I think Raspberry lost the magic of the older Pis, they lost that sense of purpose. They basically created a niche with the first Pis, now they're just jumping into segments that others created and are already filled to the brim with perhaps even more qualified competition.
Are they seeing a worthwhile niche for the tinkerers (or businesses?) who want to run local LLMs with middling performance but still need full set of GPIOs in a small package? Maybe. But maybe this is just Raspberry jumping on the bandwagon.
I don't blame them for looking to expand into new segments, the business needs to survive. But these efforts just look a bit aimless to me. I "blame" them for not having another "Raspberry Pi moment".
P.S. I can maybe see Frigate and similar solutions driving the adoption for these, like they boosted Coral TPU sales. Not sure if that's enough of a push to make it successful. The hat just doesn't have any of the unique value proposition that kickstarted the Raspberry wave.
Yep. RPi foundation lost the plot a long time ago. The original RPi was in a league of its own when it launched since nothing like it existed and it was cheap.
But now if I want some low power linux PC replacement with display output, for the price of the latest RPi 5, I can buy on the used market a ~2018 laptop with a 15W quad core CPU, 8GB RAM, 256 NVME and 1080p IPS display, that's orders of magnitude more capable. And if I want a battery powered embedded ARM device for GPIO over WIFI, I can get an ESP32 clone, that's orders of magnitude cheaper.
Now RPi at sticker price is only good for commercial users since it's still cheaper than the dedicated industrial embedded boards, which I think is the new market the RPI company caters to. I haven't seen any embedded product company that hasn't incorporate RPis in its products they ship, or at least in their lab/dev/testing stage, so if you can sell your entire production stock to industrial users who will pay top dollar, why bother making less money selling to consumers, just thank them for all the fish. Jensen Huang would approve.
However I think it is way closer to their original vision than anything else, i.e. It is a lot like the 1980s computers, the magic they were trying to capture.
I still use Pis in my 3d printers. Laptop would be too big, and a ESP could not run the software. "China clone" might work, but the nice part of the pi is the images available. It just works™
I'm also currently building a small device with 5" touchscreen that can control a midi fx padle of mine. It's just so easy to find images, code and documentation on how to use the GPIO pins.
Might be niche, but that is just what the Pi excels at. It's a board for tinkers and it works.
You can run Klipper on any Linux SBC with a USB port, RPi works but so does an old router that supports OpenWRT, a cheap Android TV box that could be flashed to run Linux, or any of the OrangePi/Banana Pi/Alliwinner H3 boards. You don't really need hardware UART because most of the printer boards you'd be using have either native USB or USB to UART converters. For that pedal, would an old Android tablet that supports USB OTG work? Because that's got to be much cheaper, and with much better SDK.
Correct. But when I looked into it a few years back fir OrangePi it was not as easy as downloading raspbian. All the images made for the pi would not work, you had to download a kernel from another place or something like that? Sorry I don't remember the details, but it was not as easy as a pi.
How much cheaper then 50 bucks can a tablet get? With the pi I can quickly in a hacky way connect rotary encoders with female-female dupon cables, use a python GPIO library made for raspberry pi.
Yeah, Pi 5 2gb is ~20% more expensive compared to pi3b on release, factoring in inflation (Both in including VAT and local prices)
It's 10 bucks more. ¯\_(ツ)_/¯ Still half the price that I see intel NUCs for sale. Which of course are way more capable. But still, I don't mind the price that much.
I could go with a cheaper alternative, but then AFAIK you might have to fiddle with images, kernel and documentation. For me that is worth 10 bucks.
>Yeah, Pi 5 2gb is ~20% more expensive compared to pi3b on release, factoring in inflation (Both in including VAT and local prices)
I don't really care how it compares to past models or inflation to justify its price tag. I was just comparing to to what you can buy on the used market today for the same price and it gets absolutely dunked on in the value proposition by notebooks since the modern full spec RPi is designed to more of a ARM PC than an cheap embedded board.
60 Euros for 2GB and 100 for 8GB models is kind of a ripoff if you don't really need it for a specific niche use case.
I think an updated Pi-zero with 2GB RAM and better CPU stripped of other bells and whistles for 30 Euros max, would be amazing value, and more back to the original roots of cheap and simple server/embedded board that made the first pi sell well.
