This project has been so impressive and exciting. I hope power management is a serious concern once they nail all the reverse engineering and functionality. Battery life would be a huge differentiator for the day to day macbook usage and traditionally Linux has been terrible at that due to lack of proprietary & highly tuned drivers
How likely is such tuning to happen within the practical lifetime of these devices? It's already been years and Linux isn't even feature complete on newer models.
Apple's indifference to OSS does not give me much hope.
I would assume that once you have fully functional support for entire historical SoCs, the effort stops being "get most things working on most chips" and starts becoming "adapt already-working things to the newest chips". Yes, it's been years, but there's still so much hardware that Asahi doesn't have good support for yet, and that's dividing the efforts to support the newest chips. I'm sure that once all the existing chips are well-supported, new chips will be fairly quick to bring up.
I've daily driven Linux for years, even on laptops. Yet they're always plugged in because I've yet to see Linux out perform battery life of native OS's. Well, except for very niche and pared down Linux's.
Rather it seems that as time goes on Linux just outgrows old devices entirely, again excluding very niche distros. Not that it makes them increasingly efficient.
Have you seen Linux getting more efficient over time? What's your use case?
The new project leadership team prioritized upstreaming the existing work first.
Once the upstreaming was largely done, they first mentioned beginning the work of adding M3 support in the February progress report.
Now we have this:
> With all of this progress coming on top of milestones already reached on M3, we are pleased to announce that we are almost ready to cut an official release! We will have more to say about this in the coming weeks, so stay tuned!
Given the enormous lifetime of these devices, I’d say chances are pretty good! The only thing that’s going to make an M series Mac obsolete is when Apple stops releasing updates. If you can run Linux on it, that thing is going to last forever.
My concern is for the devices I have, which are usually newer, and for battery life. Linux doesn't even have full hardware support, much less efficient battery life on these now several year old devices.
Maybe things will unlock soon and stay that way long enough for it to be viable for me. I'm just not holding my breath.
Of course it's great to have a supported OS for retired hardware--regardless of battery performance. That's a different use case. And I salute all those who work to keep old devices out of landfills.
Apps. We'd love a completely different OS and it is viable on desktop because web almost does everything but on mobile you're basically locked out of very essential functionality like banking, transit, messaging etc. The compatibility layers aren't good enough (yet) to offer a seamless experience.
Nothing is stopping you. You're free to use a desktop web app that either doesn't work or is terribly slow or has limited features on a mobile web browser. I'd love it if all apps were PWAs but that is not the case.
My broker has a much more fully featured web app then it does on iOS or Android. You can do things in a browser that the phone apps will send you to a browser to do. BofA is just as good of a web app. Maybe people are just used to using their phones?
Good for you. Nobody is challenging what a web app can or cannot do. I love PWAs but services I need every day either don't have a web interface at all, have limited features or are unusable on a small screen let alone having a PWA. This is not about technical abilities but about what companies actually do.
Asking for a cheque deposit is like asking for a car with a hand crank. People outside USA are going to look at you like on a member of a Sentinel tribe.
This is true but the problem is every now and again some of us have to deposit a cheque because even if we would love for cheques to go the way of the Canadian penny they are very much still a thing
No you cannot, as of yesterday. The mobile app doesn't even allow your choice of camera app, to avoid even the tiny amount of fraud that doing so might enable.
Exactly. This is how I design agents as well. I essentially treat the agent as the web/mobile app, cli tool os API library not as part of my backend even though there is where it runs. The backend doesn't treat the agent in any special way. It simply gates all action based on the permissions of the user/guest that is using the agent.
In my experience, Dart/Flutter has been so much better than React. Go has been really good for the backend. Basically if the framework/language gives you structure and one way to do things, agents tend to create less mess with less guardrails from you.
The routed experts needed for the next tokens that are not already in memory need to be read from the SSD, so the speed becomes SSD reading bound and the larger the memory, the faster the inference.
Is there a particular quant of DS v4 Flash you'd recommend that works on 64GB machines? None of the antirez versions on HF look small enough?
Also, FWIW, I've been experimenting with Laguna-S-2.1. It runs reasonably quickly (llama.cpp, IQ2_M quant) but the outputs so far aren't impressive, and it gets stuck and perseverates. Very subjectively, at that level of quantisation, it seems to perform worse than Qwen 3.6 27B at Q4_K_XL.
For a dense model this would be a limitation, but not all of a MoE model needs to be in memory, but the largest part of a MoE are the routed experts.
Some parts are needed to generated every single token and these really should fit in memory, but the router experts that are not neeed can rest on SSD and be read only if they are needed, so... you can run MoE models bigger than you memory, try the IQ2XXS.
It should work on your 64 GB after you enable SSD mode in DwarfStar (in MoEspesso it enables itself), while being slower, so... I am really hoping for good models between the 50-120 GB other than Laguna, there is a big gap right now unfortunately.
Agree on the sizing - selfishly, something like a 60B MoE would be great - fast on big machines, and a 4 or 5 bit quant should fit in 64GB and still work well.
