I made this costly mistake exactly a year ago purchasing a M2 Pro with 96GB of ram. And it was taking me 1.3min to generate an image! LOL I then decided to invest into a PC with 4090, and now I made almost 1 image per second with TurboXL - But still I need lot more power to train huge models now, so will soon purchase a H100...
@@burnesmith2 What a drawback! 4090 is a lot faster. What's the solution for interconnectivity of multiple GPUs then? It looks like all Ada Lovelace models can't use NVLink. I wonder why?
You are expecting A LOT for software that is given to you FOR FREE. Developers are not excluding Apple's hardware out of some kind of disrespect for your favourite platform, it is simply because every time you have to support a new platform, the complexity (and cost) of creating and maintaining that system increases dramatically! It is a miracle that something like Stable Diffusion works at all on Apple silicon - there is 0 financial incentive at the moment. And much, much of the work to make that possible is done by people who receive no compensation whatsoever, open source, on their own time. And I'm writing this to you from my M1 Pro, bought right after they came out. I have not had a PC at home for probably 20 years now... until last year, when I realized that if I wanted to do some AI related work at home, I didn't really have much of a choice and needed to get an NVIDIA based workstation. I use it remotely from my Mac, almost exclusively. So with respect from one Apple fan to another - it sucks, but for now, that's just the way things are.
sorry, could you share your system: remote PS unit + mac m1? I'm increasingly coming round to the idea of having to do that too (I actually have a macbook air m1 13" in conjunction with an old 27" iMac as a monitor 🙈). And I don't know yet which side to start from to finally stop torturing myself with SD on cloud services
@@Shisgara77 My Mac is just a regular MacBook Pro 16" with 32gb of RAM (2021 model). For the PC shopped at my local computer store and picked what was on sale at the time. I ended up with an AMD Ryzen 7 5700X 8-Core Processor @ 3.60 GHz with 32.0 GB of RAM. For the GPU I picked the 4070 Ti (12g). The rest is just standard stuff, pick a computer case you like, get some fast storage (NVMe). If I were to change anything I would have tried harder to find a 3090 with 24g, VRAM usually is more important. Had I had the budget, a 4090 for sure, why not. So the formula is basically set a budget, buy the best GPU you can within that budget. For remote access, I use Parsec for remote control when I'm using Windows + set up an SSH server so I can access via terminal. Then I simply run the tools and share them on a private IP. That being said If you find a good service provider in the cloud, I'm not fully convinced that having your own PC is that much more economical. Also in theory you could have access to much much more hardware than you could afford. I wanted to use it for VR stuff as well so I really needed the hardware.
@@saiborg_ see my answer to @Shisgara77 below. The short recommendation is simply set a budget, buy the best GPU you can within that budget that leaves you with enough money for a basic 32gb Windows machine (AMD or Intel, doesn't really matter, it doesn't have to the latest gen). For remote access, use Parsec, set up SSH if you are comfortable working in the terminal or Linux. Then run your tools locally and share on a local IP address.
It is natural that MacOS has much fewer software base. MacBooks/MacStudios are a small segment of PC+Laptops market(because of the fact they are hugely overpriced for their specs). If you are a coder and use Apple hardware with opensource projects - try to add your effort to make your system supported.
About open source there is no "person in charge" there is only people who most part of the time work on there free time. So please be greatfull that those things exist. And If you feel concerned feel free to write something about it. Tutorial code or else. But please don't act as there is someone who have to do something for you. There nothing in this universe who must do something for the other.
why buy a linux dedicated laptop when you can install most linux distro's on pc's which ship with windows? I installed ubuntu on an alienware with 4090 on sale from Microcenter and I was up and running in comfyui at lightning speed.
"If your making a professional machine..." - which the Macbook Pro line is. Many professionals - photographers, videographers, software developers, scientists and engineers use these Macs. You have a special case - a set of tools you want to use for generative AI, that were not initially designed to run on a Mac, but later some have them have been adapted to macOS. Just buy the right tool for the job, whether it is a Mac or something else.
