Hey everyone just one other note; I have a second channel where I test cookware, appliances, and cooking techniques: youtube.com/@platima. I'd love for you to check it out if it interests you (it's bit of a saturated TH-cam demographic it seems). Cheers 💚💚
@@sgodsellifyYep the user level is on the A7. See github.com/LuckfoxTECH/luckfox-pico/issues/99, github.com/luyi1888/rv1106-mcu and github.com/syntacore/scr1 for info on what you can do with it.
Nice to see you covering these as was holding off grabbing some as unsure if support was available so will happily visit the store once stock arrives. I see Milk-V have started getting software updated since your Mars video so would love to see a follow up video as some users are mentioning its improved since launch!
Yeah man I am so torn between these and the Duo now. The Duo is more capable though I think... I will see. And yep I have a follow-up plans, as there is so much more in the docs now with using the 2nd RTOS core, making use of the TPU, etc!
bro thought i was gonna be the only one with these, lol , i love em, i had bought heltec sufficient IOT back in feb and basically just has to figure things out with it cuz no one had vids , im glad you made this , thanks . .
For everyone considering this board: The Camera and Deep-Learning Models currently don't have Python support. Everything up until running the demo works great but sadly there's still a lot of functionality that requires significant tinkering in C++. There is Python support for almost outputs such as PWM and SPI, they have a pretty good wiki page, but there's still some work to be done before the software is fully ready.
Thanks for the great video. I have the 1103 board but not using it so far. Instead, using 3566 and 3588. Will try to do the development on 3566/3588 with NVME and then port to 1103/1106 for power/size/cost consideration.
I’ve bought one of these but it was my first experience using a headless sbc. I got it to boot but felt it was useless since I couldn’t download the os, though it seems they’ve now made it publicly available! I’ll have to give it another go
Luckfox Pico is an excellent product with such small form factor. I've compiled the Ubuntu OS with tun/tap interface with Luckfox SDK, install Openvpn client, then configured it to connect my Openvpn Server. I put it to my customer's office. I no longer need to go to the customer's office for server checking.
Hey mate that is great to hear! They certainly are bloody small but handy! Coming from a cyber-security standpoint I have some issues with that setup, but there are of course protection you can put in place such as IP-restricted jump boxes with push MFA, pubkey authentication per-session, etc
Those RISC-V board are fire! Can you do a handheld device based on one of those? Its much more compact than the Pi Zero (which has obviously more software support as its ARM based but ARM faced the same lack of support in the past). I am sure a such project would attract tons of views.
Nice job on the review! I'd love to see a demo that exercises the video camera option, including taking pictures and recording video. Do they have software support for recording video? If so, mjpeg or H.264?
Hey thanks yeah I am planning a 'part 2' that gets into more depth. I'd assume you use libcamera or even ffmpeg as a frontend for interfacing with it, and can save in any format you want. The ISP appears to support JPEG, H264 and H265 hardware-acceleration. You can see in th-cam.com/video/JIFqV1vZB2Q/w-d-xo.html how I use a library in a few ways to grab media from a connected camera - I expect the same code would likely work here with minimal changes.
@@PlatimaTinkers it would be huge if it integrates cleanly with ffmpeg. A good tell is if it provides a standard /dev/video0 interface. Looking forward to part 2!
@@leadf00tlooking at wiki.luckfox.com/Luckfox-Pico/CSI-Camera/(am going back to the workshop in a bit and can test if I can find a compatible camera) it looks like it creates /userdata/video0 haha. You can see on the Core3566 that it is fully V4L2 compatible though (wiki.luckfox.com/Core3566/Core3566-pinout/6.%20CSI%20Camera/#63-viewing-camera-information) so I'd assume so!
Hey finally one I am interested in. Always searching for SBCs with proper networking not this wireless stuff that scales like a jenga tower due to nonsense like the laws of physics.
Hey you're very welcome, and yeah no worries I just exported that small segment zoomed in if you want to have a closer look, but next time will set the terminal font much larger! th-cam.com/video/cx3rj6a_K2w/w-d-xo.html Cheers.
Loving the coverage of these small Linux boards that run Linux. I'd be really interested to know your recommendation for any which run using Power over Ethernet - I think being able to make sensors/IoT stuff that runs off PoE would be a killer app.
Hey glad to hear it mate! PoE is a very interesting one, your best bet would probably be using passive 24V POE, else the Duo IO board has some details about adding POE in the schematics for it! I might honestly design and sell a POE adapter for SBCs. Wouldn't be hard
This is A great video. I got my Luckfox Pico Plus to work and could SSH in to it and login over the network. DHCP worked right away. Worked pretty well actually! Now I am trying to use the Ubuntu image and boot from SD. I *kinda* got this working... but seems odd. Took for ever to get an IP address. But now it shows up... maybe an initial install/setup going on first time? Then to SSH login, it works but takes about 10-15 seconds to finally get to a prompt. This typical? I haven't tried another SD card yet. Need to find another one, just to make sure maybe a sketchy card. But over all it seems a bit, lethargic using the Ubuntu / SD setup ? doing A apt-get update took like 20 minutes or something to do.
Hey thanks for the feedback mate that is really awesome to hear. I've heard some people complain about this! I've personally not experienced it. First boot does always take a while though. What image were you using? I can test it out and ask their support team if need be.
@@PlatimaTinkers I was also imoressed about how easy was to set it up. I was able to recompile kernel and create my own .ko files while I never did that before. Its neat to learn how busybox is compiled on such devices.
@@PlatimaTinkers they provide VM with all tools needed. I bookmarked it somewhere I will send you. I think was in chinese but I used google translate. I am interested to see about esp_hosted as I dont know anything about it. Tool chain is ARM its a chance should not be hard?
Hi I am interested of that board (especially pico mini version). I've got two questions 1) is it possible to install full MQTT broker on that small board? 2) is it possible to solder 4 ethernet cables (100Mbit) directly to pico board? (I am interested of that because of lack of space for eth module). I know you say something about my second question - but you are speaking so fast and I'm not sure you say "can" or "can't" 4:27 regarding to soldering cables directly? Thank you in advance and sorry for my english.
Hey mate your English is fine! 1) You could probably install MQTT broker, I think it would probably run fine but I do not know how much CPU it would use. You would need to download the SDK and do a compile with it, which I can try if you want? 2) You could definitely solder the cables on, but it is a little risky and you may get a lot of TX/RX errors. I would at least recommend diodes on the TX/RX lines. This is pretty much all the ethernet module has. I say you can solder cables on, BUT you might blow it up haha. FYI I upload subtitles so hit 'c' and you can see English captions of exactly what I am saying. I know I talk too fast 🤣
@@PlatimaTinkers Thank you for the answer. I have ordered one board - the smallest pico mini version B (with flash mem)+ one ethernet board. I decided to use eth.board because of lack diodes you mentioned. It should be possible to solder both boards together to achieve a small "qube". Unfortunatelly i have no schematic of eth.board but I think it could possible to use "passive poe" by injecting +6V and gnd into not used pairs of 100Mbit ethernet cable(i will check connections and modify eth.board if necessary). On pico board there is step-down converter which accept absolute max +8V so after proper wiring it sould works in my opinion. Regarding to MQTT I'll do some tests (of course if I do not blow this crap up ;) ). I think CPU should be enough fast - 1Ghz is quite ok, but 64 megs of RAM can be a bottleneck. We will see.
@@grzesiekyogi1518 hey the ETH module is super simple; it just connects the TX+ TX- RX+ RX- pins with flyback and polarity diodes. You can see it well here on my shop page: shop.plati.ma/products/ethernet-board-for-milk-v-duo-and-luckfox-pico. You may be able to access the power pins on the bottom of the board where they are soldered. It's important to note that passive POE is 24V though, and up to 27V, so you'll need a step-down converter to bring it to 5V and connect it to VSYS. For this I have used shop.plati.ma/products/mini-360-dc-buck-converter before, as long as it's 24V passive, not 27V! You'll also want to tune that output voltage very carefully, then put some glue on the potentiometer.
@@PlatimaTinkersOn luckfox pico board there is already a dc-dc step-down converteter (EA3036C attached to VSYS) which accept absolute max +8V. So it sould be possible to inject +6V (I know - it is not according to passive POE standard, but it sould works without any additional converters and it is easy to inject by standard 2$ injector + standard power supply). If the LAN cable is not too long it should be possible even inject +5V (EA3036C converts to three voltages where 3.3V is the highest one - so I believe if there is for example 4.5V at VSYS due to cable resistance it should work anyway). I will test it soon.
@@grzesiekyogi hey yeah but I don't know where you are getting those voltages sorry - passive POE is 24-27V, and active POE is 34-57V. Eg, where are you getting your low 6V from? hah. The EA3036C looks like an awesome little unit though!
Oh my mate you're a genius! I had another look, and on www.luckfox.com/Luckfox-Pico/Luckfox-Pico-Max it has "MB", so correct, but www.luckfox.com/Luckfox-Pico/EN-Luckfox-Pico has "Mb" - meaning it is 64MB like you said. I cannot believe I missed that. Still, not great marketing in my mind haha.
for 1 i did not see the sd over the flash . I thought it was a covered bank of resistors. I like the thernet . the milkv should have done the same but it has a header spot.
Yeah I don't see why not. If you check out www.lontentech.com/products/raspberry-pi-vga-adapter-board-gpio-to-vga-module-shield-for-raspberry-pi-4-model-b-3b-2b-b that does 1080p60 using 6x GPIO pins as it only requires a ~150MHz pixel clock from memory.
Thanks mate! And yep, you use `upgrade_tool` for that. They have some instructions here: wiki.luckfox.com/Luckfox-Pico/Linux-MacOS-Burn-Image. It's very similar to the original process, but without the GUI!
