Best Single Board Computers (SBC) for A.I projects

[@Electromakerio]

If you enjoyed this video, then please support our channel by visiting our online store. We stock hundreds of single-board computers and ship worldwide: www.electromaker.io/shop/category/single-board-computers

[@andrelaurin540]

I like your presentation, but I don't understand why you did not consider some of the SBC's with built in NPUs? Like Orange PI 5, Nanopc-T6 etc...

[@danielpicassomunoz2752]

Please sir, can you list some more? (On the cheaper, but software supported side)

[@Jp-ue8xz]

because this video isn't about AI oriented SBC but just about shitty overpriced boards which don't even have competitive hardware for the task (w/ like 2 exceptions that are still grossly overpriced)

[@Electromakerio]

Thanks for your feedback! You're right; SBCs with built-in NPUs like the Orange Pi 5 and NanoPC-T6 are excellent choices. We'll make sure to include more options with NPUs in future presentations. Appreciate your input!

[@Electromakerio]

@@Jp-ue8xz Our intention was to highlight a range of options available for different needs and budgets. We'll definitely take your comments into account for future content and strive to include more competitively priced and AI-focused SBCs. Thanks for watching and sharing your thoughts!

[@jstro-hobbytech]

​@@Electromakeriothe fact current market prices or which os version runs the best is one of many of the flaws of this thinly veiled excuse to funnel people to your overpriced website. Electromaker..... people have to give it up with the anyone can be anything but buy it from us. Im on a disability pension and i bet ive donated more gear and starter kits than you have. I do it anonymously as well. No one even carries the original maxwell based jetson anymore with 4gb of ram. Perfect for teaching a robotics program to children. Ive been trying to get 2 cheap jetsons to learn ros and develop a curriculum for a local non profit for years. Even nvidia whom ive given over 50k cad to since the 90s wouldnt even throw me a deal. I got spoken down upon which sickened me. I can provide receipts for the thousands ive donated in components and pi variants, scopes, soldering stations, ul certified gear but refusing to pay over 4k to get a non profit license in canada just just to try to catch a break on costs (hopefully) defeats the purpose. Ive spent hundreds of hours scouring websites like your's for programs to get a discount to provide underprivileged people and fellow first responders and veterans with the empowerment of learning technology because they didnt get an engineering degree like me before i decided id rather be a first responder. Ive given the last 8 years and over half of my disposable income to a futile cause at my own expense so i gave away my last 600 dollar beginner lab kit away just this week and decided to just get a second engineering degree at 46 and enter the workforce again because ive completely failed at my goal. So many people need to be exposed to the joy of stem topics that can inspire them in their youth or help a combat veteran find purpose that isnt at the bottom of a bottle. My name is joey strong and i live in florence nova Scotia 902-578-1556. Put your money where your mouth is and donate a few low end sbc boards and know youve helped people who dedicated their lives to helping you. Shit, id even pay cost for a couple xavier nanos and lidars because i have the rest of the stuff. I can prove it, but ive completely lost faith in every entity who wont eat a tax write off and save a few souls. Proof, on demand. I dont lie. Ut when pwoplw who have the least have to give the most and just want to have some faith in humanity restored get fed a diet of obfuscated defelection by way of word salad places like your's unfortunately are twught to do. It's not about enabling the ext generation or helping the current one, its about charging insane markups and filing taxes in countries the companies didnt even originate in to avoid paying taxes and instead taking tax rebates because they created some jobs where people have forgotten what inspired them and the joy of sharing it with other's without the means. Prove me wrong. Be the company that restores my faith in caring about the future generations and empowering currently disenfranchised sould to find new meaning in learning. Love for engineering is infectious but any good will is marred i beaurocratic double-speak born of apathy. It's profoundly saddened me to have to stop helping people, 99 percent of them strangers who i shared an email or 2 with then backed off hopeful that they did something or bettered themselves or could even help tutor their children in subjects as simple as university prep math. I've given a similar treatise to this one so many times but this is my last as i need to care about myself if no one can trust me enough to believe in generosity born of wanting something better for people. Im a professional guitar player but i spend more time reading, making circuits, learning math. It cured my ptsd, made life exciting again and hoping i put someone one on a new path. I did most of it anonymously so ill never really know and there's nothing where i live witb people who feel like me. There's a 20 million dollar makerspace near me that is one of the best in the country ran by people who don't know what they got into after going on a massive shopping spree. I built out their electronics area then said goodbye but i fear ill be at a liquidation sale buying the awesome vintage test gear thats never been used or even turned on in some cases. They have a rare bk precision psu thats never been turned on since before 2018 and buying it will be bitter sweet bht at least ill give away my 3 channel siglent one to someone in it's place i guess.

