Re: Losing the plot
> goes and over-complicates it.
How do you mean? The article is about an *optional* daughter board.
Raspberry Pi has created a machine-learning addition for its single board computer that features the Hailo-8L AI accelerator. This AI Kit is an M.2 Hat board with a preinstalled M.2 stick that carries the Hailo-8L accelerator, marking the first time Raspberry Pi has ever sold a first-party accessory for AI. The Hailo-8L is a …
This post has been deleted by its author
I wouldn't say they were over complicating things. It just adds some more versatility to the Pi5. This AI kit is of no particular interest to me, so I won't be buying it, but it's great to have more optional functionality. I love that you can customise things to suit whatever project you want to get involved in.
More that that, I was actually kind of word that the Pi Foundation were missing the very real possibility that homebrew AI was going to be a new part of the "makerspace" that Raspberry Pi shoots for. The first Pi was leading the curve for makers by putting a computer at a pricepoint where it was competitive with microcontrollers like Arduino on small scale, but not on large scale. The Pi was a useful steeping stone for embedded, because it was easy to use standard tools and just throw extra processing at it to avoid the need to optimise firmware early on.
Now that things are becoming clearer, I can see that AI is still going to need a PC... for training. I think the Foundation are really onto something here, because no-one uses the Pi as their only computer. This means we're going to get geeks using their NVidia megabeast video card for model training, and instead of only being able to use the models on that PC, they'll able to run it on a much smaller thing.
I think the Foundation are really onto something here
Nothing to do with the charitable Foundation as such. It's Raspberry Pi Limited, soon to be Raspberry Pi Holdings PLC.
I am less inclined to believe this is about facilitating makers than it is about Raspberry Pi's desire to sell even more product into the industrial and commercial markets to boost their profits and maximise dividends for shareholders.
You may say why does this always have to happen, but I'd say, why do people never read the article and understand it before commenting.
As above, the Pi 0,1,2,3 and 4 are still available at mostly original prices. All still useful, all still simple and all very under-complicated.
I’ve been following (and buying) Pi since the very first model - and, if I’m honest, I probably have too many of them now. But they’re so small that there’s always room for one more. Now though, I’m seriously beginning to wonder whether I could daily a Pi - not just file server tasks, and the odd bit of retro gaming, but actually use one as my main computer? I think that the Pi 5 is now beefy enough, and with this doodad (and software written to take advantage of it) I’d have though it’s more than enough for most tasks.
Well, with even MS Office now available as a web app, the days of the dedicated PC are numbered. The Hailo kit pushes this further, because if local beef needs to go AI, then users don't really need much at all.
However, the Pi will never be good for training AIs, so developers are going to need PCs still....
> the Pi will never be good for training AIs
Correction: the Pi will never be good at training LLMs.
There are many, many other forms and uses of Neural Nets and other variants of Machine Learning, in all shapes and sizes - from a few matchboxes upwards. Much useful and practical work has been done using ML, especially in Machine Vision, using (what you would probably consider ludicrously bad quality) low resolution analogue cameras in the 70s and 80s - when the mini computers the devs could get their hands on were dwarfed by the processing power in a R'Pi Zero.
Look around and you should be able to find all sorts of data feeds that be used to train a model using just an R'Pi, let alone one with this accelerator connected.
Fair point, but then again, there's still a matter of training vs running. Training an AI is a lot more work than running one, and anything you could train on a Pi is not going to need the full power of the Pi, and that's whether or not you have this card in your Pi.
We're back to the days where the devs need a far more powerful computer than the users. To get a model that uses the Hailo to its fullest, you'll need a more powerful machine to train the model, probably with a NVidia 40 series card in it.
My 5 has the standard case in black and it whinges every few seconds, even when doing nothing.
I sometimes give it a shake or press down on the lid, which works from time to time.
Anyone recommend a different case?
Otherwise I shall have to leave Radio 2 on all the time.
I put a 4GB Pi5 in a standard black case in our retirement village library for basic internet and office work. It tended to run warm and was a bit noisy. Removing the standard fan and replacing it with the Active Cooler improved things considerably - The fan comes on less often, and when it does I can't hear it. If it is still too hot drill a couple of holes in the lid. Two somewhat more expensive solutions that I have for personal use are an Argon NEO 5 black case with built-in fan on a headless Pi for testing and development; and a similar Argon NEO 5 M.2 NVME PCIE case with an internal 2TB Kingston SSD on another headless Pi that I use as a file server and recorder for live TV.
You can also try oiling the fan bearing. Lift the label off the center of the fan, carefully. Add a tiny drop of machine oil to the fan bearing. Use a sewing pin or needle. And (if you can) stick the label back down. This seems to extend the life of the fan bearings somewhat, but eventually, you'll need to replace it with a new fan.
This works with the fans on my Pi 4. I'm not sure if it will work with Pi5 fans.
I wonder if calls to boycott Hailo, an Israeli company founded by ex-members of the IDF's elite technology unit, will now extend to Raspberry Pi?
That could piss on their fireworks when it comes to their imminent IPO.
Let's hope they handle things better than their "we hired a cop, fuck yeah" fiasco.