I happily ditched Apple for their shady practices. Set aside that they're the poster child for closed sourced and anti-repair... I'm glad countries outside the US have been suing apple for being horrible humans. ComfyUI for XL and turboXL models is so cheaply done in the PC world (not the razerblade you chose, that's buying a overpriced brandname, like buying into Apple). Your price comparison is not realistic kanka. The best values will be the newest Dell's (whether windows or linux). Virtual PCs in the cloud works well too (full windows install for about $.60 per hour), but requires internet. I'm currently building my nomand setup using a handheld pc with eGPU. Keep in mind that oculink or thunderbolt4 has more than enough bandwidth for generative AI. ComfyUI on the beach soon. EDIT: One of the most important things is vram, an eGPU means you can choose a non-crippled GPU with 24gb. The only laptops offering that are not mobile version of RTX4090 (mobile is different than desktop, the product names are misleading).
@@ESGamingCentral Haha... Apple now announced their version of Recall which seems even scarier. They're mining metadata of individual apps! Crazy! which PR teams think they can explain this invasive strategy away? MS, as expected, are now turning recall off by default. I wonder if Apple will do the same.
I don't get his rant: so the tensors/pytorch developers should get out or their ways just to support Mac LAPTOPS users??? for free?? You spent 5k on that Mac have you tough about sponsoring a developper to help the mac port? Apple got their money how about the devs? Nvidia has been supporting all the AI community all those years with libraries documentation AND money, what has Apple done? This "shame on the developpers" jab just doesn't sit well with me, those libraries are opensource what have YOU done? Apart from that isn't the tuxedo laptop ram replaceable? meaning you could buy 128gb of ram separately?
"Buddy, Tensorflow is owned by Google and Pytorch is owned by Facebook. Many developers, small companies, and individuals make their living by building their solutions on these available AI libraries. They earn a lot of money from Mac, Windows, and Linux users for using their technology. So, it's the other way around. Both of them have so much money that you don't need to worry about them. :)
There is Apple ecosystem and there is the world outside of it. Every PC device works on Mac but but apple device do not work on pc . You would not "switch to other ecosystem" you would get out of Apple prison.
You can call it a prison if you like, but the benefits of the design decisions seem pretty obvious to me. You mention “devices,” I’m not sure what that’s about but in the world of software MacOS is Unix-based. It can run anything that runs on Linux. Windows is the odd-man out. Windows is the OS with its own weird DOS screen instead of a normal terminal, a registry editor instead of standard dot files for config, and constraints that are at odds with improvements to CPU architecture (like max page size.) Apple controls all of that. It doesn’t have to worry about fighting with the different companies providing the CPU, GPU, laptop screen, etc. That’s why their ecosystem runs so seamlessly. They control the instruction set, the monitor resolution, and how their system architecture uses all of those resources. They intentionally make it difficult for any 3rd party to modify or build on without going full-Apple, because if a 3rd party product becomes massively popular AND operates under-the-hood in a way that Apple did not direct, they now have to consider that product and its users in every future iteration of every Apple machine
😂😂😂@@AG-ur1lj you'll never realise freedom once you're in prison, you can always say it's an Expensive prison but a prison is a prison no matter how much exquisite it is.
@@W-meme calm down drama queen, the entire point is you can’t avoid the “prison.” The idea that Apple is somehow more of a “prison” then Windows or any of the PC hardware companies you buy from is pure delusion.
How much RAM do you have on your MacBook? How much do you need to use AI locally?
I made this costly mistake exactly a year ago purchasing a M2 Pro with 96GB of ram. And it was taking me 1.3min to generate an image! LOL I then decided to invest into a PC with 4090, and now I made almost 1 image per second with TurboXL - But still I need lot more power to train huge models now, so will soon purchase a H100...
so what would be the equivalent price but better performance in windows when compared to m3 max?
H100? Damn! I'd go for multiple 4090s.
No NvLink for multiple 4090s, seems like 2 A6000s is the way.
@@burnesmith2 What a drawback! 4090 is a lot faster. What's the solution for interconnectivity of multiple GPUs then? It looks like all Ada Lovelace models can't use NVLink. I wonder why?
M2 Pro with 96GB RAM doesn't exist and TurboXL is a completely different model that can be 50x faster all by itself.
Having both Mac and PC isn't a bad thing, you should be proud!