[@jackspdr]

Good list, but you missed the Coral Dev Board with the TPU

[@jackflash6377]

You glossed over the biggest advantage of the Jetson. Cuda cores!! and an 8 core ARM CPU. Made for machine learning.

[@Electromakerio]

Great point! The Jetson's CUDA cores and 8-core ARM CPU are indeed its biggest advantages, making it a powerhouse for machine learning tasks. Thanks for highlighting this!

[@JimPlant]

Love the summary between 4:30 and 5:00 . Exactly what we found building a lego robot with Raspberry Pi. It worked but was very 'laggy' and we ended up omitting image processing (in the cloud) largely due to privacy concerns.

[@Electromakerio]

Building a Lego robot with a Raspberry Pi sounds like a fun and educational project! It's a great example of how SBCs like the Raspberry Pi can open up possibilities for creative DIY projects, even with some limitations such as processing lag, especially when handling compute-intensive tasks like image processing.

[@lukassembol7542]

What about KHADAS VIM3 Pro? It is a three years old board, but with a $139 USD price tag, dedicated 5 TOPS NPU, 4*A73@2.2GHz + 2*A53@1.8GHz, M.2 NVMe slot, 4-lane MIPI-CSI header, programmable MCU and integrated 3-axis accelerometer, there is a ton of mobile/EDGE applications where it is the best suited board there is (given any Jetson boards may be unnecessary overkill and overexpense in such applications) ;-)

[@Electromakerio]

Great point! The KHADAS VIM3 Pro is definitely a powerhouse for its price, especially with its dedicated NPU and versatile features like the M.2 NVMe slot and programmable MCU. It's an excellent choice for mobile and edge applications where you need strong performance without the overkill of something like a Jetson board. Its balance of power and cost makes it a strong contender, especially for projects that require a bit more than a typical SBC but don’t need the full capability (and cost) of higher-end solutions. Definitely still a relevant and valuable board!

[@Flare1107]

I see the Radxa rock5 is mentioned. It was what I was considering as a contender for my second suggestion, the Coral Dev board. Trying to build my own loona robot and I found the same specs listed from the loona and coral dev board. But I'm more interested in your comparison for a Compute modules line of boards by raspberry pi, orange pi, and radxa. (I'm sure there's another)

[@Electromakerio]

Thanks for sharing your insights! The Radxa Rock5 and Coral Dev Board are both solid choices. We'll definitely consider doing a detailed comparison of Compute Modules from Raspberry Pi, Orange Pi, Radxa, and other manufacturers in future videos. Your project sounds exciting-best of luck with your Loona robot build!

[@aaronslife4635]

This is purely an advertisement. No benchmarks of common use cases etc. -1

[@Octo8Brain]

Great review! I think it would be also worth considering other SBCs like the Khadas Edge2, Radxa Rock 5B or OrangePi 5 for example. They used the Rockchip RK3588 SoC with 8-core 64-bit ARM processors and a 6 TOPS NPU, and they appear to be more powerful than the Raspberry Pi 5, at least on paper. They can run real-time AI applications like Whisper for speech recognition or even small LLMs and could be great for building small personal AI assistants.

