Leaving FreeBSD with broken heart

S

startrooper

Good morning FreeBSD community!

I am traditionally a Linux user and moved to FreeBSD a little more than a year, and man.... I loved it! It is some of those moments we say "how have a lived until now without it?".
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around. I thought hard about dual boots and VMs, but I have a HW with limited capacity to start with.

Anyhow! I may be leaving but I joined this forum to remain tightened to FreeBSD looking forward to make a comeback!

Cheers!
 
startrooper said:
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around.
Click to expand...
NVIDIA ships drivers for FreeBSD. The thing with CUDA might very well be doable with the current linux emulation layer, since CUDA is essentially just a bunch of libraries.
The drivers themselves support cuda.

But sure, you use the system that fits your needs. Nothing wrong with that. But I feel like adding a couple of lines here, since I know a thing or two about ML/CUDA in the real world.
Simply because I work with mathematicians and we run a bunch of really powerful servers (Linux) for computations.

startrooper said:
but I have a HW with limited capacity to start with.
Click to expand...
Good luck with that.
If you don't have at least two current NVIDIA cards, you're not going to see a real performance boost.
And even then, I'm not sure whether or not you will even benefit from it.

99% of this stuff is just python.
Sure, there are highly specialized python libraries for concrete calculations, that will drain every bit of power from the system for weeks on end.
So much so that some of the multiple NVIDIA cards simply turn off until the next reboot.

But if you're not doing higher mathematics and you simply want to run some LLM and maybe crunch some additional training data, chances are, you're better off doing it with a current CPU.
Because many of those python libraries don't even properly make use of parallelization. Much less low level, efficient cuda stuff.
And without that, it really doesn't matter if one of your CPU cores runs on 100% while the rest idles, or if the same workload is done on a GPU.
It will take a lot of time either way.
A great example of that would be stemgen. It's faster on my laptop without cuda, than on our computing servers with multiple GPUs.
In fact, it took so long on the computing server, that i simply canceled it. While the same thing took about 10 minutes on my laptop.
 
startrooper said:
Good morning FreeBSD community!

I am traditionally a Linux user and moved to FreeBSD a little more than a year, and man.... I loved it! It is some of those moments we say "how have a lived until now without it?".
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around. I thought hard about dual boots and VMs, but I have a HW with limited capacity to start with.

Anyhow! I may be leaving but I joined this forum to remain tightened to FreeBSD looking forward to make a comeback!

Cheers!
Click to expand...
So you've just created an account over forums.freebsd.org to say that you are leaving FreeBSD ... okay.
 
startrooper said:
Good morning FreeBSD community!

I am traditionally a Linux user and moved to FreeBSD a little more than a year, and man.... I loved it! It is some of those moments we say "how have a lived until now without it?".
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around. I thought hard about dual boots and VMs, but I have a HW with limited capacity to start with.

Anyhow! I may be leaving but I joined this forum to remain tightened to FreeBSD looking forward to make a comeback!

Cheers!
Click to expand...
No worries, we are not going anywhere :)

I know people that run FreeBSD on their desktop/laptop and professionally manage Linux/UNIX/AIX services - so you are not the only one 'forced' to do other things.

You can use CUDA with Linux Compat on FreeBSD if needed:

- https://github.com/verm/freebsd-stable-diffusion

Take care.
 
I would be leaving FreeBSD because with FreeBSD's jails everything is blacklisted by default and you have to know what to whitelist whereas with Linux' firejail everything is whitelisted and you can just blacklist because you know what there is to blacklist.
 
startrooper said:
Good morning FreeBSD community!

I am traditionally a Linux user and moved to FreeBSD a little more than a year, and man.... I loved it! It is some of those moments we say "how have a lived until now without it?".
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around. I thought hard about dual boots and VMs, but I have a HW with limited capacity to start with.

Anyhow! I may be leaving but I joined this forum to remain tightened to FreeBSD looking forward to make a comeback!

Cheers!
Click to expand...
Good luck. Ill take your place. I have been using arch linux for 13 years now and i recently switched to freebsd and i love it. There is a lot to learn here and i see this forum is full of extremely knowledgeable people along with devs actively participating. Its good to be here.
 
startrooper said:
But it is with broken heart that I am planning to move back to Linux as I currently have a urge to expedite my competences in ML/AI and the lack of CUDA integration by NVIDIA
has been a obstacle difficult to get around. I thought hard about dual boots and VMs, but I have a HW with limited capacity to start with.
Click to expand...
Lack of CUDA is a shame but when expediting competencies, I am vehemently against locking myself into a single vendor. OpenCL is a bit shite but will luckily remain being shite long after CUDA has disappeared. Possibly it is one to look into instead?