@@mrlately7253 You're talking about Raspberry Pi with an RP2040 / ARM Cortex processor? Cause there's the Pico Mini, which has no wifi but is stamp sized. Or the LycheeRV Nano W, which has wifi and uses SG2002 so both ARM and RISC-V options :) shop.plati.ma/products/sipeed-lycheerv-nano-we-beta
@@PlatimaTinkers thanks for the support beyond the videos. I basically need to be able to do the same I can on a pico w but smaller. I only use a few pins. I'll check the links you shared. Thanks again
@@mrlately7253 All good! And yeah LycheeRV Nano WE would be good then! Else shop.plati.ma/products/pine64-pinenut-model-12s-wifi-ble5-stamp is tiny and cheap and has wifi. There's a few options out there. I also have some ESP32-C3 and S2 boards like littlebirdelectronics.com.au/products/esp32-c3-mini-development-board-based-on-esp32-c3fn4-single-core-processor-160mhz-running-frequency-2-4ghz-wi-fi-bluetooth-5 that work really well and can run Arduino, ESPHome, MicroPython or C! I use them for garage door controllers, etc. DFRobot make the 'Beetle' series which are the same. I have shop.plati.ma/products/esp32-cx-riser-board-v1-1-assembled-with-c3-and-led which is a prototype product but does work. Maybe not for your use case though! Really probably depends what IDF/SDK you want to use! C, Arduino, MicroPython being the big ones.
I tried compiling some programs for the plus, but you can't store any executables on the sd card because it has to be in FAT32 which has issues with file permissions ... that was a bummer
Yeah in Pt 2 I'll cover a bit more like that. Best to store within the image, eg copy it across with `adb push` after doing a cross compile, else you can probably slap another partition on the SD card as ExFAT... I'll have a tinker and let you know!
@@AboubakrAyeah no worries mate I've got one connected out at the workshop that I'm heading to shortly. Will try it out and let you know! You're also very welcome. Cheers
@@PlatimaTinkers BTW I noticed you're asking for the VM in another comment, I tried adding the link in the comment but it got deleted for some reason this is it, it has a VM containing the toolchain for cross-compilation, I've given it a try and it works perfectly drive.google.com/drive/folders/11Kc5Kogjjoqe5yWO0VI1-x4RMYv4m98L
I'm trying to send the link for the VM that has the cross-compilation toolchain in a comment but it gets deleted, I can send it on email if you're interested
We dont need no stinkin' FTDI, Silicon Labs CP2104 for the win! True story, at my work we use a cp2108 chip in our diagnostic board. Its the same thing as the 2104 but it has THREE serial endpoints.
Oooh that is pretty sweet - I could definitely see some value there, especially with some devices having separate serial interfaces for a) flashing then b) debug output.
@@PlatimaTinkers oh .. the cp2108 has at least four, endpoints numbered 0 through 3. I set up some udev rules on Linux to symlink the endpoints based on USB device attributes.
No POE version, but not hard to mod I think! Power usage is in the description. I think there is not enough RAM to run a service like that though? Not too sure!
so it's the size of a Pi Pico, or a bit bigger with a full on network jack, and runs Linux. So I assume you can run micropython or something from that Linux terminal to control the GPIO? I've only got basic familiarity with microcontrollers.
Yo yeah pretty close. It supports Python 3 as of just recently - comes with the image. MicroPython itself has been ported to ARM (github.com/micropython/micropython/tree/master/ports/bare-arm) and that MIGHT work, but would probably require a few tweaks. CircuitPython runs on the STM32F746 though, which is also Cortex-M7, so that could also work. Just not sure how the ethernet support would go with those binaries - maybe better to stick to the Linux base layer!
@@PlatimaTinkers That's neat! I'm mostly used to Arduinos that don't even have an OS of their own, so the idea of SSH'ing into one to use it is kinda new. Thanks for the info.
@@DFX2KXyeah I do like Arduinos, but I'm Linux at the core, so using SBCs like this that are _basically_ embedded systems but still have a shell makes me pretty damn happy! And you're very welcome mate. Have an awesome weekend
Glad to hear mate! Hope you ordered it from me 😋 (all good if not, #shamelessselfplug) Keep your eye out for more too, sounds like they've got some neat products in the works!
@@PlatimaTinkers ah do you ship to Sweden? Sorry I did not order from you, but I will say your prices are almost half of what they are here in Sweden for the Luckfox Pico Max (your shop says 165SEK, and I paid 300SEK here in Sweden) , so if you do, I can order from you next time :)
@@AndrewTSq Hey mate yep I do, and all good! The prices will probably go up a little soon as they are based on USD-AUD for my buying, but it definitely won't be that much!
Wooow, such a great video! Do you know how does the camera work? I'm trying to do object recognition but the only thing I know over the camera is that generates something like a .jpg file, does that file updates all the time? that is how does cameras works on linux? I'm a newbie in this world, thank you!!!
@ericcaudillobarbosa586 Hey thanks! The camera does not have to generate a JPG - it's just a data stream that you can interpret any way you want; save images, save videos, perform realtime detection, stream it to the network, etc. There's some example info on it here: wiki.luckfox.com/Luckfox-Pico/CSI-Camera and
Dude...this probably isn't going to be the ideal board for object recognition. object detection maybe. If object recognition is what you're interested in. Buy a raspberry pi 4 or 5 or pick up an Nvidia jetson. The jetson was made for AI applications. The most important thing to know is that you can't make chicken salad out of chicken shit, and these under 20 dollar boards are chicken shit when it comes to these type applications. If you're building an AI system or application and your board doesn't support python, you're going to hate it. peace out bro.
@rickhunt3183 I think you've missed a bit of the specs, as you're not correct all sorry mate. It'd be good for you to check out how the ISP pipeline and TPU work, plus how RKNN stitches it all together. Unlike the Pi 4, the board has 0.5 TOPs NPU and a 4MP 30fps ISP. It can do object detection AND recognition, and you can even do facial recognition at a decent clip. This RV1103 (lower end to 1106 and 1109) SoC can run YOLOV5 at about 10fps from memory, RetinaFace at a very good speed, plus plenty of other image recognition, text recognition, car plate recognition, speed recognition and similar models all at a solid pace. And it supports Python, comes with it out of the box in fact. Really not sure where you're getting any of your info from sorry.
what do you mean? i’ve been working on a Duo project but i’m stuck on SPI. it feels like reading/dev/spidevx.0 results in weird, unexpected outputs is this because of the microSD? 🫣
@@Onkoe I would assume so. I asked on the discuss forum and was told that the same signals run both ports. The schematic agrees. You get one or the other.
Nice video as always. Thank you for the video. This board is similar to the Milk-V Duo, eh? I personally like boards with WiFi built-in for convenience.
Hey thank you kindly, and yeah similar to Milk-V Duo and Ox64 in a number of ways! I prefer wifi too hah, but you could add USB or SPI wifi to this, else LoRa I'd say.
It's u-boot, so OS default from SD card or onboard SPI flash, but you can boot from ethernet or USB if you wanted! The default image is Debian based, the same as Ubuntu is, but you could easily roll-your-own using `debootstrap` / `cdebootstrap` if you specifically wanted the Ubuntu flavour. Just note that only Debian has significant RISC-V support so far, so best sticking to that ecosystem and the vendor apt repos.
I think you asked on Reddit? :P And yeah, USB-C addons work fine as long as the Linux kernel has the driver, and it could probably run Alpine but it'd take a lot of kerjiggering I think!
Please consider zooming in on the text .. very difficult to read .. really don't need to show the whole thing top to bottom and w/ all of the empty space on the right side. ☮
Yeah mate no worries - what part were you referring to? It can really depend on the resolution of your screen but I try to make it readable at 1080! Cheers for the feedback
@@PlatimaTinkers 7:17 onward .. for example (what I'm talking about can be avoided several times). Thanks for being kind about what I hope is constructive criticism. ☮ Great video .. don't take me wrong. Thanks for sharing your knowledge and experience!
@@THE-X-Forcehey yeah I get what you mean now - the terminal output! I just had a look and it's quite readable here, you might want to click the cog icon for the settings and make sure it's set to 1080p50 - that was the recording resolution, so that playback fullscreen should look close to 'native'. I have increased the PuTTy font size though so it's easier in future videos! Any criticism is absolutely appreciated - I am rubbish at this so value any input or feedback. Cheers!
@@PlatimaTinkers Ok .. my res is good and I _can_ make it out (barely on this 17" laptop at 1080), but maybe you will agree that having nearly half the screen show nothing but empty black space is .. less than ideal? I'm suggesting cropping most of that dead space out .. and since that would make the video vertical (aka horrible), lopping off some of the top text that the viewer has already seen, and doesn't need to focus on. Hope that makes sense. Maybe try it and see what you think. In any event, it's no big deal. Thank you again, and you are definitely not rubbish at this. Best of luck from NY☮
@@THE-X-ForceYeah mate had not considered that at all, I'm on a 27" haha. You're absolutely right about the waste of screen real estate though; if I make the font size 20pt it looks like it fills the width nearly perfectly. TH-cam's compression should massacre it less than too! Good luck with the freezing temps there, we've got 86-104 here!
Yea mate absolutely; the built in ISP, reserved RAM, and 0.5 TOPS NPU is designed exactly for that! See wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-RKNN-Test for more info!
Hey they have no wifi built in, but they do support USB device mode so you could easily connect USB or SDIO wifi as long as you have the drivers. They have MIPI CSI 2-lane for camera connectivity, so you should be able to do 4K@30FPS but the capability of the ISP is limited to about 4M so you'd only get QHD (2560x1440) unless you dropped the FPS further. Back of the napkin math would be 4K@10FPS to JUST fit within that hardware capability. The Pico Pro and Pico Max support 5M30FPS ISP so you should be able to get 15FPS which is what our eyes/brain usually need to perceive something as animated. Enjoy!
