered on the CPU.
I had spent a month to port my two apps(One runs on iPadOS, another one runs on macOS) from OpenGL to Metal successfully, I have a few experiences to share.
At first, I used Apple's math library which included in the 'MigratingOpenGLCodeToMetal' sample code, it is mainly a left-hand math library, it is elegant and fast, but I gave up the library finally.
The reason is, 3d models usually have right-hand matrices and counter-clockwise order triangles stored in the file, I can't use them directly, the conversion lead to many mistakes.
So I decided to use the same math library written for OpenGL, to my surprise, the math library works so well, I just need to convert the OpenGL API to Metal API, it is relatively simple than converting the math library.
In the macOS version, I uses 'if else' statement to support both OpenGL and Metal, if the machine does not support Metal, it uses OpenGL, otherwise, it uses Metal, OpenGL and Metal use the same data structure, the only difference is, in OpenGL, the app culls the back face, in Metal, it culls the front face.
To port the shader, I convert them line by line manually, I spent about two days to port the shaders.
At first, I used Apple's math library which included in the 'MigratingOpenGLCodeToMetal' sample code, it is mainly a left-hand math library, it is elegant and fast, but I gave up the library finally.
The reason is, 3d models usually have right-hand matrices and counter-clockwise order triangles stored in the file, I can't use them directly, the conversion lead to many mistakes.
So I decided to use the same math library written for OpenGL, to my surprise, the math library works so well, I just need to convert the OpenGL API to Metal API, it is relatively simple than converting the math library.
In the macOS version, I uses 'if else' statement to support both OpenGL and Metal, if the machine does not support Metal, it uses OpenGL, otherwise, it uses Metal, OpenGL and Metal use the same data structure, the only difference is, in OpenGL, the app culls the back face, in Metal, it culls the front face.
To port the shader, I convert them line by line manually, I spent about two days to port the shaders.
Actually Cheetah's rendering itself shouldn't use OpenGL. It's the viewports that do (where you model, create textures etc).
So, M1 is beating midrange desktop CPUs. Also cinebench scores for M1 are just behind maxed out MBP 16”. This is for *entry level* machines.
What will people complain about when Macs are better, faster, and cheaper than PCs?
What will people complain about when Macs are better, faster, and cheaper than PCs?
The price and Apple's policy to make their customers fully dependent from them as you can't even replace a simple hd anymore (they delibarately try to make it impossible to repair their gadgets through 3rd-party firms).
We don't complain about the tech. But the entry level isn't very interesting for work with 3d, so there isn't much to be excited about for most of us. And I for myself aren't excited about any hardware anymore for decades but for the possibilities it opens for me. So if Apple will present the fastest pro computer in 2 years or so, I simply don't care if it's as expensive as today's macpros, because it will be almost as unreachable as a cray (thinking about it, wouldn't be bad to have one in the attic for rendering purposes ...).
I’m not normally interested in “entry level” devices. But this “entry level” device clobbers high-end devices.
forums.macrumors.com
FCPX performance on M1 chip
This is from a photographer in Australia. He got the device today and posted this on Weibo in Chinese. I guess I’ll translate a bit. Exporting H.264 Sony 10 bit 422 footage with just one rec709 lut takes: - 11 minutes and 30 seconds on his iMac Pro with Vega 56 and 128gigs of ram. - 10 minutes...
forums.macrumors.com
Martin
0
Thanks for the link. I was already curiously awaiting the Cinebench 23 score since they give a good indication how fast Cheetah3Ds will be. Raytracing is a quite specific workload due to it's very random memory access pattern. So I was very curious how the M1 chip handles it. Things like geekbench always give a bad prediction for that workload.
The 7,500 points of the Apple M1 are actually pretty good for an laptop like the MacBook Pro. That's almost as fast as my MacPro 2013 (12 core) which achievs 8260 points. That's pretty cool for on the go. But it's also slower than the AMD Ryzen 4800U ( U= 10-25W low power part) with 10,100 points. So this result is a little bit a mixed bag.
I think it's no surprise that Apple can built very competitive mobile chips. But I'm curious how they want to produce big fat workstation CPUs with their very limited sales of the MacPro. Designing such a big chip for such a low volume product makes the chip crazy expensive. Maybe they will design their own server CPUs too and will use them or they go multi-chip.
