Llama Habitat Continues To Expand, Now Includes The PSP

August 17, 2025 by Tyler August 14 Comments

Organic Llamas have a rather restricted range, in nature: the Andes Mountains, and that’s it. Humans weren’t content to let the fluffy, friend-shaped creatures stay in their natural habitat, however, and they can now be found on every continent except Antarctica. The Llama2 Large Language Model is like that: while it may have started on a GPU somewhere, thanks to enterprising hackers like [Caio Madeira], who has ported Llama2 to the PlayStation Portable (PSP), the fluffiest LLM can be found just about anywhere.

The AI, in all its glory, dooming yet another system.

Ultimately this project has its roots in Llama2.c by [karpathy], a project we’ve seen used on Pentium II under Windows 98, DOS machines running 486 processors, and even the venerable Commodore 64, of all impossible things. Now, it’s the PSP’s turn. This implementation uses the same 260K tinystories model as the C64 port, upon which it is based. Of course the PSP’s RAM has room for a much larger model, but [Ciao] apparently prefers to run the tiny model faster on this less-ancient gaming hardware.

Its getting to the point that it’s harder to find systems that won’t run LLMs than those that do. Given that Llama2 seems to be the new DOOM, it’s probably only a matter of time before their virtual fur is all over all our old equipment. Fortunately for allergy sufferers, virtual fur cannot trigger a histamine response.

If you know of another system getting LLMs (Alpaca-adjacent or otherwise), send in a tip.

It’s A Pi, But It’s Not Quite A Raspberry Pi

August 17, 2025 by Jenny List 17 Comments

When is a Raspberry Pi not a Raspberry Pi? Perhaps when it’s a Pi Pico-shaped board with an RP3A0 SoC from a Raspberry Pi Zero 2, made by [jonny12375].

Back in the early days of the Raspberry Pi, there was a offering from the Korean manufacturer Odroid, which wasn’t merely a similar machine with a different SoC, but a full clone in a smaller form factor featuring the same BCM2385 chip as the original. It was electrically and software-wise identically to the real thing, which we suspect didn’t go down very well with the Pi folks in Cambridge. The supply of Broadcom chips dried up, and ever since then the only way to get a real Pi has been from the official source. That’s not quite the end of the unofficial Pi story though, because a few hardy experimenters have made Pi clones like this one using chips desoldered from the real thing.

It’s the fruit of a reverse-engineering project to find the chip’s pinout, and it’s a proof of concept board rather than the intended final target of the work. The process involved painstakingly sanding down each layer of a Zero 2 board to reveal the traces and vias. The current board has a few quirks but it boots, making this an impressive piece of work on all counts. We’re looking forward to seeing whatever the final project will be.

If you’re hungry for more Pi-derived goodness, we’ve also seen one using the part form a Pi 3.

Commodore Is Back Selling New C64s, But Should You Buy Them?

August 17, 2025 by Maya Posch 107 Comments

It’s hard to argue with nostalgia, but you can toss a bucket of cold facts over it. In the case of the recent rescuing of the Commodore brand from the clutches of relabeling of generic electronics by [Perifractic] of Retro Recipes, we got [The Retro Shack] doing the proverbial bucket dumping in a new video. Basically the question is whether the fresh Commodore 64 offerings by the new-and-improved Commodore are what you really want, or need.

The thing is that over the decades many people have created all the bits that you need to build your own classical C64, or even buy one off-the-shelf, with people like [bwack] having reverse-engineered the various C64 mainboards. These can be populated with drop-in replacements for chips like the SID, VIC-II, CIAs and others that are readily available, along with replica cases and keyboards. If you crave something less bulky and complex, you can run a bare metal C64 emulator like BMC64 on a Raspberry Pi, or just run the VICE emulator on your platform of choice. There’re also options like the full-sized TheC64 and Ultimate 64 Elite II systems that you can buy ready to go.

Basically, there is a whole gamut of ways to get some part of the C64 experience, ranging from emulator-only to a full hardware DIY or pre-assembled format. Each of which come with their own price tag, starting at $0 for running VICE on your existing system. With so much choice we can only hope that the renewed Commodore company will become something more than Yet Another C64 Experience.

Continue reading “Commodore Is Back Selling New C64s, But Should You Buy Them?” →

A Second Chance For The Single Wheel Monorail?

August 16, 2025 by Heidi Ulrich 27 Comments

Lately, this peculiar little single wheel monorail came to our attention. Built by [extraglide1976], all from Meccano. His build started with modest tests: one gyro obviously flopped. Two gyros geared together ran slightly better. But when he adds active gimbal control, things suddenly come to life – the model shudders, catches itself, and carries on. The final green-roofed locomotive, with LEDs signalling ‘system go’, trundles smoothly along a single rail on [extraglide1976]’s deck.

To be fair, it houses a lot of mechanics and engineering which we don’t find in the monorails of today. We do have quite a few monorails in our world, but none of them balance on a single wheel like this one. So, where did this invention derail?

Outside of theme parks, Japan is one of the few countries where monorails are still used as serious urban transport: though Germany’s century-old Wuppertal Schwebebahn, the lesser-known C-Bahn, China’s sprawling Chongqing and Shanghai systems, Malaysia’s Kuala Lumpur line, Brazil’s São Paulo network, the US links in Seattle and Las Vegas, and India’s Mumbai Monorail prove the idea has quietly taken root elsewhere.

The thing you’ll see in nearly all these monorails is how the carriages are designed to clamp onto the tracks. This is of course the most safe option, but it loses out on speed to the ones that sit on top of the tracks, balancing on one wheel. Such a train was actually invented, in 1910, by Louis Brennan. His original monorail promised faster, cheaper transport, even using existing rails. The carriages leaned into turns like a motorbike, without any intervention from the driver. Two counter-rotating gyroscopes kept the carriage upright, cancelling precession forces like a mechanical Jedi trick.