Hello, i tried to stream over usb. Its working. Is it possible to stream over wifi or something by connecting with esp32? If its possible please help me to stream wirelessly using esp32 as a wifi midule!
I saw a video that compared RISCV against ARM and the ARM core was consistently better so I think the ARM processors give you better performance. I also saw a video that complained a lot about the support for a lot of these Chinese SoC carriers and the No. 1 issue was making them work with established code libraries etc, so for me PI may be more expensive but life is short.
That's not really comparing Apples to Apples though, that's more saying you want two oranges but being given a lime and an orange. They're quite different in how they work, and RISC-V is quite new. Yes, if you want raw power, go ARM, but you miss a fair bit of the point then, but it's not always about power and these 1Ghz+ processors are a tad insane; that amount of power in an embedded device is just not necessary 99% of the time.
Uhhh, I guess, but I cannot fathom the use case there. It's not really designed to have a GUI, but you could probably enable swap then install X and VNC to it from another computer. It'll run like a 1995 Dell. If you want to run X or VNC on the LuckFox SBC you probably need to implement a VGA interface by bitbanging GPIOs.
@@InikoPro No worries. Completely depends on what you want to achieve though - do you need wifi or ethernet, do you need NPU or not, etc. They both compute, but both specialise in different functions.
Può essere dovuto a molte cose, ad esempio l'immagine della scheda SD, l'alimentazione, ecc. È meglio provare i forum su forums.luckfox.com/ o il supporto tecnico su wiki.luckfox.com/intro. Buona fortuna!
I wonder, since I never really see people use the tpu in such boards, yet on paper some of them should give compute power in speciffic AI related workloads close to around a quarter of the performance of a rtx4060(one of those new beagle bone boards, not cuda however), while using insane amounts lower power. so why doesn't seem anyone to be using those TPU's or C7x DSP's, fpga and such for AI tasks such as language models and image generation, just like the raspberry pi's they probably could be quite doable to cluster and might even be added on a single pcie card, I wonder what kind of performances could be reached that way, even though ram might be a issue, but for everything there are solutions. I mostly just wonder if there really isn't any development for AI to run on such hardware, since it would be much more efficient than how it is done now, and they already did make things to run it on integrated gpu's and old cpu's which technically seen would be a lot slower at such things in basically any way since they also don't benefit from the cuda cores or such. this has also been proven by intell adding a very small npu into their new cpu's, and using that NPU actually was much faster than the cpu and the gpu in those apu's, also noticably beating the amd apu's, despite that integrated NPU being a very small very low power part. to make it more serious, some of the SBC's from beagleboard actually have a quite a lot faster NPU in them than those intel APU's have, intels next gen was supposed to have faster NPU's, but those results where gotten by people using the old integrated NPU which actually was slower than the NPU in some of those new beagleboard computers. so honnestly it seems like big AI and such really miss a lot of opportunity by getting stuck in those big gpu's and such rather than trying to find ways to make it work properly on such small formfactor low power hardware.
Hey mate bit of a long comment and I'm slightly rushed sorry, but I think you're missing a fair bit of information there from what I can understand of what you've said. The terms are all right, but you're seemingly not using them in the right context. The TPU/NPUs in these might do 0.3-5 TOPs @ INT8, which is great for some basic machine learning, but absolutely nothing close to what an RTX4060 can do. An RTX card is hundreds of TOPS and also using a 16 bit floating point number as opposed to integer, so you can use less space for faster transformations or inferences with a greater degree of accuracy given the wider gamut of number values that can be represented. Those can do some complex DNN functionality like text to image generation, large language models, etc. Even further to that, RTX 3k and 4k series supports a newer type of floating point that is specifically designed to work better with AI (bfloat16). These SBCs I develop on and show off are small units that are better at ML for tasks such as facial recognition, and then embedding them in smart door bells, IP cameras, robots, gardens, etc. There's then also a bit of an issue with memory; to run a 7B parameter LLM on a GPU you're going to need a hefty card and then minimum 12GB of VRAM, and even then you'll only get 1-8 tokens/second. These things have 16-512MB of RAM, so you'd be lucky to get a 500,000M param LLM running, and besides it talking garbage, it would be extremely slow due to the IO bandwidth available, and that's if you can even quantise a fp16 model to run on int8, in which case you'd likely lose some of the LLM "magic" in that case. Long story short; different power level, different 'neural' technology, different use case, different market and price. That is just my understanding though, and I could very well be getting some details wrong there! That being said, I have 2x GTX 1070 Ti's, an RTX 2080, an RTX 3060 and an RTX 3080 that I run different LLM and diffusion models on, so I do have a fair bit of familiarity with the tech... Just still saving for an RTX 4090 one day 🤘 Thanks for your input either way!
@@PlatimaTinkers well if you are saving for a 4090 you might wanna wait for the rtx 5000 series which might launch not to far away are supposed to be much faster than the 4000 series. for the TPU's from most of them I didn't actually know their speed, However I speciffically based it on the beaglebone AI-64. the beaglebone AI-64 is suposed to reach quite high numbers and also has multiple different accelerators from which some are really fast. even though I indeed do not know all about such things. but I do know things could be much more efficient, but that would also largely depend on the AI itself being designed to be efficient instead of based around normal hardware. however lately I was thinking that since AI like stablediffusion uses essentially weighter random noice, that it might be accelerated and made more efficienty by using a second model at first which instead of training noice clouds would train more on fractal clouds, that way if that works right it might be able to skip many steps in the proces and only do a few steps at the end to make them some less fractalish.
@@ted_van_loon I don't think there is going to be huge jump from the 4K to 5K range from what I can see - eg minimal new data types, just a smaller lithography, extra pipelines and cores, faster memory, etc. I'd say once the 5K comes out, I'll grab a 4080 Ti or 4090 as the price will drop, then maybe wait for the 6K when they do something mental like 1nm lithography, GDDR7, etc. Re the BeagleBone AI-64, it's a pretty bloody powerful board - it'd definitely go a long way towards something like a high-end video editing system if utilised properly, but it is still not designed to run models like LLMs. It's got an 8TOPS TPU, high bandwidth ISP, a decent little GPU, and 4GB RAM (ignoring the DSPs). This is higher than the Fire specs we just discussed, but more designed for robotics and automotive. Eg it would be great at real-time processing of multiple video signals for things like depth mapping, object recognition, collision prediction, etc. You can find examples of the models at github.com/TexasInstruments/edgeai-modelzoo and get an idea for how the board may be used. If you wanted to compare it to a GPU for running diffusion models like SDXL or LLMs such as Mixtral 0.2, you might need a few hundred of them, and then you end up with something less power efficient and less compact plus the challenge of trying to get the data between them using very limited interconnects like 1Gbps LAN (AI infrastructure will often run 40Gbps - 400Gbps fibre interconnects). Large companies such as Mistral, OpenAI, Alphabet, Meta, etc, are all working on making these inference and diffusion models more efficient, however, the long and short of it is that how they work requires massive amounts of data to be multiplied at blazing speeds, which requires a lot of fast RAM, high bandwidth, and seriously huge parallel pipelines. CPUs are crap at this, TPUs are okay at this, GPUs are great at this, and AI accelerators do it excellently. I get where your idea is coming from, however, unfortunately it does not work anything like that, and the AI is not based around any hardware; the goals are typically based around creating the fastest, most accurate, and smallest models possible, and it really doesn't matter what hardware it runs on at that point; GPUs are just the easiest for consumers, where as for high-end inference you'll want H100's or similar. They are called "GPUs" but really, they aren't. They're not meant for playing games, you can't plug a display into them, they just multiply matrices amazingly efficiently. We all started using GPUs for this because rendering images and textures just happens to use a similar pipeline. I hope all that information helps you understand a bit more!
Yep everyone agrees haha. They kind of just make hardware, and leave the rest (software, doco, etc) to the community. Even their new/current website was made by someone voluntarily 😐 Pros and cons!
Being a hobbyist and just noodling with this stuff, this is like a "gross amount of power in a microcontroller" So when do they just start shipping these with PCIe interfaces and stop pretending these aren't computers? Wait... do any of these have a PCIe interface?
Hahahah yeah some of these SBC's / eSBC's support PCIe. Eg the Milk-V Mars (th-cam.com/video/HuU0LbnTbFk/w-d-xo.html) has an E-keyed M.2 slot, which is PCIe 2x. VisionFive 2 much the same. Then a lot of the Pine64 products include a PCIe slot: pine64.com/product-category/single-board-computers/ Just insane!
How are we supposed to burn your image file in the links? Its not supported by any sd-card writer I have tested? edit: i was able to burn ubuntu from luckfox wiki, but they do not tell you the root password, I can ssh to it, but luckfox does not work as password :/ edit: I must fix my eyes, its pico and not root as username lol. now it works
Hahah sorry I didn't get to this sooner and help you out, was sleeping :P On Linux and Mac I dump the image on the SD card just using `dd`, but you can probably use Disk Utility - I think there's an option to restore/write image. On Windows I use Balena Etcher. Just un-gzip it first if using DD!
its the community, raspberry and arduino both have a big community and therefore easy to solve problems and copy and paste some projects, riscV it just started, i just now hear more of it.
Probably not the correct place but an idea have say 4+ GPU's or parallel processors e.g RISC type CPU's mounted on a duel sided mo'board, xxGB ram running embedded Linux kernel with a free open source AI, Lora style PTLM (purpose trained language model) the idea is a portable pocket-sized assistant capable of reinforced/ self learning with onboard file access to the user private data, emails, and journals. The reasoning for multiple gpu's is to reduce the heat /energy requirements by cascading the load across them to keep them at a preset optimal temperature and service longevity. My free open source idea would be similar size to an oldskool pager, Wi-Fi enabled, portable, self-contained AI assistant.