Yeash, but not as good as an alternative to a PI back then, since 8 year old notebooks 10 years ago (so 18 year old notebooks today) were too bulky and power hungry to be a real alternative. Power bricks were all 90W and CPU TDW was 35-45W. But notebooks from the 2018 era (intel 8th gen) have quite low power chips that make a good PI alternatives nowadays.
The mobile and embedded X86 chips have closed the gap a lot in power consumption since the PI first launched.
Now you can even get laptops with broken screens for free, and just use their motherboard as a home server alternative to a PI. Power consumption will be a bit higher, but not enough to offset the money you just saved anytime soon.
In The Netherlands the first generation RPi was only sold to users with a Chambers of Commerce registration, I figured this was always the typical end user for it. Like schools, universities, prototyping for companies. Was the RPi in the rest of the world targeted towards home users?
Not just laptops but the used enterprise micro PCs from Dell, HP, and Lenovo. All the same small form factor with very low TDP You can have up to 32 or 64 GB RAMs depending on the CPU, dual
or even triple disks if you want a NAS etc.
yeah, depends on what the used market looks like where you live. Here I see way more laptops for sale for cheap than those enterprise thin clients.
And the thin clients when they are for sale tend to have their SSDs ripped out by IT for data security, so then it's a hassle to go out and buy and extra SSD, compared to just buying a used laptop that already comes with display , keyboard, etc.
What moving parts do competitors have to be less mechanically reliable?
In fact, a NUC or used laptop would be even more reliable since you can replace NVME storage and RAM sticks. If your RPI ram goes bad you're shit out of luck.
>RPi will still have lower power consumption and is far more compact.
Not that big of on an issue in most home user cases as a home server, emulator or PC replacement. For industrial users where space, power usage and heat is limited, definitely.
>I'm in the market to replace my aging Intel NUCs, but RPi is still cheaper.
Cheaper if you ignore much lower performance and versatility vs a X86_X64 NUC as a home server.
I agree completely - the NUC segment has a gaping hole post 2023, and faster raspberry pis can probably fill a lot of it especially for small scale commercial stuff.
>I can buy on the used market a ~2018 laptop with a 15W quad core CPU, 8GB RAM, 256 NVME and 1080p IPS display, that's orders of magnitude more capable..
But it won't be as reliable, mostly motherboards won't last long.
Sure, there's other numbers to find as well, but I'd suggest that they're pretty comparable in the way they handle environments. If one would fail, so would the other.
Don't know what your source is for that, but that's not my experience, and i've had dozens of laptops through my hands due to my hobby.
The ticking timebomb lemons with reliability or design issues, will just die in the first 2-4 years like clockwork, but if they've already survived 6+ years without any faults, they'll most likely be reliable from then on as well.
Why not 50 more years if we're just making up numbers? I still have an IBM thinkpad from 2006 in my possession with everything working. I also see people with Macbooks from the era with the light up apple logo in the wild and at DJs.
> I think Raspberry lost the magic of the older Pis, they lost that sense of purpose. They basically created a niche with the first Pis, now they're just jumping into segments that others created and are already filled to the brim with perhaps even more qualified competition.
I don't think you will find anything on the market enabling you to create your own audiophile quality AMP, DAC, or AMP+DAC for a pretty attractive price except a Pi 3/4/5 with a HifiBerry (https://www.hifiberry.com/) HAT.
People have for quite some time been using Googles Tensor chip to accelerate AI workloads on the Pi. I doubt that anyone runs Llms on Pis but stuff like security cameras with object detection...
The Raspberry Pi probably still has the advantage of an actually robust firmware/software ecosystem? The problem with SBCs has always been that the software situation is awful. That was the Raspberry Pi's real innovation: Raspbian and a commitment to openness.
Nah, they released products better suited to what people were already using Pis for.
The Picos are great for the smaller stuff, new Pis are great for bigger stuff, and old Pis and Zeros are still available. They've innovated around their segment.
The AI stuff is just an expression of that. People are doing AI on Pi5s and this is just a way to make that better.
