Computers Built to Last
datagubbe.se
datagubbe.se
Because you need a user base to make the software carry itself.
Also from a power point of view 7W is pretty much what we have to play with if everyone is going to have one powered on permanently.
Get a good case for it so that you can passively cool it and you're good to go for atleast 100 years if the heatpaste doesn't dry up:
http://move.rupy.se/file/pi_4.jpg
I recently found this case, but it's still in the mail:
Maybe covering the solder surfaces with thick, hard resin could help. (So goes repairability.)
[1]: https://www.microcontrollertips.com/when-tin-whiskers-sponta...
Even for commodity parts (Cell phones and Laptops) PCBs and components are built to prevent problems with Sn. Certainly, in Automotive it's a big deal!
A good thermal paste (Arctic Cooling MX series, for example) is good for at least 10 years, and it has almost infinite shelf life, so renewing paste every decade is not bad.
Another viable case is Cooler Master's Pi Case 40 [0].
The nice thing is, it's protected from dust, and lower tray files are open, so you can modify the design and reprint if you want to adapt it to a previous or next generation Pi compatible board.
It also has a hardware button, which you can program.
[0]: https://www.coolermaster.com/catalog/cases/raspberry-pi/pi-c...
I'm using Arctic Cooling's MX series thermal pastes for over a decade, and my eight year old MX-2 installation works as good as day one. It's taking heat away from an Intel i7 3770K, which is not a cool chip, esp. under load, and mine has seen some hotter days under constant load (with a scientific code which is designed to get as much performance possible from a processor, no less).
I have a couple of these Cooler Master cases too, and their out of the box pastes are doing a good job for now. So, I'd rather not fix something if it's not broken. If the temps start to climb, I already have more than enough spare of the paste I use in my systems.
Ah you’re right. I was thinking of experiments with indium foil I had seen[1]. Plus gallium is used in the pastes apparently.
So if the goal is a very long term thermal interface material, I don’t think pastes are the best. A homogenous solid is likely to last longer.
Indium foil fits the bill nicely. Also thermal pads or graphite pads, but I’m not sure how long they last.
[1] https://www.reddit.com/r/buildapc/comments/99fq4y/experiment...
Arctic Cooling states that MX-5 is guaranteed to keep its performance its for 8 years [0]. Considering this stuff is made to interface with 200W+ CPUs, cooking it dry with a Raspberry Pi is very unlikely.
As I said before, I'm cooking MX-2 in two different systems (Q6600 & i7 3770K) for 8 years, and they didn't lose their transfer performance a bit.
From my experience, thermal pads and solid interfaces do not transfer with the same efficiency and cannot absorb heat spikes as well as a good thermal paste. I run my OrangePi Zero with a thermal pad interface to a small heat-sink and that thing runs uncomfortably hot. I can't transfer data faster than 40mbps for a long time or run CPU intensive stuff, because it gets hot, fast. Its form factor and case doesn't allow for a more robust solution either.
Of course the choice is yours, but in my experience, when compared to older paste, or to ordinary white stuff, current generation stuff is leaps and bounds ahead. Especially, Arctic Cooling.
OTOH, used properly, liquid (liquefying) metal should have a better transfer efficiency than paste.
If you have issues, you can also try lapping the surfaces.
As I've said, it's attached to a OrangePi Zero which has a very small, epoxy packaged CPU. So tension-mounting, lapping, or using higher end compounds are out of question.
The whole point of OrangePi Zero is being small all over, so it's happy temp-wise with its normal load, I'm not planning to work on it further for now.
Raspberry Pi 4 is different and has a whole-body heatsink, and its paste is good for now. If it starts to lose cooling capacity, that'll be a different matter all-over.
The entire metal case becomes the heatsink for the processor, and no fan is needed. It's strong and sturdy.
For PSU I use overdimensioned Mean Well that should last 50 years atleast.
You always need a window for failure, redundancy is your only friend in this case!
The SD cards have peaked at 1TB but they are getting cheaper. (so far)
Only use Intel SATA SSDs, I bought X25-E SLC drives from 2011 to get that 65nm 100.000 writes per bit.
But yes, media will have to be replaced, as a whole with open-source software and distributed databases everything will be saved.
If the hardware works, but you can't get software there anymore how usable will it be.
The lower you go from 65nm the brittler it becomes, also smaller and that's how the flash MLC has reached 1TB on the same surface (micro SD) that only allowed 8GB SLC.
My X25-E drives are only 64GB!!!
Photon/Quantum processors will never match electrical in Gflops/W/durability/$.
Magnetic drives are not an option because of energy consumption.
Also with peak hydrocarbons we don't have the energy to keep growing development/building of new things:
So yes your concern is valid, what happens if SD/SSD media explodes in price and eventually is unavailable?
My answer is buy the media you can right now and plan for it's use, charge ALOT for writes! Don't log anything. And try to get SLC while it's still available to buy:
I have calculated that for the 1 million customers I plan to have (I will cap my user base, but since the software in open-source others will hopefully fill the gaps if the product is successful) I have enough drive durability to store and update their records once per day for 500 years.