Back then, it failed commercially, but today? With cheap sensors, brushless motors, and microcontrollers, and intelligent software, why not let it make a comeback? It could carry freight through narrow urban tunnels. With high-speed single-rail pods?

Investors killed Brennan’s idea, but we live in a different time now. You could start out with a gimmicky ‘snacks and beer’ highline from your fridge to your garage. Share your take on it in the comments!

Continue reading “A Second Chance For The Single Wheel Monorail?” →

From Smartphone To A Home Server

August 16, 2025 by Tyler August 27 Comments

Some people like their homelabs to be as big and fancy as possible, with racks of new or surplus server hardware sucking down power. [Hardware Haven] evidently has the opposite idea, given he just made a video about making the cheapest, smallest server possible: an Android phone.

Sure, it’s not going to be streaming terabytes of data at multiple gigabytes per second, but that’s not everyone’s use case. Don’t forget, flagship phones had multiple cores and gigabytes of RAM a decade ago, so even an old and busted smartphone has more than enough power for something like Home Assistant, which is what gets installed in this video.

After considering loading termux and rooting his device for Docker-on-Android, he opted for postmarket OS, the premiere Linux for old smartphones. That’s not because the Linux environment you get with termux wouldn’t work; it’s just that he wanted something native. To that end, he bought a somewhat worse-for-wear Xiaomi Mi A1 from eBay to get hardware Alpine-based postmarket could use.

Software wise, it was just a matter of following instructions and reading manuals — Linux is Linux, after all. The firewall proved to be his main challenge, though trying to branch out from Home Assistant to run Minecraft Server did run into Java issues [Hardware Haven] had no interest in troubleshooting. Hardware wise, though, well — do you want to leave a phone plugged in permanently? Smokey the Bear suggests you not, especially if you live near a forest. Besides, you probably don’t want your server on WiFi, and at least this smartphone wouldn’t charge when using a networking dongle.

That meant phone surgery: the battery came out, and 5 V from an old USB charger was piped into the battery charge controller via a diode. The diode was used for its voltage drop, to bring the 5 V supply down to a believable battery voltage — a buck converter might have been better, but you use what you have, and the diode drop doesn’t dissipate much power. Power dissipation is still one watt at idle, six during a stress test.

Given how cheap the phone was, and how little power this thing sips, [Hardware Heaven] has an excellent answer to those who say homelabbing is a rich person’s hobby. This project also reminds us that while our phones might not be as hackable as we’d like, they’re still far from totally locked down. You can even run NixOS on (some of) them.

Continue reading “From Smartphone To A Home Server” →

MorPhlex: The TPU Filament That Goes Soft After You Print It

August 16, 2025 by Maya Posch 6 Comments

In FDM 3D printing cycles TPU is a bit of a special filament. Not so much because of its properties, but because it’s rather stretchy even as a filament, which makes especially printing certain hardness grades of TPU into somewhat of an nightmare. An interesting new contender here comes from a company called BIQU, who reckon that their ‘MorPhlex’ TPU solves many of those problems. Recently the [ModBot] channel on YouTube got sent a spool of the filament for testing.

The BIQU MorPhlex TPU filament being turned into squishy slippers. (Credit: ModBot, YouTube)

The ‘magic’ here is that this TPU claims to be a 90A TPU grade while on the spool, but after printing it becomes 75A, meaning a lot softer and squishier. Perhaps unsurprisingly, a big selling point on their product page is that you can print squishy shoes with it. Beyond this is claims to be compatible with ‘most FDM printers’, and the listed printing parameters are typical for TPU in terms of extruder and bed temperature.

After drying the filament as recommended for TPU in general, test prints were printed on a Bambu Lab H2D. Here BIQU recommends not using the AMS, but rather the dedicated TPU feeding channel. For the test prints some slippers were printed over the course of two days. In hindsight glue stick should have been applied to make parts removal easier.

The slippers were indeed squishy, but the real test came in the form of a Shore A hardness meter and some test cube prints. This showed an 80 – 85A for the BIQU MorPhlex test cube depending on whether to test the side or top. As the product datasheet indicates a final hardness of 75A +/- 3A, one could argue that it’s kind-of in spec, but it mostly raises questions on how parameters like temperature and extrusion speed affect the final result.

2025 One Hertz Challenge: STM32 Blinks In Under 50 Bytes

August 16, 2025 by Lewin Day 9 Comments

Many of us have run a Blink program on a microcontroller before. It’s effectively the “Hello, World!” of the embedded space. However, few of us have ever thought about optimizing our Blink code to be as miniscule as possible. But that’s precisely what [Rudra Lad] did for this entry into the 2025 One Hertz Challenge!

This example of Blink, delay_blinky_13, is built specifically for the STM32F4 Discovery microcontroller development board. [Rudra] notes the code is “highly optimized” and compiles down to a binary size of under 50 bytes. The code doesn’t even use RAM, and it aims to get the blink as close to 1 Hz as possible. Many optimizations were used to crunch it down as small as possible. For example, the standard startup code isn’t used, with the entire program instead written in the Reset_Handler to save space. Bit-band is also used to write to peripheral registers to blink the LED, since this uses less instructions than the typical methods. Meanwhile, with many tweaks to the delay counting routine, [Rudra] was eventually able to get the blink frequency to 1.00019 Hz, as measured on a logic analyzer. That’s pretty darn close!

While it’s rare that you have only 50 bytes of binary space to blink an LED, work like this is a great way to flex your coding muscles. Code is on Github for the curious, and if you’ve worked up your own impressive tiny binaries, don’t hesitate to let us know!