So Apple still has some miles to go to become king of desktop if you compare it with the 74,000 Cinebench R23 points of an AMD Threadripper. And much faster Zen 3 Threadrippers will be released soon. So I can't see Apples 3x, 5x, 7x ... performance domination yet. But I would be happy if they prove me wrong.
Exciting times ahead. More competition is always good for the customer.
I'm not sure but I think Cinebench is only using the CPU cores while a Cheetah3D Metal version would use also the GPU cores of the M1 chip and therefore be faster in comparison?
ArsTechnica found the M1 to be as fast or faster than high end desktop CPUs: (Not the fastest desktop CPUs.)
arstechnica.com
It seems to me that Apple can address the workstation market by making one design that services the high-end iMac through to workstation SKUs. It won’t be the fastest system they *could* make, but it will still be fast enough. And the margins will be ridiculous…
It’s also not clear how the economies of scale work in this case. Can Apple make 16GB amd 8GB parts on the same wafers and just vary the mix based on demand? In such a case Apple has a lot of flexibility.
Finally, Apple is a massive consumer of datacenter computers. As long as five years ago, Facebook and Google were among the largest PC manufacturers in the US. (Amazon is likely bigger still.) Apple’s own internal servers might allow economies of scale…
Hands-on with the Apple M1—a seriously fast x86 competitor [Updated]
Apple's M1 proves that ARM can compete with x86 in high-end systems.
arstechnica.com
It seems to me that Apple can address the workstation market by making one design that services the high-end iMac through to workstation SKUs. It won’t be the fastest system they *could* make, but it will still be fast enough. And the margins will be ridiculous…
It’s also not clear how the economies of scale work in this case. Can Apple make 16GB amd 8GB parts on the same wafers and just vary the mix based on demand? In such a case Apple has a lot of flexibility.
Finally, Apple is a massive consumer of datacenter computers. As long as five years ago, Facebook and Google were among the largest PC manufacturers in the US. (Amazon is likely bigger still.) Apple’s own internal servers might allow economies of scale…
If you compare actual hardware with Apple's silicon of the future, that's certainly true. Nvidia bought ARM and AMD has the same licenses as Apple. So we plain don't know how the hardware will compare in two years or so as those firms probably aren't that much behind (Apple still has an advantage). To really own the market (which they could) Apple would have to sell their machines at somehow reasonable prices (which they probably won't). So, we plain don't know. I somehow tend to think that they will be best in class but it's pure speculation, could as well start reading tea leaves and polish my crystal ball..
Last edited:
M1 single core performance emulating x86 is better than any intel Mac…
www.macrumors.com
Hasdrubal: yes, I basically agree. But, setting aside crystal balls, here’s the history — Apple has built and sustained significant leads design, usability, manufacturing, supply-chains. And now they have an entry-level cpu that blows away anything even vaguely comparable.
Apple Silicon M1 Emulating x86 is Still Faster Than Every Other Mac in Single Core Benchmark
The first native benchmarks of Apple's M1 chip appeared on the Geekbench site last week showing impressive native performance. Today, new...
www.macrumors.com
Hasdrubal: yes, I basically agree. But, setting aside crystal balls, here’s the history — Apple has built and sustained significant leads design, usability, manufacturing, supply-chains. And now they have an entry-level cpu that blows away anything even vaguely comparable.
(I certainly wouldn't buy any intel stock, that's for sure)
The "history" part is what somehow hems my enthusiasm: the prices, always the prices.
The good thing is, with blender probably out of the way for a while, I hope Cheetah will gain some more users out of this (it's one of today's few products that aren't run by greed for which we can't Martin applaud enough), so he'll get something out of so much work.
The "history" part is what somehow hems my enthusiasm: the prices, always the prices.
The good thing is, with blender probably out of the way for a while, I hope Cheetah will gain some more users out of this (it's one of today's few products that aren't run by greed for which we can't Martin applaud enough), so he'll get something out of so much work.
Swizl
0
Thanks for the update Martin! Unlike some developers, you've always kept your users informed and I (we) greatly appreciate that. We're all here because we love the app. I know it's difficult to work around these obstacles Apple keeps throwing in you (the developers) way, but I think once the transition is over, C3d will be even better for it. We all understand you are single person developer and that it will take a little more time to convert the code over. Especially now that you have two issues to contend with.
Whenever v8 comes out, I'm here with my wallet ready to buy that next version.
Whenever v8 comes out, I'm here with my wallet ready to buy that next version.