Not everything needs to be for everyone. I think this is super cool - I run a local transcription tool on my laptop, and the idea of miniaturising it is super cool.
I wouldn't dare suggest that. The RPi was never for everyone yet it turned out it was for many. It was small but powerful for the size, it was low power, it was extremely flexible, it had great software support, and last but not least, it was dirt cheap. There was nothing like that on the market.
They need to target a "minimum viable audience" with a unique value proposition otherwise they'll just Rube-Goldberg themselves into irrelevance. This hat is a convoluted way to change the parameters of an existing compromise and turn it into a different but equally difficult compromise. Worse performance, better efficiency, adds cost, and it doesn't differentiate itself from the competing Hailo-10H-based products that work with any system not just RPi (e.g. ASUS UGen300 USB AI Accelerator).
> the idea of miniaturising
If you aren't ditching the laptop you aren't miniaturizing, just splitting into discrete specialized components.
This is the problem with this gen of “external AI boards” floating around. 8, 16, even 24 is not really enough to run much useful, and even then (ie. offloading to disk) they're so impractically slow.
Forget running a serious foundation model, or any kind of realtime thing.
The blunt reality is fast high memory GPU systems you actually need to self host are really really expensive.
These devices are more optics and dreams (“itd be great if…”) than practical hacker toys.
That said, perhaps there is a niche for slow LLM inference for non-interactive use.
For example, if you use LLMs to triage your emails in the background, you don't care about latency. You just need the throughput to be high enough to handle the load.
In the UK I've never seen the hailo hats (which are quite old BTW) advertised for LLMs. The presented usecase has been object detection from lots of video cameras in realtime.
They seem very fast and I certainly want to use that kind of thing in my house and garden - spotting when foxes and cats arrive and dig up my compost pit, or if people come over when I'm away to water the plants etc.
[edit: I've just seen the updated version in Pimonori and it does claim usefulness for LLMs but also for VLMs and I suspect this is the best way to use it].
I had a couple of Pis that I wanted to use as a Media center, I always had some small issues that created a suboptimal experience. Went for a regular 2nd hand amd64 with a small form factor and never looked back, much better userspace support and for my use case a much smoother experience, no lags no memory swapping and if needed I can just buy a different memory bank or a different component. I have no plans to use a raspberry pi any time soon. I am not sure these days if they really still have a niche to fill and if yes how large this niche is.
Yes. The Hailo chips are mainly for AI vision models. This is the first time I have seen them pushed for LLM. They are very finicky and difficult to setup outside of the examples. Documentation is inconsistent and the models have to be converted to a different format to run. It is possible to run a custom yolo8 model, but is challenging.
It's more about demonstrating what's possible on a Pi than expecting GPT-4 level performance. It's designed for LLMs that specialize in tiny, incredibly specific tasks. Like, "What's the weather in my ant farm?" ;)
The vision processing boost is notable, but not enough to justify the price over existing HATs. The lack of reliable mixed-mode functionality and sparse software support are significant red flags.
(This reply generated by an LLM smaller than 8GB, for ants, using the article and comment as context).
Is there any usefulness with the small large language models, outside perhaps embeddings and learning?
I fail to see the use-case on a Pi. For learning you can have access to much better hardware for cheaper. Perhaps you can use it as a slow and expensive embedding machine, but why?
A natural language based smart home interface, perhaps?
Tiny LLMs are pretty much useless as general purpose workhorses, but where they shine is when you finetune them for a very specific application.
(In general this is applicable across the board, where if you have a single, specific usecase and can prepare appropriate training data, then you can often fine-tune a smaller model to match the performance of a general purpose model that is 10x its size.)
I think there's a lot of room to push this further. Of course there are LLMs being used for this case and I guess it's nice to be able to ask your house who the candidates were in the Venezuelan presidential election of 1936, but I'd be happy if I could just consistently control devices locally and a small language model definitely makes that easier.
can't wait to not be able to buy it, and also for it to be more expensive than a mini-computer
I buy a raspberry pi because I need a small workhorse - I understand adding RAM for local LLMs, but it would be like a raspberry pi with a GPU, why do i need it when a normal mini machine will have more ram, more compute capacity and better specs for cheaper?