Corgi Digm - would you please find out how long it takes to render a SINGLE image, using 1024 *1024 pixels using “Img2img” on your M3 mac?
I just watched the whole video. can't code. but enjoyed your video! ty
I see you have a performance monitor installed. Were you able to install the crystools node without any problems? I can't do it on my m2max
You are expecting A LOT for software that is given to you FOR FREE. Developers are not excluding Apple's hardware out of some kind of disrespect for your favourite platform, it is simply because every time you have to support a new platform, the complexity (and cost) of creating and maintaining that system increases dramatically!
It is a miracle that something like Stable Diffusion works at all on Apple silicon - there is 0 financial incentive at the moment. And much, much of the work to make that possible is done by people who receive no compensation whatsoever, open source, on their own time.
And I'm writing this to you from my M1 Pro, bought right after they came out. I have not had a PC at home for probably 20 years now... until last year, when I realized that if I wanted to do some AI related work at home, I didn't really have much of a choice and needed to get an NVIDIA based workstation. I use it remotely from my Mac, almost exclusively.
So with respect from one Apple fan to another - it sucks, but for now, that's just the way things are.
I am in the same boat you are. Looking at investing in an NVIDIA PC. Can you recommended what machine you have setup and your remote workflow?
sorry, could you share your system: remote PS unit + mac m1? I'm increasingly coming round to the idea of having to do that too (I actually have a macbook air m1 13" in conjunction with an old 27" iMac as a monitor 🙈). And I don't know yet which side to start from to finally stop torturing myself with SD on cloud services
@@Shisgara77 My Mac is just a regular MacBook Pro 16" with 32gb of RAM (2021 model). For the PC shopped at my local computer store and picked what was on sale at the time. I ended up with an AMD Ryzen 7 5700X 8-Core Processor @ 3.60 GHz
with 32.0 GB of RAM. For the GPU I picked the 4070 Ti (12g). The rest is just standard stuff, pick a computer case you like, get some fast storage (NVMe).
If I were to change anything I would have tried harder to find a 3090 with 24g, VRAM usually is more important. Had I had the budget, a 4090 for sure, why not.
So the formula is basically set a budget, buy the best GPU you can within that budget.
For remote access, I use Parsec for remote control when I'm using Windows + set up an SSH server so I can access via terminal. Then I simply run the tools and share them on a private IP.
That being said If you find a good service provider in the cloud, I'm not fully convinced that having your own PC is that much more economical. Also in theory you could have access to much much more hardware than you could afford. I wanted to use it for VR stuff as well so I really needed the hardware.
@@saiborg_ see my answer to @Shisgara77 below. The short recommendation is simply set a budget, buy the best GPU you can within that budget that leaves you with enough money for a basic 32gb Windows machine (AMD or Intel, doesn't really matter, it doesn't have to the latest gen). For remote access, use Parsec, set up SSH if you are comfortable working in the terminal or Linux. Then run your tools locally and share on a local IP address.
Sorry, the video is too long and I want to ask you something, have you tried Apple MLX with stable diffusion? is it usable?
It is natural that MacOS has much fewer software base.
MacBooks/MacStudios are a small segment of PC+Laptops market(because of the fact they are hugely overpriced for their specs).
If you are a coder and use Apple hardware with opensource projects - try to add your effort to make your system supported.
Please enable video subtitles.🙏🏻
راهی برای اجبار کردن زیرنویس ها وجود دارد اما پیچیده است آیا واتس اپ دارید تا بتوانم این موضوع را راحت تر برای شما توضیح دهم
About open source there is no "person in charge" there is only people who most part of the time work on there free time.
So please be greatfull that those things exist.
And If you feel concerned feel free to write something about it. Tutorial code or else.
But please don't act as there is someone who have to do something for you.
There nothing in this universe who must do something for the other.
why buy a linux dedicated laptop when you can install most linux distro's on pc's which ship with windows? I installed ubuntu on an alienware with 4090 on sale from Microcenter and I was up and running in comfyui at lightning speed.
Don't forget some Win Laptops have CPU cap, that limit the performance of the gpu
"If your making a professional machine..." - which the Macbook Pro line is. Many professionals - photographers, videographers, software developers, scientists and engineers use these Macs. You have a special case - a set of tools you want to use for generative AI, that were not initially designed to run on a Mac, but later some have them have been adapted to macOS.
