That's not to say the two men don't think AI will be helpful in the future. Indeed, Torvalds noted one good side effect already: "NVIDIA has gotten better at talking to Linux kernel developers and working with Linux memory management," because of its need for Linux to run AI's large language models (LLMs) efficiently.
Linux
From Wikipedia, the free encyclopedia
Linux is a family of open source Unix-like operating systems based on the Linux kernel, an operating system kernel first released on September 17, 1991 by Linus Torvalds. Linux is typically packaged in a Linux distribution (or distro for short).
Distributions include the Linux kernel and supporting system software and libraries, many of which are provided by the GNU Project. Many Linux distributions use the word "Linux" in their name, but the Free Software Foundation uses the name GNU/Linux to emphasize the importance of GNU software, causing some controversy.
Rules
- Posts must be relevant to operating systems running the Linux kernel. GNU/Linux or otherwise.
- No misinformation
- No NSFW content
- No hate speech, bigotry, etc
Related Communities
Community icon by Alpár-Etele Méder, licensed under CC BY 3.0
so THATS why we are getting better nvidia support.
i knew it just couldnt be from the goodness of their newly converted hearts.
You didn't know GPUs are used for training/running DNN?
sure, but am i surprised this is the only reason they improved desktop drivers for us? no.
am i disappointed? yeah, a little.
Hahaha. I love it. Fuck closed source hardware gatekeepers.
Nice to see them groveling for performance.
Kneel!!
C'mon, I can joke. Such a cathartic paragraph to read. Intractable cunts.
Did not know the thing about purposefully adding rogue tabs to kconfig files to catch poorly written parsers. That's fucking hilarious and I'd love to have the kind of clout to get away with something like that rather than having to constantly work around other people's mistakes.
I write a lot of scripts that engineers need to run. I used to really try to make things 'fail soft' so that even if one piece failed the rest of the script would keep running and let you know which components failed and what action you needed to take to fix the problem.
Eventually I had so many issues with people assuming that any errors that didn't result in a failure were safe to ignore and crucial manual steps were being missed. I had to start making them 'fail hard' and stop completely when a step failed because it was the only way to get people to reliably perform the desired manual step.
Trying to predict and account for other people's behavior is really tricky, particularly when a high level of precision is required.
It is a developer milestone :) when you learn to be a resilient applicant is about recovery situation you perfect understanding. Fail fast everything else. Repeat 1000 times, you have something
soft failures add complexity and ambiguity to your system, as it creates many paths and states you have to consider. It's generally a good idea to keep the exception handling simple, by failing fast and hard.
here is a nice paper, that highlights some exception handling issues in complex systems
https://www.usenix.org/system/files/conference/osdi14/osdi14-paper-yuan.pdf
Always fail soft in underlying code and hard in user space IMHO
This is why I enjoy programming libraries only I will ever use. "Do I need to account for user ignorance and run a bunch of early exit conditions at the beginning of this function to avoid throwing an exception? Naww, fuck it, I know what I'm doing."
It's the quickest way to prove to yourself that you know what you're doing... Most of the time, anyway...
Sounds familiar, haha.
Hohndel agreed but added that the industry needs to support these smaller projects -- and not only with money. "Companies need to engage with these projects. Have your company adopt a couple of such projects and just participate. Read the code, review the patches, and provide moral support to the maintainers. It's as simple as that."
Really glad he said this, I keep seeing posts about how all these big companies could solve the problem by just throwing money at small projects and while that is better than nothing it would help way more to have their own developers helping to review and fix issues.
Is there a link to this talk (or interview, or whatever this is) but in a video format, or at least a text without all those «SEE ALSO» self ads?
Maybe it's this one (I'm in a rush here 🙂) ? https://youtube.com/watch?v=VHHT6W-N0ak Someone in the comments writes that the full interview is in the channel of Linux Foundation : https://piped.video/channel/UCfX55Sx5hEFjoC3cNs6mCUQ
Sadly no, that one is three months old. Hopefully they'll publish it on the Linux Foundation yt channel or something.
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
Here is an alternative Piped link(s):
https://piped.video/watch?v=VHHT6W-N0ak
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
not yet, you should run an adblocker
If you're still interested it seems that they've uploaded the keynote, see link in my comment:
https://lemmy.deedium.nl/comment/115389
Thanks!
Wait a minute... BS stands for "Beautiful Science" now?
In addition, hardware developers reinvent old ways of doing things and only learn by making all the same mistakes that have been made before. It's sad, but true.
This same criticism is validly launched at software devs all the time lol.
One thing I've anecdotalally seen and heard is hardware guys indicating that something is rock solid and solved because it's old, so building on top of it isn't a problem. Obviously we have to build on the old to get to the new, but if we just skip auditing hardware due to age we end up deploying vulnerable hardware globally. Spectre and Meltdown are an interesting example where I've heard from at least one distinguished professor that "everyone" believed branch prediction design/algorithms were essentially done. Was it adequately assessed from a security POV? Clearly not, but was it assessed from a security POV in general? I have no idea, but it would be nice as a tech enthusiast and software guy to see the other side of the fence take these things seriously in a more public way, in particular when it comes to assessing old hardware for new attack vectors.
Spectre and Meltdown are an interesting example where I’ve heard from at least one distinguished professor that “everyone” believed branch prediction design/algorithms were essentially done.
Interesting to hear this.
Was it adequately assessed from a security POV? Clearly not, but was it assessed from a security POV in general? I have no idea, but it would be nice as a tech enthusiast and software guy to see the other side of the fence take these things seriously in a more public way, in particular when it comes to assessing old hardware for new attack vectors.
Right.
It seems they've uploaded the keynote, see:
Keynote: Linus Torvalds, Creator of Linux & Git, in Conversation with Dirk Hohndel
Cool, thanks for sharing!
Here is an alternative Piped link(s):
Keynote: Linus Torvalds, Creator of Linux & Git, in Conversation with Dirk Hohndel
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.