Set a limit to what you want to do.
Apply the same to monitors, keyboards, mouses, power supplies, cases (passively cooled), batteries (lead-acid), routers, switches, cables etc.
If you play it right the glass in the fiber strand will need replacing before your hardware stack does.
And it probably will get replaced for high latency / low bandwidth messages by P2P radio: http://radiomesh.org
The hardware is wonderfully documented and easy to repair. My Amiga has been running for a quarter of a century now.
And then it all stopped, seemingly because building equipment became cheap enough to do that they started worrying about duplicates. Now you end up with errors that you'd never get if people were still able to learn how nice stuff works, by making it dangerous to publish the designs publicly now no one has detailed implementation references. Sure, things like the RPi and Arduino exist so you can find open hardware but it's nothing like when almost everything was open by default.
I wish things had gone down differently but things made post 2000 I end up throwing out far more often than stuff from the 1980s. Some of that is selection bias but a lot of it is having access to detailed repair manuals.
Hm? Don't 386s do this too? And BeagleBones? MSP430s?
1999 isn't all of a quarter of a century, but there's a Quadra in the other room which will be 30 this year and also still sees some use.
It's not that the new stuff isn't nice, but I still haven't managed to exhaust the potential of stuff from 20 years ago for my music at least. So I can just stay off the upgrade cycle, which feels quite good.
For non-music, save for TLS, Hackernews could of course also work well on old hardware. Most websites you could also get the functionality, if not the exact experience. But alas, that's not the way the world went, and although a ten-year-old PC in 2022 is a lot more "current" than a 10-year-old PC was in 2012, it still seems like we will keep being dragged forwards by developers wanting more shiny things and hardware manufacturers who want to sell hardware.
I did see a YouTube vid last year of an Amiga booting up what I presume to be an old version of Debian. It got there - EVENTUALLY - at least proving that AmigaOS was a vastly more efficient of OS (over Debian, at least).
> Playing against a perfect tic-tac-toe competitor, and playing well, can only ever result in a draw. This may be computationally impressive the first time, but it isn't very fun. Opponents, computer or human, become interesting when they make mistakes--or more accurately, when it becomes clear that they might make mistakes under certain circumstances. Such mistakes highlight weaknesses, which players can exploit as part of a strategy.
The book goes onto explain how Pong on the Atari 2600 achieves a similar level of unreliability, since it would be trivial for the computer to follow the y-position of the ball and never lose. Every 8 frames, the computer neglects to track the ball, drifting behind the position, but recovering its position when the ball ricochets off the top or bottom of the screen. This way the "AI" of Pong occasionally makes mistakes, in order not to be too harsh to the human's ego.
Thorough breakdown of Relay-Moe's logic with photos is here: (3.5MB PDF) https://www.vintagecomputer.net/cisc367/Radio%20Electronics%...
I was honestly expecting him to mention vintage (ie. IBM/pre-lenovo) thinkpads. Those have a cult following behind them, supposedly for their serviceability, durability, and features (nice keyboards/trackpoint). They're certainly more useful and easier to repair than the Amiga. Speaking of which, can something really be considered easy to repair if replacement parts have to be custom manufactured by "hobbyists and small scale businesses", and parts need to be reimplemented via FPGAs? At that point you're probably creating more e-waste than using a raspberry pi or something.
For example: [1] - I'd expect the clock to be circular instead of oval?
[1]: https://upload.wikimedia.org/wikipedia/commons/f/f4/Amiga_Wo...
Later versions of AmigaOS supported higher resolutions, including square-pixel resolutions, but a lot of the UI artwork was still designed with lower-resolutions in mind so it still looks weird.
640x200, doubled to 640x400, but the clock still looks weird: http://toastytech.com/guis/amiga1apps.png
640x512, with a reasonable-looking circle at the top, but the icons in the middle look ridiculous: http://toastytech.com/guis/amiga2bounce.png
640x480 (square pixels!) with a nice clock, a nice colour wheel, and nice icons, but the title-bar buttons and the keyboard shortcuts in the menu look weird: http://toastytech.com/guis/amiga35utils.png
It came out as new, the only issue is one of the display segmemts on fifth number (from left) not lighting up, luckily it is the right bottom one, so it is still "readable".
My HP 28-C (circa 1987) is still going strong, and it is 34 years old by now.
Another thing a computer built to last needs are blob-free mainline drivers. It is sad to have a hardware you just can't use with a better kernel because it is not supported.
Interesting. I knew about the Framework laptop but the Reform seems to go quite farther in being usable and repairable for a long time. Bulky and less advanced, but I like how one can just buy spare battery cells for cheap on Amazon or Aliexpress.
I keep a G4 iBook, iMac, and a couple of Minis around as backup computers because they still just work. I've been thinking about switching some of them to Linux/BSD now that TenFourFox is dead, but they're still useful for a lot of tasks on Leopard. My iBook definitely plays DVDs much more conveniently than most Macs of the last decade.
Also, use whatever tech is used in deep space probes.