Just buy the right tool for the job, whether it is a Mac or something else.
get the vision pro as well, no need for extra screens
I happily ditched Apple for their shady practices. Set aside that they're the poster child for closed sourced and anti-repair... I'm glad countries outside the US have been suing apple for being horrible humans. ComfyUI for XL and turboXL models is so cheaply done in the PC world (not the razerblade you chose, that's buying a overpriced brandname, like buying into Apple). Your price comparison is not realistic kanka. The best values will be the newest Dell's (whether windows or linux). Virtual PCs in the cloud works well too (full windows install for about $.60 per hour), but requires internet. I'm currently building my nomand setup using a handheld pc with eGPU. Keep in mind that oculink or thunderbolt4 has more than enough bandwidth for generative AI. ComfyUI on the beach soon. EDIT: One of the most important things is vram, an eGPU means you can choose a non-crippled GPU with 24gb. The only laptops offering that are not mobile version of RTX4090 (mobile is different than desktop, the product names are misleading).
Well enjoy Microsoft Recall.
@@ESGamingCentral Haha... Apple now announced their version of Recall which seems even scarier. They're mining metadata of individual apps! Crazy! which PR teams think they can explain this invasive strategy away? MS, as expected, are now turning recall off by default. I wonder if Apple will do the same.
@@Superfandotfan Samesame. Microsoft seems clueless while trying to innovate. Apple just seems shady in general.
@@Superfandotfan you missed the 📝 memo cause it’s the opposite of Recall.
you should buy the Mac Studio with M4 Ultra that will come out next june. The M4 family will be optimized for this AI processes
I don't get his rant: so the tensors/pytorch developers should get out or their ways just to support Mac LAPTOPS users??? for free??
You spent 5k on that Mac have you tough about sponsoring a developper to help the mac port? Apple got their money how about the devs?
Nvidia has been supporting all the AI community all those years with libraries documentation AND money, what has Apple done?
This "shame on the developpers" jab just doesn't sit well with me, those libraries are opensource what have YOU done?
Apart from that isn't the tuxedo laptop ram replaceable? meaning you could buy 128gb of ram separately?
"Buddy, Tensorflow is owned by Google and Pytorch is owned by Facebook. Many developers, small companies, and individuals make their living by building their solutions on these available AI libraries. They earn a lot of money from Mac, Windows, and Linux users for using their technology. So, it's the other way around. Both of them have so much money that you don't need to worry about them. :)
There is Apple ecosystem and there is the world outside of it. Every PC device works on Mac but but apple device do not work on pc . You would not "switch to other ecosystem" you would get out of Apple prison.
😅 that is quite sarcastic
this. Apple is the only one who has a real ecosystem with their devices, the other "ecosystems" are a joke
You can call it a prison if you like, but the benefits of the design decisions seem pretty obvious to me.
You mention “devices,” I’m not sure what that’s about but in the world of software MacOS is Unix-based. It can run anything that runs on Linux. Windows is the odd-man out. Windows is the OS with its own weird DOS screen instead of a normal terminal, a registry editor instead of standard dot files for config, and constraints that are at odds with improvements to CPU architecture (like max page size.)
Apple controls all of that. It doesn’t have to worry about fighting with the different companies providing the CPU, GPU, laptop screen, etc. That’s why their ecosystem runs so seamlessly. They control the instruction set, the monitor resolution, and how their system architecture uses all of those resources. They intentionally make it difficult for any 3rd party to modify or build on without going full-Apple, because if a 3rd party product becomes massively popular AND operates under-the-hood in a way that Apple did not direct, they now have to consider that product and its users in every future iteration of every Apple machine
😂😂😂@@AG-ur1lj you'll never realise freedom once you're in prison, you can always say it's an Expensive prison but a prison is a prison no matter how much exquisite it is.
@@W-meme calm down drama queen, the entire point is you can’t avoid the “prison.” The idea that Apple is somehow more of a “prison” then Windows or any of the PC hardware companies you buy from is pure delusion.
Apple has been selling overpriced/underperforming systems for over 25 years. Should've done your homework first. Nothing new here.
The M series Max laptops are hardly underperforming.