Also, if you are just now playing catchup with ML/AI, you might want to instead put your feelers out at what the next hype may be and prepare to cash in on that instead?
 
canihazsekooriti said:
It gives a much better starting point. In the end, you can achieve the same on both systems, I reckon.
Click to expand...
well, you can block anything you like, but when after an update suddenly telnet is active and you don't block that - you get 'problems'.
 
This must be some kind of bait. If OP really joined FreeBSD a year ago and loved it and "didn't know how he lived without FreeBSD", why didn't he join a year ago? Why didn't he ask beginner questions, or CUDA questions?

I've been active on some topics regarding CUDA and AI on FreeBSD, and I don't even have a strong Nvidia card! That's because I joined the FreeBSD community for real a little over a year ago and I really don't know how I managed to get by without it.

If this guy was for real, he would get one server with Nvidia cards on Linux, and a workstation with FreeBSD, but that is not interesting or gets attention in this forum like ragebaits :(
 
canihazsekooriti said:
It gives a much better starting point. In the end, you can achieve the same on both systems, I reckon.
Click to expand...
Of course, it's always easier to get things up and running, if everything is allowed by default, instead of you need to think of each you need to allow, and explicitely allow it first.
And in theory - when you respect everything - it makes no difference, yes.
But practically, that's not the case. You don't respect everything.
You're not going through the whole list of things in question, and decide for each to allow or block it (if you even know what each point is.)

So, by principle it's always (way) more secure to blacklist/block everything by default, and whitelist/allow only explicitely selected things you know what it is and why you allow it.
 
omardevonlittle said:
This must be some kind of bait. If OP really joined FreeBSD a year ago and loved it and "didn't know how he lived without FreeBSD", why didn't he join a year ago? Why didn't he ask beginner questions, or CUDA questions?

I've been active on some topics regarding CUDA and AI on FreeBSD, and I don't even have a strong Nvidia card! That's because I joined the FreeBSD community for real a little over a year ago and I really don't know how I managed to get by without it.

If this guy was for real, he would get one server with Nvidia cards on Linux, and a workstation with FreeBSD, but that is not interesting or gets attention in this forum like ragebaits :(
Click to expand...
I dont think thats how it works. I joined these forums few weeks ago, but just recently started using freebsd. I have zero questions about anything because its all well documented and pretty much every question is already answered in the forums.
 
SKull said:
NVIDIA ships drivers for FreeBSD. The thing with CUDA might very well be doable with the current linux emulation layer, since CUDA is essentially just a bunch of libraries.
The drivers themselves support cuda.

But sure, you use the system that fits your needs. Nothing wrong with that. But I feel like adding a couple of lines here, since I know a thing or two about ML/CUDA in the real world.
Simply because I work with mathematicians and we run a bunch of really powerful servers (Linux) for computations.


Good luck with that.
If you don't have at least two current NVIDIA cards, you're not going to see a real performance boost.
And even then, I'm not sure whether or not you will even benefit from it.

99% of this stuff is just python.
Sure, there are highly specialized python libraries for concrete calculations, that will drain every bit of power from the system for weeks on end.
So much so that some of the multiple NVIDIA cards simply turn off until the next reboot.

But if you're not doing higher mathematics and you simply want to run some LLM and maybe crunch some additional training data, chances are, you're better off doing it with a current CPU.
Because many of those python libraries don't even properly make use of parallelization. Much less low level, efficient cuda stuff.
And without that, it really doesn't matter if one of your CPU cores runs on 100% while the rest idles, or if the same workload is done on a GPU.
It will take a lot of time either way.
A great example of that would be stemgen. It's faster on my laptop without cuda, than on our computing servers with multiple GPUs.
In fact, it took so long on the computing server, that i simply canceled it. While the same thing took about 10 minutes on my laptop.
Click to expand...
Thanks for the message! At work I have a proper high end environment to develop (also Linux based). I want to use my personal computer to test some concept with more freedom, hence, I am not after performance really. At least for now! About the libraries, yes, I have the same impression as you and possibly I could give a try when I get some more time available :)
 
SirDice said:
That's definitely a shame. And I don't get why Nvidia won't enable it on the FreeBSD version of their drivers. Terribly annoying to be honest.
Click to expand...
I hope that will change. I work with linux in many projects and products and time to time I tell people, hey! freebsd seems more stable and reliable. Hope NVIDIA will find a positive business case using freebsd, :)
 
You must log in or register to reply here.
Back
Top