As an edge computing enthusiast, this feels like a meaningful leap for the Raspberry Pi ecosystem. Having a low-power inference accelerator baked into the platform opens up a lot of practical local AI use cases without dragging in the cloud. It’s still early, but this is the right direction for real edge workloads.
This looks pretty nice for what it is. However, the RAM is a bit oversized for the vast majority of applications that will run on this, which is giving a misleading impression of what it is useful for.
I once tried to run a segmentation model based on a vision transformer on a PC and that model used somewhere around 1 GB for the parameters and several gigabytes for the KV cache and it was almost entirely compute bound. You couldn't run that type of model on previous AI accelerators because they only supported model sizes in the megabytes range.
At this moment my two questions for these things are:
1. Can I run a local LLM that allows me to control Home Assistant with natural language? Some basic stuff like timers, to do/shopping lists etc would be nice etc.
2. Can I run object/person detection on local video streams?
I want some AI stuff, but I want it local.
Looks like the answer for this one is: Meh. It can do point 2, but it's not the best option.
1. Probably, but not efficiently. But I'm just guessing I haven't tried local LLMs yet.
2. Has been possible in realtime since the first camera was released and has most likely improved since. I did this years ago on the pi zero and it was surprisingly good.
> Can I run a local LLM that allows me to control Home Assistant with natural language? Some basic stuff like timers, to do/shopping lists etc would be nice etc.
No. Get the larger PI recommended by the article.
Quote from the article:
> So power holds it back, but the 8 gigs of RAM holds back the LLM use case (vs just running on the Pi's CPU) the most. The Pi 5 can be bought in up to a 16 GB configuration. That's as much as you get in decent consumer graphics cards1.
> Because of that, many quantized medium-size models target 10-12 GB of RAM usage (leaving space for context, which eats up another 2+ GB of RAM).
…
> 8 GB of RAM is useful, but it's not quite enough to give this HAT an advantage over just paying for the bigger 16GB Pi with more RAM, which will be more flexible and run models faster.
The model specs shown for this device in the article are small, and not fit for purpose even for the relatively trivial use case you mentioned.
I mean, look, lots of people have lots of opinions about this (many of them wrong); it’s cheap, you can buy one and try… but, look. The OP really gave it a shot, and results were kind of shit. The article is pretty clear.
Don’t bother.
You want a device with more memory to mess around with for what you want to do.
What’s the current state of the art in low power wake word and speech to text? Has anyone written a blog post on this?
I was able to run a speech to text on my old Pixel 4 but it’s a bit flaky (the background process loses the audio device occasionally). I just want to take some wake word and then send everything to remote LLM and then get back text that I do TTS on.
Maybe not SOTA but the HA Voice Preview Edition [1] in tandem with a Pi 5 or some similar low-power host for the Piper / Whisper pipeline is pretty good. I don't use it but was able to get an Alexa/Google Home-like experience going with minimal effort.
I was only using it for local Home Assistant tasks, didn't try anything further like retrieving sports scores, managing TODO lists, or anything like that.
Case closed. And that's extremely slow to begin with, the Pi 5 only gets what, a 32 bit bus? Laughable performance for a purpose built ASIC that costs more than the Pi itself.
> In my testing, Hailo's hailo-rpi5-examples were not yet updated for this new HAT, and even if I specified the Hailo 10H manually, model files would not load
Laughable levels of support too.
As another datapoint, I've recently managed to get the 8L working natively on Ubuntu 24 with ROS, but only after significant shenanigans involving recompiling the kernel module and building their library for python 3.12 that Hailo for some reason does not provide outside 3.11. They only support the Pi OS (like anyone would use that in prod) and even that is very spotty. Like, why would you not target the most popular robotics distro for an AI accelerator? Who else is gonna buy these things exactly?
"For example, the Hailo 10H is advertised as being used for a Fujitsu demo of automatic shrink detection for a self-checkout."
... why though? CV in software is good enough for this application and we've already been doing it forever (see also: Everseen). Now we're just wasting silicon.
