The 12” model shown specifically had a lot of hard drive failures. My 17” model (actually I think aluminum, not titanium, but it was a G4) had a motherboard failure very early on. But after that was fixed it still works fine 22 years later! It even operated for several years continually as a server circa 2009. At one point it had an uptime of over 1 year. Never had any problems!
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Software Engineer Runs Generative AI on 20-Year-Old PowerBook G4
- Thread starter MacRumors
- Start date
- Sort by reaction score
Aw come on, that stuff was over by 2005 anyway, we were busy ramping up for the real party that hit in 2008.
AltiVec snd the G4/G5 was very powerful.
BeOS ran so fast on the G3 compared in MacOS. There just wasn’t a lot you could do with it.
You are aware this is a small 110M parameter model? It's very limited in capabilities, which is why it works.
Not saying optimization isn't important, but this isn't enough optimization to make even ChatGPT 2 work on this, which is a 1.5B parameter model.
I’ve said it over and over again. Back in the early 90’s the things you could get a Sinclair Spectrum to do with only 128k of memory was amazing. Over time we stopped optimising and started relying on Moore’s Law instead. The G4 and G5 PPC chips and the ones that followed that ended up in the PS3 and XBox360, were classed as supercomputing chips at the time. Even today, projects like TenFourFox have shown the PPC chips (and the Intel core Duo which superseded them) to be still remarkably capable if just given optimisation efforts. G4 tends to surprise the most because people don’t account for the AltiVec effect.
Yes, the vector unit was impressive at a time when PowerPC was also impressive.
Then again, we were working on AI in the late 1970s and early 1980s. With enough effort, storage, and working memory, you can do many things thought impossible.
Then again, we were working on AI in the late 1970s and early 1980s. With enough effort, storage, and working memory, you can do many things thought impossible.
BeOS was a slick piece of work. Too bad the software for it scared people away. ☠️BeWare.☠️🙃🙃
A comparison between the Altivec vs the AVX instructions they added in the 2012 processors would be interesting.
And yet I get work done just fine on Linux. Diminishing returns of useful features set in long ago.
Perhaps, but IBM/Motorola unfortunately never made a G5 for a laptop. One of the reasons Apple switched to Intel
I always found it absurd that Linux had better out of the box driver support for my laptop than Windows does. I switched to Fedora for roughly a year (and tried a few other distros). Each time I installed it, WiFi just worked, I went back to Windows and had to buy a USB WiFi dongle just to complete the install.
I completely agree with that. Software optimization can sometimes be hard as in this example, but it pays off in the end. Most software runs, but completely inefficient. I’ve spent the last 10 years optimizing websites and the amount of resources wasted is just incredible. If all websites are properly optimized, the World Wide Web would just need about 20% of all the resources currently in use (that’s storage, CPU, RAM, energy). Yes, those optimizations cost time and money, but the savings might be just as big, but that’s not what’s being billed for as the investment doesn’t pay out the next day.
Taking it to the extreme: If you look at a classic GameBoy game, those games are < 1 MB in size, because everything was programmed down to bare metal. If someone would make a clone of GameBoy game these days, they most likely do that in JavaScript and your install would already need 300 MB, and it would require 100 MB of RAM minimum instead of the 8 + 8 KB and each action would cost hundreds if not thousands of CPU cycles, because layers are build on top of other layers and no single layer is completely optimized.
Last edited:
Compared to now? Yeah. Simpler times. I was there. It was scary, uncertain, the death knell to the 90s and a seemingly innocent childhood. But not every day felt like an existential crisis unfolding that threatened the very fabric of society in the United States so yes, simpler times.
It's been available for a while. No jailbreak needed as AS and macOS isn't locked-down:
Asahi Linux
asahilinux.org
Last edited:
BeOS ran very well on my 604e but it didn't do much of anything.
At some point, they realized that they were going to add a lot of code to fill in the blanks, and then, they started working on the x86 version.
I suspect that everyone running a Mac was glad that the renewed Apple was going to use NeXTStep instead of BeOS as a new basis.
I hope it’s a job at him in the most dorky product launch ever. That picture of tim on Vanity Fair wearing the Apple Vision Pro instantly made all Apple vision pros a buzz kill for anybody who wants to be a non-dork.
It’s almost as bad as when the Segway was launched. The first video we all saw was Bill Gates driving one ( with a bicycle helmet that made him look as cool as Michael Dukakis wearing a helmet riding in a tank during his presidential)
It should’ve been Puff Daddy in his bathrobe. …
This is quite the response! I am really enjoying the discussion here.
This is nothing more than a tech demo, but a fun one nonetheless. Many of you have correctly pointed out that this 110M model is puny in comparison to some of the larger multi-billion parameters. This setup would be lucky to produce a token in an hour, if we could even somehow fit the model into a 32-bit address space.
I have to agree that these 12" PowerBook G4s are among my favorite laptop designs. I love a 4:3 display for coding, and the edge to edge keyboard makes good use of the available horizontal space. I have been looking for a nice quality 12" (or less) laptop for years and really nothing comes close. I daily drive an M2 MacBook Air 13" these days, but back in 2006-2010, one of these PowerBook G4s was my daily driver.
It does get out sometimes! Here's a pic from mid-2023 when it flew with me from San Francisco to Toronto.
I also fabricated what I think is probably the only USB-C to Apple ring-style power connector. This uses a USB-C trigger module to request 20V from a power bank. My trusty Anker 737 successfully ran this laptop for nearly 5 hours while I happily played StarCraft.
This LLM effort is a fun little project. Thanks for all of the supportive comments.
This is nothing more than a tech demo, but a fun one nonetheless. Many of you have correctly pointed out that this 110M model is puny in comparison to some of the larger multi-billion parameters. This setup would be lucky to produce a token in an hour, if we could even somehow fit the model into a 32-bit address space.
I have to agree that these 12" PowerBook G4s are among my favorite laptop designs. I love a 4:3 display for coding, and the edge to edge keyboard makes good use of the available horizontal space. I have been looking for a nice quality 12" (or less) laptop for years and really nothing comes close. I daily drive an M2 MacBook Air 13" these days, but back in 2006-2010, one of these PowerBook G4s was my daily driver.
It does get out sometimes! Here's a pic from mid-2023 when it flew with me from San Francisco to Toronto.
I also fabricated what I think is probably the only USB-C to Apple ring-style power connector. This uses a USB-C trigger module to request 20V from a power bank. My trusty Anker 737 successfully ran this laptop for nearly 5 hours while I happily played StarCraft.
This LLM effort is a fun little project. Thanks for all of the supportive comments.