My original idea was to create a portable personalised AI assistant, fully capable of online team gameplay and also fully individualised home and work assistant, similar to those portrayed in SCIFI films and old skool games.
Possibly, but unfortunately these just aren't the tech for it. The CPUs and even their dedicated NPUs are just not capable of running those sorts of models. Eg consider that my RTX3080 can only to 2 tokens / second with a 7B model, and you'd be lucky to get even 1 token per day if you clustered 100 of these.
@@delicacydelightThere's actually already some of those out there now! If you ignore Siri which now has transformer-based DNN behind it, or even Google Assistant which I think is now Gemini-powered, then you've still got startups like the Rabbit R1 which are pushing into these arenas!
@@PlatimaTinkersyeah, dabbled a little with them on my pc. The comment is in reference to smaller LLMs, Lora style purpose trained models, which don't need general training on every bit of human knowledge. Fully customise for Individual usage before it's adapted on to one's device - when this concept becomes a thing. All models have baseline training and filters, privacy, ethics etc Every profession is stored as pretrained LLM stored onboard M.2. Just say an engineering student researching a topic, it would load only need that engineering module. Common law module, e.g. draft legal documents to challenge a personal case. Property Law module when purchasing, it could do the conveyancing and review the contract terms conditions. Medical LLMs - General Practitioner who can pre-diagnose a condition........
@@delicacydelightYeah you'd have a few solid obstacles to overcome, but it's doable - hence some companies looking into it to some degree. Cloud-powered would possibly be better, but an offline dumber/more basic version for those privacy-minded folks could work. Note that LORAs are of no use on their own; they still require a base model to operate on. A Luckfox Pico or Milk-V Duo simply would not cut it for these purposes though, you'd need to start with something like Jetson Nano. I'd estimate you could prototype something like what you're describing for maybe $10k R&D costs, not including labour. I may be off there, but that's based on a cursory glance of parts needed.
Ain't wrong but that part of it works like any other - connect MIPI CSI camera, run your PyTorch project/model. There are good docs on it here: wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-RKNN-Test but either way I have a note in my "Part 2" production ideas to show it off a bit more. I am also planning to use it in this project: th-cam.com/video/BAabA__iIbo/w-d-xo.html
Hey you're very welcome! You shouldn't need DriverAssistant on macOS, and using my SD image you won't need SoC ToolKit either. If you want to flash the SPI you will, but you can probably try cross compiling github.com/xboot/xrock and then just write each of those binary files at the corresponding address as shown in the screenshot or listed in the .env file!
You'd need to specify exactly which camera. As long as it's supported by the kernel, eg IMX219, and has the right connector (Luckfox Pico has a 20-pin FFC, so might need an adapter cable), then at most all you'd have to do is modify the device tree. There is a bit of doco on it here wiki.luckfox.com/Luckfox-Pico/CSI-Camera. The camera it's designed for works out of the box with RKNN etc, and is probably best stuck to for documentation and support reasons, but of course depends what you want to achieve.
How ever much you put in 😊I usually use these faster 64GB cards: shop.plati.ma/products/sandisk-ultra-64gb-microsd-sdhc-sdxc-uhs-i-memory-card-140mb-s-class-10-speed-no-adapter
Hey what do you mean? I don't actually see a huge amount of difference between them - LuckFox and Milk-V seem to be competing in a very similar space, with very similar capabilities!
Haha, why's that? One of the best I've ever seen for an SoC! You can use Balena with the .img.gz I've provided if need be, but it will give you errors/warnings due to how the image is laid out.
@@PlatimaTinkers it's a windows only tool? Other boards don't need a special image generation tool at all. Even the bl808 has Linus and macOS versions of their programming tool.
@@davidwillmore You shouldn't need DriverAssistant on macOS, and using my SD image you won't need SoC ToolKit either. If you want to flash the SPI you will, but you can probably try cross compiling github.com/xboot/xrock and then just write each of those binary files at the corresponding address as shown in the screenshot or listed in the .env file! Bit of a shame there I guess, but at least there is some solution!
This board annoys me. You apparently need Windows just to make a boot SD card. I have a virtual Windows machine, but put off doing it, because it broke my sense of reality 😊 Why should I need Windows to install Linux??? Why would they not have a way to do it in Linux? 😂 Crazy talk. 😂
Hey nah no Windows needed mate, though I can see how it looks that way! You can either grab a pre-build SD card image like I've provided, then just `dd` it onto the SD card (bootloader on SPI flash should already be burned), or you can follow their Linux and macOS instructions here: wiki.luckfox.com/Luckfox-Pico/Linux-MacOS-Burn-Image I just use this Windows laptop as it's what I have laying out at the workshop, although I develop in Debian, and my couch laptop is a MacBook haha. Hoping that helps!
Yeah, bit Linux/Mac instructions say that this only works for SPI NAND FLASH and not tf card (sd card). I'll try the sd image link you provided, although I think I already did that, and it wouldn't boot. Thanks.
@@brettb.345 Yeah if you download the gz file, un-gzip it, then dump it on the SD like `tar -zxvf Pico_SD_Image-20231114.img_.gz -C ~ && dd if=~/img.img of=/dev/sdb bs=1M status=progress` that should do it. You are writing it to the whole card, not a specific partition, as it contains the partitions on it. Just make sure the card is... I forget, maybe 32GB or bigger? I cannot recall if I shrunk this root partition before I imaged it hah.
Hey everyone just one other note; I have a second channel where I test cookware, appliances, and cooking techniques: youtube.com/@platima. I'd love for you to check it out if it interests you (it's bit of a saturated TH-cam demographic it seems). Cheers 💚💚
Wooo that's me!
Is it booting from the ARM cortex A7 core first? If so then how do you use the Risc-v chip?
@@sgodsellifyYep the user level is on the A7. See github.com/LuckfoxTECH/luckfox-pico/issues/99, github.com/luyi1888/rv1106-mcu and github.com/syntacore/scr1 for info on what you can do with it.
Nice to see you covering these as was holding off grabbing some as unsure if support was available so will happily visit the store once stock arrives.
I see Milk-V have started getting software updated since your Mars video so would love to see a follow up video as some users are mentioning its improved since launch!
Yeah man I am so torn between these and the Duo now. The Duo is more capable though I think... I will see.
And yep I have a follow-up plans, as there is so much more in the docs now with using the 2nd RTOS core, making use of the TPU, etc!
bro thought i was gonna be the only one with these, lol , i love em, i had bought heltec sufficient IOT back in feb and basically just has to figure things out with it cuz no one had vids , im glad you made this , thanks . .
Hey man that's awesome. Yeah these were such a random find haha. So damn good though! Enjoy
For everyone considering this board: The Camera and Deep-Learning Models currently don't have Python support. Everything up until running the demo works great but sadly there's still a lot of functionality that requires significant tinkering in C++.
There is Python support for almost outputs such as PWM and SPI, they have a pretty good wiki page, but there's still some work to be done before the software is fully ready.
Very good to know, thanks for the input!
Thanks for the great video.
I have the 1103 board but not using it so far. Instead, using 3566 and 3588.
Will try to do the development on 3566/3588 with NVME and then port to 1103/1106 for power/size/cost consideration.
Yeah that is probably the way to go; compile time on these would hurt haha. And you're very welcome!
Waiting for my 1106 boards to arrive now :D
I’ve bought one of these but it was my first experience using a headless sbc. I got it to boot but felt it was useless since I couldn’t download the os, though it seems they’ve now made it publicly available! I’ll have to give it another go
OH awesome, which one did you get?
I'd definitely give it a go!
The algorithm has done well this day. Subbed.
awesome to hear mate, cheers
Awesome. Thanks for this board reviews 🔥. By the way, MCU is short for Microcontroller Unit
Hey very welcome and thanks for the support. Yep that's the ticket - my lumpy potato brain really struggles with the amount of TLA's these days 😕
Luckfox Pico is an excellent product with such small form factor. I've compiled the Ubuntu OS with tun/tap interface with Luckfox SDK, install Openvpn client, then configured it to connect my Openvpn Server. I put it to my customer's office. I no longer need to go to the customer's office for server checking.
Hey mate that is great to hear! They certainly are bloody small but handy!
Coming from a cyber-security standpoint I have some issues with that setup, but there are of course protection you can put in place such as IP-restricted jump boxes with push MFA, pubkey authentication per-session, etc
@@PlatimaTinkers I think VPN is more secure than Teamviewer or Anydesk
Those RISC-V board are fire! Can you do a handheld device based on one of those? Its much more compact than the Pi Zero (which has obviously more software support as its ARM based but ARM faced the same lack of support in the past). I am sure a such project would attract tons of views.
Yeah you absolutely could! I just don't have any ideas for one? Hah
@@PlatimaTinkers Adding WiFi would be a great start! :)
@@o0julek0o yuuuup!
Nice job on the review! I'd love to see a demo that exercises the video camera option, including taking pictures and recording video. Do they have software support for recording video? If so, mjpeg or H.264?
Hey thanks yeah I am planning a 'part 2' that gets into more depth. I'd assume you use libcamera or even ffmpeg as a frontend for interfacing with it, and can save in any format you want. The ISP appears to support JPEG, H264 and H265 hardware-acceleration. You can see in th-cam.com/video/JIFqV1vZB2Q/w-d-xo.html how I use a library in a few ways to grab media from a connected camera - I expect the same code would likely work here with minimal changes.
@@PlatimaTinkers it would be huge if it integrates cleanly with ffmpeg. A good tell is if it provides a standard /dev/video0 interface. Looking forward to part 2!
@@leadf00tlooking at wiki.luckfox.com/Luckfox-Pico/CSI-Camera/(am going back to the workshop in a bit and can test if I can find a compatible camera) it looks like it creates /userdata/video0 haha. You can see on the Core3566 that it is fully V4L2 compatible though (wiki.luckfox.com/Core3566/Core3566-pinout/6.%20CSI%20Camera/#63-viewing-camera-information) so I'd assume so!