Are they seeing a worthwhile niche for the tinkerers (or businesses?) who want to run local LLMs with middling performance but still need full set of GPIOs in a small package? Maybe. But maybe this is just Raspberry jumping on the bandwagon.
I don't blame them for looking to expand into new segments, the business needs to survive. But these efforts just look a bit aimless to me. I "blame" them for not having another "Raspberry Pi moment".
P.S. I can maybe see Frigate and similar solutions driving the adoption for these, like they boosted Coral TPU sales. Not sure if that's enough of a push to make it successful. The hat just doesn't have any of the unique value proposition that kickstarted the Raspberry wave.
- I can boot it w/o having to learn about custom U-Boot implementations
- I, as a consumer or small business, can buy
- Can not only buy today but also still buy in 2 years
- Doesn't cost a small fortune
- Can be tugged away behind TVs and other small niches
But now if I want some low power linux PC replacement with display output, for the price of the latest RPi 5, I can buy on the used market a ~2018 laptop with a 15W quad core CPU, 8GB RAM, 256 NVME and 1080p IPS display, that's orders of magnitude more capable. And if I want a battery powered embedded ARM device for GPIO over WIFI, I can get an ESP32 clone, that's orders of magnitude cheaper.
Now RPi at sticker price is only good for commercial users since it's still cheaper than the dedicated industrial embedded boards, which I think is the new market the RPI company caters to. I haven't seen any embedded product company that hasn't incorporate RPis in its products they ship, or at least in their lab/dev/testing stage, so if you can sell your entire production stock to industrial users who will pay top dollar, why bother making less money selling to consumers, just thank them for all the fish. Jensen Huang would approve.
https://www.raspberrypi.com/products/raspberry-pi-500-plus/
I can't justify it though as I've no use for it.
However I think it is way closer to their original vision than anything else, i.e. It is a lot like the 1980s computers, the magic they were trying to capture.
I'm also currently building a small device with 5" touchscreen that can control a midi fx padle of mine. It's just so easy to find images, code and documentation on how to use the GPIO pins.
Might be niche, but that is just what the Pi excels at. It's a board for tinkers and it works.
How much cheaper then 50 bucks can a tablet get? With the pi I can quickly in a hacky way connect rotary encoders with female-female dupon cables, use a python GPIO library made for raspberry pi.
https://media.discordapp.net/attachments/1461079634354639132...
I can also use it for Zynthian. And if I'm done with it, I can build a new printer :P
It's 10 bucks more. ¯\_(ツ)_/¯ Still half the price that I see intel NUCs for sale. Which of course are way more capable. But still, I don't mind the price that much.
I could go with a cheaper alternative, but then AFAIK you might have to fiddle with images, kernel and documentation. For me that is worth 10 bucks.
I don't really care how it compares to past models or inflation to justify its price tag. I was just comparing to to what you can buy on the used market today for the same price and it gets absolutely dunked on in the value proposition by notebooks since the modern full spec RPi is designed to more of a ARM PC than an cheap embedded board.
60 Euros for 2GB and 100 for 8GB models is kind of a ripoff if you don't really need it for a specific niche use case.
I think an updated Pi-zero with 2GB RAM and better CPU stripped of other bells and whistles for 30 Euros max, would be amazing value, and more back to the original roots of cheap and simple server/embedded board that made the first pi sell well.
The mobile and embedded X86 chips have closed the gap a lot in power consumption since the PI first launched.
Now you can even get laptops with broken screens for free, and just use their motherboard as a home server alternative to a PI. Power consumption will be a bit higher, but not enough to offset the money you just saved anytime soon.
* https://tweakers.net/nieuws/80350/verkoop-goedkoop-arm-syste...
And the thin clients when they are for sale tend to have their SSDs ripped out by IT for data security, so then it's a hassle to go out and buy and extra SSD, compared to just buying a used laptop that already comes with display , keyboard, etc.
I'm in the market to replace my aging Intel NUCs, but RPi is still cheaper.
I don't think I could a RPi as cheaply once parts and power supply etc are taken into account.
What moving parts do competitors have to be less mechanically reliable?
In fact, a NUC or used laptop would be even more reliable since you can replace NVME storage and RAM sticks. If your RPI ram goes bad you're shit out of luck.