This is definitely an Ai vision board. The NPU is as big as the Orange Pi. An inference demo would be cool.
Yeah pretty damn awesome. I am aiming to do demos in Part 2 sometime in the future. Such a long backlog of content haha
It's all about the documentation in my view. :)
Weeeeelllllllllllll without SDK, the documentation is useless.... Without documentation, the SDK is useless 🤣
Hey finally one I am interested in. Always searching for SBCs with proper networking not this wireless stuff that scales like a jenga tower due to nonsense like the laws of physics.
hey yeah I'd say right now I am torn between this, and the Milk-V Duo; th-cam.com/video/YqUtGk0DHbQ/w-d-xo.html. Both VERY capable tiny little buggers
Stupid laws of physics!
@THE-X-Force yeah gets me every time, especially after about 9 beers
Thanks for all the great info. Could you please increase the font size so we can read it on smaller screens? Thanks, again.
Hey you're very welcome, and yeah no worries I just exported that small segment zoomed in if you want to have a closer look, but next time will set the terminal font much larger! th-cam.com/video/cx3rj6a_K2w/w-d-xo.html Cheers.
Loving the coverage of these small Linux boards that run Linux.
I'd be really interested to know your recommendation for any which run using Power over Ethernet - I think being able to make sensors/IoT stuff that runs off PoE would be a killer app.
Hey glad to hear it mate! PoE is a very interesting one, your best bet would probably be using passive 24V POE, else the Duo IO board has some details about adding POE in the schematics for it!
I might honestly design and sell a POE adapter for SBCs. Wouldn't be hard
I need Power over Wifi.
thanks for reviewing it ! was looking on ali for purchase, but the lack of info and overall interest pushed me away
Hey mate you're very welcome. Hoping you ended up getting some, as they're damn awesome little units!
I just tried to order a few, but they don't have any payment methods for the USA. That's a bummer.
What?! Not even Paypal? That seems so odd. Watch this space for when my stock arrives!
could the MCU be an embedded microcontroller, much like the PRU units on the Beaglebone Black? This would be very useful.
Yeah I think so, but completely closed. Eg I think it is essentially the ISP and NPU
This is A great video. I got my Luckfox Pico Plus to work and could SSH in to it and login over the network. DHCP worked right away.
Worked pretty well actually!
Now I am trying to use the Ubuntu image and boot from SD.
I *kinda* got this working... but seems odd.
Took for ever to get an IP address. But now it shows up... maybe an initial install/setup going on first time?
Then to SSH login, it works but takes about 10-15 seconds to finally get to a prompt.
This typical? I haven't tried another SD card yet. Need to find another one, just to make sure maybe a sketchy card.
But over all it seems a bit, lethargic using the Ubuntu / SD setup ?
doing A apt-get update took like 20 minutes or something to do.
Hey thanks for the feedback mate that is really awesome to hear.
I've heard some people complain about this! I've personally not experienced it. First boot does always take a while though.
What image were you using? I can test it out and ask their support team if need be.
If you plug camera it turns into IP camera streaming. And it gives you notifications in console if something happens in the image.
Yeah that is so awesome, I started playing with that afterwards and am leaving that side of it for a 'Part 2'
@@PlatimaTinkers I was also imoressed about how easy was to set it up. I was able to recompile kernel and create my own .ko files while I never did that before. Its neat to learn how busybox is compiled on such devices.
@@darkobul1Oh wow that is awesome. What guide did you follow? I'm yet to do so, but was thinking about how hard it would be to port esp_hosted...
@@PlatimaTinkers they provide VM with all tools needed. I bookmarked it somewhere I will send you. I think was in chinese but I used google translate.
I am interested to see about esp_hosted as I dont know anything about it. Tool chain is ARM its a chance should not be hard?
@@darkobul1 Appreciated! And yeah I'd not think it too hard either - instructions look fairly straight forward.
Hi I am interested of that board (especially pico mini version). I've got two questions 1) is it possible to install full MQTT broker on that small board? 2) is it possible to solder 4 ethernet cables (100Mbit) directly to pico board? (I am interested of that because of lack of space for eth module). I know you say something about my second question - but you are speaking so fast and I'm not sure you say "can" or "can't" 4:27 regarding to soldering cables directly? Thank you in advance and sorry for my english.
Hey mate your English is fine!
1) You could probably install MQTT broker, I think it would probably run fine but I do not know how much CPU it would use. You would need to download the SDK and do a compile with it, which I can try if you want? 2) You could definitely solder the cables on, but it is a little risky and you may get a lot of TX/RX errors. I would at least recommend diodes on the TX/RX lines. This is pretty much all the ethernet module has.
I say you can solder cables on, BUT you might blow it up haha. FYI I upload subtitles so hit 'c' and you can see English captions of exactly what I am saying. I know I talk too fast 🤣
@@PlatimaTinkers Thank you for the answer. I have ordered one board - the smallest pico mini version B (with flash mem)+ one ethernet board. I decided to use eth.board because of lack diodes you mentioned. It should be possible to solder both boards together to achieve a small "qube". Unfortunatelly i have no schematic of eth.board but I think it could possible to use "passive poe" by injecting +6V and gnd into not used pairs of 100Mbit ethernet cable(i will check connections and modify eth.board if necessary). On pico board there is step-down converter which accept absolute max +8V so after proper wiring it sould works in my opinion. Regarding to MQTT I'll do some tests (of course if I do not blow this crap up ;) ). I think CPU should be enough fast - 1Ghz is quite ok, but 64 megs of RAM can be a bottleneck. We will see.
@@grzesiekyogi1518 hey the ETH module is super simple; it just connects the TX+ TX- RX+ RX- pins with flyback and polarity diodes. You can see it well here on my shop page: shop.plati.ma/products/ethernet-board-for-milk-v-duo-and-luckfox-pico. You may be able to access the power pins on the bottom of the board where they are soldered.
It's important to note that passive POE is 24V though, and up to 27V, so you'll need a step-down converter to bring it to 5V and connect it to VSYS. For this I have used shop.plati.ma/products/mini-360-dc-buck-converter before, as long as it's 24V passive, not 27V! You'll also want to tune that output voltage very carefully, then put some glue on the potentiometer.
@@PlatimaTinkersOn luckfox pico board there is already a dc-dc step-down converteter (EA3036C attached to VSYS) which accept absolute max +8V. So it sould be possible to inject +6V (I know - it is not according to passive POE standard, but it sould works without any additional converters and it is easy to inject by standard 2$ injector + standard power supply). If the LAN cable is not too long it should be possible even inject +5V (EA3036C converts to three voltages where 3.3V is the highest one - so I believe if there is for example 4.5V at VSYS due to cable resistance it should work anyway). I will test it soon.
@@grzesiekyogi hey yeah but I don't know where you are getting those voltages sorry - passive POE is 24-27V, and active POE is 34-57V. Eg, where are you getting your low 6V from? hah. The EA3036C looks like an awesome little unit though!
Can the RISC thingo be used in digital signal processing to take some of the work away from CPU
Yeah I think that's how it works - eg it does all the media processing
maybe they wrote ram in mbit? 512mbit = 64 mbytes
Oh my mate you're a genius! I had another look, and on www.luckfox.com/Luckfox-Pico/Luckfox-Pico-Max it has "MB", so correct, but www.luckfox.com/Luckfox-Pico/EN-Luckfox-Pico has "Mb" - meaning it is 64MB like you said. I cannot believe I missed that. Still, not great marketing in my mind haha.
This is epic. The new PS4 jailbreak craze is at an all time high rn! Awesome video my dude 😎👍🏻
Oh, shoutout to PCBWay! They're absolutely great!
Is it? News to me 🤣🤣
Cheers either way mate!
for 1 i did not see the sd over the flash . I thought it was a covered bank of resistors. I like the thernet . the milkv should have done the same but it has a header spot.
Yeah it's an interesting design that's for sure!
Sorry about the delay, nearly missed this comment hah
Interesting possibilities with this thing. Much more than the Ox64. Use the RISC-V as VGA GPU? Will be checking the camera options.
Yeah I don't see why not. If you check out www.lontentech.com/products/raspberry-pi-vga-adapter-board-gpio-to-vga-module-shield-for-raspberry-pi-4-model-b-3b-2b-b that does 1080p60 using 6x GPIO pins as it only requires a ~150MHz pixel clock from memory.
Great movie . Is it possible to flash sd card or nand under Linux? Which application should I use in this case?
Thanks mate! And yep, you use `upgrade_tool` for that. They have some instructions here: wiki.luckfox.com/Luckfox-Pico/Linux-MacOS-Burn-Image. It's very similar to the original process, but without the GUI!
@@PlatimaTinkers thank you :)
@@PlatimaTinkers I think it will be good idea for a new one movie :) Their explanation is about nothing :)
@@ruhmachino any time!
@@ruhmachino Possibly very true! THank you
Is it possible to make a pico w micro? Like rp2040 on one side and wifi module on the other. 5 gpio. No USB but pins for flashing. Love your pro board
Yeah I've been thinking about designing something similar. Like a SG2002 with a W600 module.
Cheers mate!
@@PlatimaTinkers I would love a half sized pico w
@@mrlately7253 You're talking about Raspberry Pi with an RP2040 / ARM Cortex processor?
Cause there's the Pico Mini, which has no wifi but is stamp sized. Or the LycheeRV Nano W, which has wifi and uses SG2002 so both ARM and RISC-V options :)
shop.plati.ma/products/sipeed-lycheerv-nano-we-beta
@@PlatimaTinkers thanks for the support beyond the videos.
I basically need to be able to do the same I can on a pico w but smaller. I only use a few pins. I'll check the links you shared. Thanks again
@@mrlately7253 All good!
And yeah LycheeRV Nano WE would be good then! Else shop.plati.ma/products/pine64-pinenut-model-12s-wifi-ble5-stamp is tiny and cheap and has wifi.
There's a few options out there. I also have some ESP32-C3 and S2 boards like littlebirdelectronics.com.au/products/esp32-c3-mini-development-board-based-on-esp32-c3fn4-single-core-processor-160mhz-running-frequency-2-4ghz-wi-fi-bluetooth-5 that work really well and can run Arduino, ESPHome, MicroPython or C! I use them for garage door controllers, etc. DFRobot make the 'Beetle' series which are the same.
I have shop.plati.ma/products/esp32-cx-riser-board-v1-1-assembled-with-c3-and-led which is a prototype product but does work. Maybe not for your use case though!
Really probably depends what IDF/SDK you want to use! C, Arduino, MicroPython being the big ones.
I tried compiling some programs for the plus, but you can't store any executables on the sd card because it has to be in FAT32 which has issues with file permissions ... that was a bummer
Yeah in Pt 2 I'll cover a bit more like that. Best to store within the image, eg copy it across with `adb push` after doing a cross compile, else you can probably slap another partition on the SD card as ExFAT... I'll have a tinker and let you know!
@@PlatimaTinkers yes please, if it's possible with the ExFAT let me know, thanks a lot for video BTW
@@AboubakrAyeah no worries mate I've got one connected out at the workshop that I'm heading to shortly. Will try it out and let you know!
You're also very welcome. Cheers
@@PlatimaTinkers BTW I noticed you're asking for the VM in another comment, I tried adding the link in the comment but it got deleted for some reason
this is it, it has a VM containing the toolchain for cross-compilation, I've given it a try and it works perfectly
drive.google.com/drive/folders/11Kc5Kogjjoqe5yWO0VI1-x4RMYv4m98L
I'm trying to send the link for the VM that has the cross-compilation toolchain in a comment but it gets deleted, I can send it on email if you're interested
We dont need no stinkin' FTDI, Silicon Labs CP2104 for the win!
True story, at my work we use a cp2108 chip in our diagnostic board. Its the same thing as the 2104 but it has THREE serial endpoints.
Oooh that is pretty sweet - I could definitely see some value there, especially with some devices having separate serial interfaces for a) flashing then b) debug output.
@@PlatimaTinkers oh .. the cp2108 has at least four, endpoints numbered 0 through 3. I set up some udev rules on Linux to symlink the endpoints based on USB device attributes.
@@kayakMike1000 Aaah nice. I'm currently developing with a CP2102 - turns out they make quite the array of bridges!
whats the power usage? is there a poe version? something like this for a pihole or adguard-home would be awesome.
No POE version, but not hard to mod I think! Power usage is in the description. I think there is not enough RAM to run a service like that though? Not too sure!
so it's the size of a Pi Pico, or a bit bigger with a full on network jack, and runs Linux. So I assume you can run micropython or something from that Linux terminal to control the GPIO? I've only got basic familiarity with microcontrollers.
Yo yeah pretty close. It supports Python 3 as of just recently - comes with the image. MicroPython itself has been ported to ARM (github.com/micropython/micropython/tree/master/ports/bare-arm) and that MIGHT work, but would probably require a few tweaks. CircuitPython runs on the STM32F746 though, which is also Cortex-M7, so that could also work. Just not sure how the ethernet support would go with those binaries - maybe better to stick to the Linux base layer!
@@PlatimaTinkers That's neat! I'm mostly used to Arduinos that don't even have an OS of their own, so the idea of SSH'ing into one to use it is kinda new. Thanks for the info.
@@DFX2KXyeah I do like Arduinos, but I'm Linux at the core, so using SBCs like this that are _basically_ embedded systems but still have a shell makes me pretty damn happy! And you're very welcome mate. Have an awesome weekend
I just ordered the LuckFox Pico Max 256MB 1106, after watching this!. Will be fun to use something else than a Raspberry Pi :)
Glad to hear mate! Hope you ordered it from me 😋 (all good if not, #shamelessselfplug)
Keep your eye out for more too, sounds like they've got some neat products in the works!
@@PlatimaTinkers ah do you ship to Sweden? Sorry I did not order from you, but I will say your prices are almost half of what they are here in Sweden for the Luckfox Pico Max (your shop says 165SEK, and I paid 300SEK here in Sweden) , so if you do, I can order from you next time :)
@@AndrewTSq Hey mate yep I do, and all good! The prices will probably go up a little soon as they are based on USD-AUD for my buying, but it definitely won't be that much!
There is also mini ver.
Very true! Have ordered one
Wooow, such a great video!
Do you know how does the camera work? I'm trying to do object recognition but the only thing I know over the camera is that generates something like a .jpg file, does that file updates all the time? that is how does cameras works on linux? I'm a newbie in this world, thank you!!!
@ericcaudillobarbosa586 Hey thanks! The camera does not have to generate a JPG - it's just a data stream that you can interpret any way you want; save images, save videos, perform realtime detection, stream it to the network, etc. There's some example info on it here: wiki.luckfox.com/Luckfox-Pico/CSI-Camera and
Dude...this probably isn't going to be the ideal board for object recognition. object detection maybe. If object recognition is what you're interested in. Buy a raspberry pi 4 or 5 or pick up an Nvidia jetson. The jetson was made for AI applications. The most important thing to know is that you can't make chicken salad out of chicken shit, and these under 20 dollar boards are chicken shit when it comes to these type applications. If you're building an AI system or application and your board doesn't support python, you're going to hate it. peace out bro.
@rickhunt3183 I think you've missed a bit of the specs, as you're not correct all sorry mate. It'd be good for you to check out how the ISP pipeline and TPU work, plus how RKNN stitches it all together.
Unlike the Pi 4, the board has 0.5 TOPs NPU and a 4MP 30fps ISP. It can do object detection AND recognition, and you can even do facial recognition at a decent clip. This RV1103 (lower end to 1106 and 1109) SoC can run YOLOV5 at about 10fps from memory, RetinaFace at a very good speed, plus plenty of other image recognition, text recognition, car plate recognition, speed recognition and similar models all at a solid pace.
And it supports Python, comes with it out of the box in fact. Really not sure where you're getting any of your info from sorry.
You can have SD and SPI? Nice that's better than the DUO.
what do you mean? i’ve been working on a Duo project but i’m stuck on SPI. it feels like reading/dev/spidevx.0 results in weird, unexpected outputs
is this because of the microSD? 🫣
@@Onkoe I would assume so. I asked on the discuss forum and was told that the same signals run both ports. The schematic agrees. You get one or the other.
I believe this true, but I believe that it ALSO physically gets in the way :(
@@davidwillmore I believe this true, but I believe that it ALSO physically gets in the way :(
@@Onkoe They are physically in the way of each other too :(
Nice video as always. Thank you for the video.
This board is similar to the Milk-V Duo, eh?
I personally like boards with WiFi built-in for convenience.
Hey thank you kindly, and yeah similar to Milk-V Duo and Ox64 in a number of ways!
I prefer wifi too hah, but you could add USB or SPI wifi to this, else LoRa I'd say.
Cool. OS only boos from SSD ? It support Ubuntu?
It's u-boot, so OS default from SD card or onboard SPI flash, but you can boot from ethernet or USB if you wanted!
The default image is Debian based, the same as Ubuntu is, but you could easily roll-your-own using `debootstrap` / `cdebootstrap` if you specifically wanted the Ubuntu flavour.
Just note that only Debian has significant RISC-V support so far, so best sticking to that ecosystem and the vendor apt repos.
cool!
Indeed!
Can it run alpine Linux and can I connect a USB soundcard via usb c?
I think you asked on Reddit? :P And yeah, USB-C addons work fine as long as the Linux kernel has the driver, and it could probably run Alpine but it'd take a lot of kerjiggering I think!
This SoC have support for linux users? (I mean if i can program it from linux system)
Yep!
Please consider zooming in on the text .. very difficult to read .. really don't need to show the whole thing top to bottom and w/ all of the empty space on the right side. ☮
Yeah mate no worries - what part were you referring to? It can really depend on the resolution of your screen but I try to make it readable at 1080! Cheers for the feedback
@@PlatimaTinkers 7:17 onward .. for example (what I'm talking about can be avoided several times).
Thanks for being kind about what I hope is constructive criticism. ☮
Great video .. don't take me wrong. Thanks for sharing your knowledge and experience!
@@THE-X-Forcehey yeah I get what you mean now - the terminal output! I just had a look and it's quite readable here, you might want to click the cog icon for the settings and make sure it's set to 1080p50 - that was the recording resolution, so that playback fullscreen should look close to 'native'. I have increased the PuTTy font size though so it's easier in future videos!
Any criticism is absolutely appreciated - I am rubbish at this so value any input or feedback. Cheers!
@@PlatimaTinkers Ok .. my res is good and I _can_ make it out (barely on this 17" laptop at 1080), but maybe you will agree that having nearly half the screen show nothing but empty black space is .. less than ideal?
I'm suggesting cropping most of that dead space out .. and since that would make the video vertical (aka horrible), lopping off some of the top text that the viewer has already seen, and doesn't need to focus on. Hope that makes sense. Maybe try it and see what you think.
In any event, it's no big deal. Thank you again, and you are definitely not rubbish at this.
Best of luck from NY☮
@@THE-X-ForceYeah mate had not considered that at all, I'm on a 27" haha. You're absolutely right about the waste of screen real estate though; if I make the font size 20pt it looks like it fills the width nearly perfectly. TH-cam's compression should massacre it less than too!
Good luck with the freezing temps there, we've got 86-104 here!
Hmm ... This LuckFox Pico is an interesting board.
Ain't wrong mate - found some other cool stuff too which I'll put in a Part 2 video!
hey, can I use this board for object detection?