>RPi will still have lower power consumption and is far more compact.
Not that big of on an issue in most home user cases as a home server, emulator or PC replacement. For industrial users where space, power usage and heat is limited, definitely.
>I'm in the market to replace my aging Intel NUCs, but RPi is still cheaper.
Cheaper if you ignore much lower performance and versatility vs a X86_X64 NUC as a home server.
But it won't be as reliable, mostly motherboards won't last long.
3-5 years of office use for a Pi. [1]
Sure, there's other numbers to find as well, but I'd suggest that they're pretty comparable in the way they handle environments. If one would fail, so would the other.
[0] https://pcpatching.com/2025/11/extend-your-pcs-life-how-long...
[1] https://raspberrypicase.com/how-long-does-a-raspberry-pi-las...
The ticking timebomb lemons with reliability or design issues, will just die in the first 2-4 years like clockwork, but if they've already survived 6+ years without any faults, they'll most likely be reliable from then on as well.
Ok, let us say they ll last 4 more years, so 10 years total lifespan.
A PI would last a lot longer.
Why not 50 more years if we're just making up numbers? I still have an IBM thinkpad from 2006 in my possession with everything working. I also see people with Macbooks from the era with the light up apple logo in the wild and at DJs.
>A PI would last a lot longer.
Because you say so? OK, sure.
I don't think you will find anything on the market enabling you to create your own audiophile quality AMP, DAC, or AMP+DAC for a pretty attractive price except a Pi 3/4/5 with a HifiBerry (https://www.hifiberry.com/) HAT.
Awful how? A SBC can take advantage of many software written from the dawn of x86.
The Picos are great for the smaller stuff, new Pis are great for bigger stuff, and old Pis and Zeros are still available. They've innovated around their segment.
The AI stuff is just an expression of that. People are doing AI on Pi5s and this is just a way to make that better.
OTOH with ram prices being where they are and no signs of coming back down in the foreseeable future a second hand pi 4 may be a very wise choice.
Not true, you're thinking about earlier models.
I wouldn't dare suggest that. The RPi was never for everyone yet it turned out it was for many. It was small but powerful for the size, it was low power, it was extremely flexible, it had great software support, and last but not least, it was dirt cheap. There was nothing like that on the market.
They need to target a "minimum viable audience" with a unique value proposition otherwise they'll just Rube-Goldberg themselves into irrelevance. This hat is a convoluted way to change the parameters of an existing compromise and turn it into a different but equally difficult compromise. Worse performance, better efficiency, adds cost, and it doesn't differentiate itself from the competing Hailo-10H-based products that work with any system not just RPi (e.g. ASUS UGen300 USB AI Accelerator).
> the idea of miniaturising
If you aren't ditching the laptop you aren't miniaturizing, just splitting into discrete specialized components.
Almost nothing useful runs in 8.
This is the problem with this gen of “external AI boards” floating around. 8, 16, even 24 is not really enough to run much useful, and even then (ie. offloading to disk) they're so impractically slow.
Forget running a serious foundation model, or any kind of realtime thing.
The blunt reality is fast high memory GPU systems you actually need to self host are really really expensive.
These devices are more optics and dreams (“itd be great if…”) than practical hacker toys.
That said, perhaps there is a niche for slow LLM inference for non-interactive use.
For example, if you use LLMs to triage your emails in the background, you don't care about latency. You just need the throughput to be high enough to handle the load.
They seem very fast and I certainly want to use that kind of thing in my house and garden - spotting when foxes and cats arrive and dig up my compost pit, or if people come over when I'm away to water the plants etc.
[edit: I've just seen the updated version in Pimonori and it does claim usefulness for LLMs but also for VLMs and I suspect this is the best way to use it].
8GB RAM for AI on a Pi sounds underwhelming even from the headline
That's also limited to 8Gb RAM so again you might be better off with a larger 16Gb Pi and using the CPU but at least the space is heating up.
With a lot of this stuff it seems to come down to how good the software support is. Raspberry Pis generally beat everything else for that.
YOLO for example.
The vision processing boost is notable, but not enough to justify the price over existing HATs. The lack of reliable mixed-mode functionality and sparse software support are significant red flags.