Yea mate absolutely; the built in ISP, reserved RAM, and 0.5 TOPS NPU is designed exactly for that! See wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-RKNN-Test for more info!
can they process 4K cam? and if so can they stream over some wireless?
Hey they have no wifi built in, but they do support USB device mode so you could easily connect USB or SDIO wifi as long as you have the drivers.
They have MIPI CSI 2-lane for camera connectivity, so you should be able to do 4K@30FPS but the capability of the ISP is limited to about 4M so you'd only get QHD (2560x1440) unless you dropped the FPS further. Back of the napkin math would be 4K@10FPS to JUST fit within that hardware capability.
The Pico Pro and Pico Max support 5M30FPS ISP so you should be able to get 15FPS which is what our eyes/brain usually need to perceive something as animated.
Enjoy!
Hello, i tried to stream over usb. Its working. Is it possible to stream over wifi or something by connecting with esp32? If its possible please help me to stream wirelessly using esp32 as a wifi midule!
@alfaskc729 You need to read the documentation friend. I am not a personal ChatGPT sorry.
I saw a video that compared RISCV against ARM and the ARM core was consistently better so I think the ARM processors give you better performance.
I also saw a video that complained a lot about the support for a lot of these Chinese SoC carriers and the No. 1 issue was making them work with established code libraries etc, so for me PI may be more expensive but life is short.
That's not really comparing Apples to Apples though, that's more saying you want two oranges but being given a lime and an orange. They're quite different in how they work, and RISC-V is quite new.
Yes, if you want raw power, go ARM, but you miss a fair bit of the point then, but it's not always about power and these 1Ghz+ processors are a tad insane; that amount of power in an embedded device is just not necessary 99% of the time.
0:28 I think you’re right about people wanting these, but probably not the for the reason you had in mind when making this video
Oh?
turns out you can use it to jailbreak the Playstation 4 with it, think their sales went through the roof thanks to that.
@@erwinvb70 Oh I had no idea! I'll have to get more stock haha
since when the amount of memory is in bits not bytes?
I think it's more common in embedded devices TBH
Can you use VNC Server or/and Desktop on Luckfox Plus?
Uhhh, I guess, but I cannot fathom the use case there. It's not really designed to have a GUI, but you could probably enable swap then install X and VNC to it from another computer. It'll run like a 1995 Dell. If you want to run X or VNC on the LuckFox SBC you probably need to implement a VGA interface by bitbanging GPIOs.
Ok thanks, should I buy Luckfox Pico Plus or Raspberry Pi Zero 2W?
@@InikoPro No worries. Completely depends on what you want to achieve though - do you need wifi or ethernet, do you need NPU or not, etc. They both compute, but both specialise in different functions.
That means that you already forgot about the milk-v? 😢
Hah no, very different use cases! Very similar form and price though!
Ciao, ho riscontrato un problema
Il mio LUCKFOX funziona
Per soli 10 secondi poi si spegne
Esiste una soluzione? Grazie
Può essere dovuto a molte cose, ad esempio l'immagine della scheda SD, l'alimentazione, ecc. È meglio provare i forum su forums.luckfox.com/ o il supporto tecnico su wiki.luckfox.com/intro. Buona fortuna!
It seems that exist a 256MB variant (2048MiB)
Yeah mate those are the Max and Pro! They also have RV1106 SOC. So damn good
I wonder, since I never really see people use the tpu in such boards, yet on paper some of them should give compute power in speciffic AI related workloads close to around a quarter of the performance of a rtx4060(one of those new beagle bone boards, not cuda however), while using insane amounts lower power.
so why doesn't seem anyone to be using those TPU's or C7x DSP's, fpga and such for AI tasks such as language models and image generation, just like the raspberry pi's they probably could be quite doable to cluster and might even be added on a single pcie card, I wonder what kind of performances could be reached that way, even though ram might be a issue, but for everything there are solutions. I mostly just wonder if there really isn't any development for AI to run on such hardware, since it would be much more efficient than how it is done now, and they already did make things to run it on integrated gpu's and old cpu's which technically seen would be a lot slower at such things in basically any way since they also don't benefit from the cuda cores or such.
this has also been proven by intell adding a very small npu into their new cpu's, and using that NPU actually was much faster than the cpu and the gpu in those apu's, also noticably beating the amd apu's, despite that integrated NPU being a very small very low power part.
to make it more serious, some of the SBC's from beagleboard actually have a quite a lot faster NPU in them than those intel APU's have, intels next gen was supposed to have faster NPU's, but those results where gotten by people using the old integrated NPU which actually was slower than the NPU in some of those new beagleboard computers. so honnestly it seems like big AI and such really miss a lot of opportunity by getting stuck in those big gpu's and such rather than trying to find ways to make it work properly on such small formfactor low power hardware.
Hey mate bit of a long comment and I'm slightly rushed sorry, but I think you're missing a fair bit of information there from what I can understand of what you've said. The terms are all right, but you're seemingly not using them in the right context.
The TPU/NPUs in these might do 0.3-5 TOPs @ INT8, which is great for some basic machine learning, but absolutely nothing close to what an RTX4060 can do. An RTX card is hundreds of TOPS and also using a 16 bit floating point number as opposed to integer, so you can use less space for faster transformations or inferences with a greater degree of accuracy given the wider gamut of number values that can be represented. Those can do some complex DNN functionality like text to image generation, large language models, etc. Even further to that, RTX 3k and 4k series supports a newer type of floating point that is specifically designed to work better with AI (bfloat16).
These SBCs I develop on and show off are small units that are better at ML for tasks such as facial recognition, and then embedding them in smart door bells, IP cameras, robots, gardens, etc. There's then also a bit of an issue with memory; to run a 7B parameter LLM on a GPU you're going to need a hefty card and then minimum 12GB of VRAM, and even then you'll only get 1-8 tokens/second. These things have 16-512MB of RAM, so you'd be lucky to get a 500,000M param LLM running, and besides it talking garbage, it would be extremely slow due to the IO bandwidth available, and that's if you can even quantise a fp16 model to run on int8, in which case you'd likely lose some of the LLM "magic" in that case.
Long story short; different power level, different 'neural' technology, different use case, different market and price. That is just my understanding though, and I could very well be getting some details wrong there! That being said, I have 2x GTX 1070 Ti's, an RTX 2080, an RTX 3060 and an RTX 3080 that I run different LLM and diffusion models on, so I do have a fair bit of familiarity with the tech... Just still saving for an RTX 4090 one day 🤘 Thanks for your input either way!
@@PlatimaTinkers well if you are saving for a 4090 you might wanna wait for the rtx 5000 series which might launch not to far away are supposed to be much faster than the 4000 series.
for the TPU's from most of them I didn't actually know their speed,
However I speciffically based it on the beaglebone AI-64.
the beaglebone AI-64 is suposed to reach
quite high numbers and also has multiple different accelerators from which some are really fast.
even though I indeed do not know all about such things.
but I do know things could be much more efficient, but that would also largely depend on the AI itself being designed to be efficient instead of based around normal hardware.
however lately I was thinking that since AI like stablediffusion uses essentially weighter random noice, that it might be accelerated and made more efficienty by using a second model at first which instead of training noice clouds would train more on fractal clouds, that way if that works right it might be able to skip many steps in the proces and only do a few steps at the end to make them some less fractalish.
@@ted_van_loon I don't think there is going to be huge jump from the 4K to 5K range from what I can see - eg minimal new data types, just a smaller lithography, extra pipelines and cores, faster memory, etc. I'd say once the 5K comes out, I'll grab a 4080 Ti or 4090 as the price will drop, then maybe wait for the 6K when they do something mental like 1nm lithography, GDDR7, etc.
Re the BeagleBone AI-64, it's a pretty bloody powerful board - it'd definitely go a long way towards something like a high-end video editing system if utilised properly, but it is still not designed to run models like LLMs. It's got an 8TOPS TPU, high bandwidth ISP, a decent little GPU, and 4GB RAM (ignoring the DSPs). This is higher than the Fire specs we just discussed, but more designed for robotics and automotive. Eg it would be great at real-time processing of multiple video signals for things like depth mapping, object recognition, collision prediction, etc.
You can find examples of the models at github.com/TexasInstruments/edgeai-modelzoo and get an idea for how the board may be used. If you wanted to compare it to a GPU for running diffusion models like SDXL or LLMs such as Mixtral 0.2, you might need a few hundred of them, and then you end up with something less power efficient and less compact plus the challenge of trying to get the data between them using very limited interconnects like 1Gbps LAN (AI infrastructure will often run 40Gbps - 400Gbps fibre interconnects).
Large companies such as Mistral, OpenAI, Alphabet, Meta, etc, are all working on making these inference and diffusion models more efficient, however, the long and short of it is that how they work requires massive amounts of data to be multiplied at blazing speeds, which requires a lot of fast RAM, high bandwidth, and seriously huge parallel pipelines. CPUs are crap at this, TPUs are okay at this, GPUs are great at this, and AI accelerators do it excellently.
I get where your idea is coming from, however, unfortunately it does not work anything like that, and the AI is not based around any hardware; the goals are typically based around creating the fastest, most accurate, and smallest models possible, and it really doesn't matter what hardware it runs on at that point; GPUs are just the easiest for consumers, where as for high-end inference you'll want H100's or similar. They are called "GPUs" but really, they aren't. They're not meant for playing games, you can't plug a display into them, they just multiply matrices amazingly efficiently.
We all started using GPUs for this because rendering images and textures just happens to use a similar pipeline. I hope all that information helps you understand a bit more!
The thing I don't like aboutpin64 is documentation and no guides. I own a pine64 cluster board and othe stuff that has never worked.
Yep everyone agrees haha. They kind of just make hardware, and leave the rest (software, doco, etc) to the community. Even their new/current website was made by someone voluntarily 😐 Pros and cons!