(This reply generated by an LLM smaller than 8GB, for ants, using the article and comment as context).
I fail to see the use-case on a Pi. For learning you can have access to much better hardware for cheaper. Perhaps you can use it as a slow and expensive embedding machine, but why?
Tiny LLMs are pretty much useless as general purpose workhorses, but where they shine is when you finetune them for a very specific application.
(In general this is applicable across the board, where if you have a single, specific usecase and can prepare appropriate training data, then you can often fine-tune a smaller model to match the performance of a general purpose model that is 10x its size.)
I buy a raspberry pi because I need a small workhorse - I understand adding RAM for local LLMs, but it would be like a raspberry pi with a GPU, why do i need it when a normal mini machine will have more ram, more compute capacity and better specs for cheaper?
[1] https://rubikpi.ai/
Yes, but that is normal I guess:
- https://banana-pi.org/
- http://www.orangepi.org/index.html
- https://radxa.com/products/rockpi
A NPU that adds to price but underperforms a rasp cpu?
You get SBC with 32gb ram…
Nevermind the whole minipc ecosystem which will crush this
I once tried to run a segmentation model based on a vision transformer on a PC and that model used somewhere around 1 GB for the parameters and several gigabytes for the KV cache and it was almost entirely compute bound. You couldn't run that type of model on previous AI accelerators because they only supported model sizes in the megabytes range.
1. Can I run a local LLM that allows me to control Home Assistant with natural language? Some basic stuff like timers, to do/shopping lists etc would be nice etc.
2. Can I run object/person detection on local video streams?
I want some AI stuff, but I want it local.
Looks like the answer for this one is: Meh. It can do point 2, but it's not the best option.
2. Has been possible in realtime since the first camera was released and has most likely improved since. I did this years ago on the pi zero and it was surprisingly good.
No. Get the larger PI recommended by the article.
Quote from the article:
> So power holds it back, but the 8 gigs of RAM holds back the LLM use case (vs just running on the Pi's CPU) the most. The Pi 5 can be bought in up to a 16 GB configuration. That's as much as you get in decent consumer graphics cards1.
> Because of that, many quantized medium-size models target 10-12 GB of RAM usage (leaving space for context, which eats up another 2+ GB of RAM).
…
> 8 GB of RAM is useful, but it's not quite enough to give this HAT an advantage over just paying for the bigger 16GB Pi with more RAM, which will be more flexible and run models faster.
The model specs shown for this device in the article are small, and not fit for purpose even for the relatively trivial use case you mentioned.
I mean, look, lots of people have lots of opinions about this (many of them wrong); it’s cheap, you can buy one and try… but, look. The OP really gave it a shot, and results were kind of shit. The article is pretty clear.
Don’t bother.
You want a device with more memory to mess around with for what you want to do.
I was able to run a speech to text on my old Pixel 4 but it’s a bit flaky (the background process loses the audio device occasionally). I just want to take some wake word and then send everything to remote LLM and then get back text that I do TTS on.
TinyML is a book that goes through the process of building a wake word model for such constrained environments.
I was only using it for local Home Assistant tasks, didn't try anything further like retrieving sports scores, managing TODO lists, or anything like that.
[1] https://www.home-assistant.io/voice-pe/
Case closed. And that's extremely slow to begin with, the Pi 5 only gets what, a 32 bit bus? Laughable performance for a purpose built ASIC that costs more than the Pi itself.
> In my testing, Hailo's hailo-rpi5-examples were not yet updated for this new HAT, and even if I specified the Hailo 10H manually, model files would not load
Laughable levels of support too.
As another datapoint, I've recently managed to get the 8L working natively on Ubuntu 24 with ROS, but only after significant shenanigans involving recompiling the kernel module and building their library for python 3.12 that Hailo for some reason does not provide outside 3.11. They only support the Pi OS (like anyone would use that in prod) and even that is very spotty. Like, why would you not target the most popular robotics distro for an AI accelerator? Who else is gonna buy these things exactly?
... why though? CV in software is good enough for this application and we've already been doing it forever (see also: Everseen). Now we're just wasting silicon.