SoC's seems to be a lot more complex compared to microcontrollers.
Oooh yeah mate, but more capable too. Eg Atmega32U4 - 16MHz 8-bit RISC-V, 2.5KB RAM, 32KB Flash, 12-channel 10-bit ADC, USB 2.0 integrated for flashing. RV1103 - 1.2GHz 32-bit RISC-V, 64MB+ RAM, 26 GPIOs, TPU, ISP, Ethernet, MMU, MIPI DSI/CSI, USB Host & Device, SDIO for SD or Wifi, PCIe, SPI NOR flash controller, etc.
@@PlatimaTinkers If their increased complexity didn't bring more capabilities the designers f'ed up.
@@bogmaerkehahaha ain't fkn wrong mate
Being a hobbyist and just noodling with this stuff, this is like a "gross amount of power in a microcontroller"
So when do they just start shipping these with PCIe interfaces and stop pretending these aren't computers?
Wait... do any of these have a PCIe interface?
Hahahah yeah some of these SBC's / eSBC's support PCIe. Eg the Milk-V Mars (th-cam.com/video/HuU0LbnTbFk/w-d-xo.html) has an E-keyed M.2 slot, which is PCIe 2x. VisionFive 2 much the same. Then a lot of the Pine64 products include a PCIe slot: pine64.com/product-category/single-board-computers/
Just insane!
How are we supposed to burn your image file in the links? Its not supported by any sd-card writer I have tested? edit: i was able to burn ubuntu from luckfox wiki, but they do not tell you the root password, I can ssh to it, but luckfox does not work as password :/ edit: I must fix my eyes, its pico and not root as username lol. now it works
Hahah sorry I didn't get to this sooner and help you out, was sleeping :P
On Linux and Mac I dump the image on the SD card just using `dd`, but you can probably use Disk Utility - I think there's an option to restore/write image. On Windows I use Balena Etcher. Just un-gzip it first if using DD!
its the community, raspberry and arduino both have a big community and therefore easy to solve problems and copy and paste some projects, riscV it just started, i just now hear more of it.
Yeah pretty damn awesome; both are starting to adopt more RV32 and RV64 which is awesome!
Probably not the correct place but an idea have say 4+ GPU's or parallel processors e.g RISC type CPU's mounted on a duel sided mo'board, xxGB ram running embedded Linux kernel with a free open source AI, Lora style PTLM (purpose trained language model) the idea is a portable pocket-sized assistant capable of reinforced/ self learning with onboard file access to the user private data, emails, and journals. The reasoning for multiple gpu's is to reduce the heat /energy requirements by cascading the load across them to keep them at a preset optimal temperature and service longevity.
My free open source idea would be similar size to an oldskool pager, Wi-Fi enabled, portable, self-contained AI assistant.
My original idea was to create a portable personalised AI assistant, fully capable of online team gameplay and also fully individualised home and work assistant, similar to those portrayed in SCIFI films and old skool games.
Possibly, but unfortunately these just aren't the tech for it. The CPUs and even their dedicated NPUs are just not capable of running those sorts of models.
Eg consider that my RTX3080 can only to 2 tokens / second with a 7B model, and you'd be lucky to get even 1 token per day if you clustered 100 of these.
@@delicacydelightThere's actually already some of those out there now!
If you ignore Siri which now has transformer-based DNN behind it, or even Google Assistant which I think is now Gemini-powered, then you've still got startups like the Rabbit R1 which are pushing into these arenas!
@@PlatimaTinkersyeah, dabbled a little with them on my pc. The comment is in reference to smaller LLMs, Lora style purpose trained models, which don't need general training on every bit of human knowledge.
Fully customise for Individual usage before it's adapted on to one's device - when this concept becomes a thing.
All models have baseline training and filters, privacy, ethics etc
Every profession is stored as pretrained LLM stored onboard M.2.
Just say an engineering student researching a topic, it would load only need that engineering module.
Common law module, e.g. draft legal documents to challenge a personal case.
Property Law module when purchasing, it could do the conveyancing and review the contract terms conditions.
Medical LLMs - General Practitioner who can pre-diagnose a condition........
@@delicacydelightYeah you'd have a few solid obstacles to overcome, but it's doable - hence some companies looking into it to some degree. Cloud-powered would possibly be better, but an offline dumber/more basic version for those privacy-minded folks could work.
Note that LORAs are of no use on their own; they still require a base model to operate on.
A Luckfox Pico or Milk-V Duo simply would not cut it for these purposes though, you'd need to start with something like Jetson Nano. I'd estimate you could prototype something like what you're describing for maybe $10k R&D costs, not including labour. I may be off there, but that's based on a cursory glance of parts needed.
Do you think it could handle recording 1080p60 video and storing it? Does this handle overclocking?
In the video description is a title 'Documentation' with a link next to it. Click that.
Must be a slow down in printer (and 3D printers) sales.. and drones (a regulatory thing)
Yeah possibly!
So your thumbnail saying “ai vision” and you did not show how it works.
Ain't wrong but that part of it works like any other - connect MIPI CSI camera, run your PyTorch project/model. There are good docs on it here: wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-RKNN-Test but either way I have a note in my "Part 2" production ideas to show it off a bit more. I am also planning to use it in this project: th-cam.com/video/BAabA__iIbo/w-d-xo.html
hmm, is there a Linux/macOS version of that little tool? if not, this board is some hot garbage to me 🔥
in any case, thanks for the great video!! ☺️
Hey you're very welcome! You shouldn't need DriverAssistant on macOS, and using my SD image you won't need SoC ToolKit either. If you want to flash the SPI you will, but you can probably try cross compiling github.com/xboot/xrock and then just write each of those binary files at the corresponding address as shown in the screenshot or listed in the .env file!
does any one know if this supports the pi camera
You'd need to specify exactly which camera. As long as it's supported by the kernel, eg IMX219, and has the right connector (Luckfox Pico has a 20-pin FFC, so might need an adapter cable), then at most all you'd have to do is modify the device tree.
There is a bit of doco on it here wiki.luckfox.com/Luckfox-Pico/CSI-Camera. The camera it's designed for works out of the box with RKNN etc, and is probably best stuck to for documentation and support reasons, but of course depends what you want to achieve.
what is SD card memory space?
How ever much you put in 😊I usually use these faster 64GB cards: shop.plati.ma/products/sandisk-ultra-64gb-microsd-sdhc-sdxc-uhs-i-memory-card-140mb-s-class-10-speed-no-adapter
@@PlatimaTinkers Thank you
@@user-lj8kf8tu9h Any time!
I got a similar one milk but without Bluetooth/wifi I don't use it, too much friction
Yeah these have no BT/Wifi unfortunately. I'll definitely be working on that in another video!
Little b in 512 Mb = 512 megabit divided by 8 = 64 megabyte memory
Yep!
Then the term "mebibytes", which is technically the right one, just make it confusing too haha.
Curti só pelo arroto!🤣
Hahahahaha! Felices vacaciones / navidad 🍻
I dont see this being a killer if they do not have a linux based uploader.
the MilkV duo can boot from sd card andallows alot nore
Hey what do you mean? I don't actually see a huge amount of difference between them - LuckFox and Milk-V seem to be competing in a very similar space, with very similar capabilities!
I missed to identify the purpose of the video ... 🤔
That you did
No Wifi in 2023???
Not everyone needs wifi :P But yeah, it does feel a tad odd!
512 Mb = 64 MB (bits vs bytes)
Yep that's what the other guy said.
That SD card creation process looks horrid.
Haha, why's that? One of the best I've ever seen for an SoC! You can use Balena with the .img.gz I've provided if need be, but it will give you errors/warnings due to how the image is laid out.
@@PlatimaTinkers it's a windows only tool? Other boards don't need a special image generation tool at all. Even the bl808 has Linus and macOS versions of their programming tool.
@@davidwillmore You shouldn't need DriverAssistant on macOS, and using my SD image you won't need SoC ToolKit either. If you want to flash the SPI you will, but you can probably try cross compiling github.com/xboot/xrock and then just write each of those binary files at the corresponding address as shown in the screenshot or listed in the .env file! Bit of a shame there I guess, but at least there is some solution!
@@davidwillmore Luckfox engineering just gave me this new link: wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-MacOS-Burn-Image/ Enjoy!
@@PlatimaTinkers cool, thanks.
This board annoys me. You apparently need Windows just to make a boot SD card. I have a virtual Windows machine, but put off doing it, because it broke my sense of reality 😊 Why should I need Windows to install Linux??? Why would they not have a way to do it in Linux? 😂 Crazy talk. 😂
Hey nah no Windows needed mate, though I can see how it looks that way!
You can either grab a pre-build SD card image like I've provided, then just `dd` it onto the SD card (bootloader on SPI flash should already be burned), or you can follow their Linux and macOS instructions here: wiki.luckfox.com/Luckfox-Pico/Linux-MacOS-Burn-Image
I just use this Windows laptop as it's what I have laying out at the workshop, although I develop in Debian, and my couch laptop is a MacBook haha. Hoping that helps!
Yeah, bit Linux/Mac instructions say that this only works for SPI NAND FLASH and not tf card (sd card). I'll try the sd image link you provided, although I think I already did that, and it wouldn't boot. Thanks.
@@brettb.345 Yeah if you download the gz file, un-gzip it, then dump it on the SD like `tar -zxvf Pico_SD_Image-20231114.img_.gz -C ~ && dd if=~/img.img of=/dev/sdb bs=1M status=progress` that should do it. You are writing it to the whole card, not a specific partition, as it contains the partitions on it. Just make sure the card is... I forget, maybe 32GB or bigger? I cannot recall if I shrunk this root partition before I imaged it hah.
@@PlatimaTinkers , I'll try that when I get a chance. Thanks very much!
@@brettb.345No worries at all!