[@Electromakerio]

Thank you for the insightful addition! You're absolutely right; the Khadas Edge2, Radxa Rock 5B, and OrangePi 5 represent exciting alternatives to the Raspberry Pi, especially for users who require more processing power for AI-driven projects. The Rockchip RK3588 SoC they're equipped with is indeed a potent piece of hardware, combining 8-core 64-bit ARM processors with a 6 TOPS (trillion operations per second) NPU (Neural Processing Unit). This makes them especially suitable for demanding AI tasks.

[@Octo8Brain]

@@Electromakerio Indeed, and those SBCs are already more than 1 year old, new ones have been released since, for example: Radxa Rock 5C, Orange Pi 5 Pro (both with the same form factor as the Pi 5) or the Radxa ZERO 2 Pro (6-core 64-bit ARM, 5 TOPS).

[@MrKelaher]

Kendryte K210 based boards for machine vision applications. M5 have one for example - M5StickV K210 AI Camera

[@Electromakerio]

Thanks for the suggestion! The Kendryte K210-based boards, like the M5StickV K210 AI Camera, are great for machine vision applications. We'll look into covering these in future videos. Appreciate your input!

[@Iswimandrun]

Pi 5 with a Coral TPU over the pcie and booting from USB 3.0 with a 2.5" SATA SSD to USB converter. I did it just had to build some debian packages from source, mess with the device tree, enable the pcie port via some config file edits, install some dependencies, install pyenv and us it to make the right python version from source configure your system to use that version then install the python packages. There might of been some udev changes or I might be mis remembering from a different project anyway look it up it's possible. I used a pineberrypi pcie board to interface with the Coral TPU.

[@Electromakerio]

That's an impressive setup you've managed with the Pi 5, especially with the integration of the Coral TPU via PCIe. Building Debian packages from source and tweaking the device tree definitely requires a good understanding of the system's inner workings! The use of a pineberrypi PCIe board for interfacing with the Coral TPU sounds like an interesting solution for enhancing AI capabilities, thanks for sharing your experience and the steps involved. We would love to see a project write-up submitted to our project hub 😃

[@Iswimandrun]

@@Electromakerio sure where is this hub. I did it for work originally but. I just got a duplicate set of the bill of materials to do it for myself so I can have two units one in the office and one at home to work into a testing automation project I am spearheading.

[@Iswimandrun]

So I got the kit to do it again and I will ensure my write-up is as complete as I can.

[@Iswimandrun]

Your site's contact me function is hard to use with my DNS filtering maybe, idk didn't check what was getting blocked but the behavior was the thing that tries to determine if I am not a bot keeps not connecting to ask me the question oh well I will try tomorrow.

[@Electromakerio]

Here: www.electromaker.io/projects

[@jacksimba]

Incredible explanations and context! Thanks !

[@Electromakerio]

Thank you so much! I'm glad you found the explanations helpful. Your support means a lot!

[@jolianWassof]

Why you dont talk about orange pi 5 plus

[@Electromakerio]

Great point! The Orange Pi 5 Plus is a solid board with impressive specs. We’ll look into including it in future videos. Thanks for the suggestion!

[@osbert]

13:25

[@Nathan-q6y]

Radxa X2l would have made it to my list definitely however I do get that not too many others are so lucky to actually possess one of these beautiful pieces of tech.. 🎉

[@Electromakerio]

Thanks for the suggestion! The Radxa X2l is indeed a fantastic piece of tech. We'll definitely consider including it in future videos. Appreciate your input!

[@MyPhone-qg2eh]

Just buy 10 pi5s

[@Electromakerio]

Buying 10 Raspberry Pi 5 units is quite the commitment! The Pi 5, with its enhanced performance and capabilities, offers a lot of potential. I'm curious about what you're aiming to achieve with such a setup.

[@robinmitchell6803]

First

[@richelliot99]

Forth!

[@midnightmoves7976]

nice work thanks, and you said "further more" and not more over. happy days.

[@Electromakerio]

Thanks for the kind words! And good catch-I'll keep that in mind for next time. Happy days!

[@spacetechtips]

What about a mini PC with a 4060?

[@Electromakerio]

Using a mini PC equipped with an NVIDIA GeForce RTX 4060 (assuming you're projecting into the future as the 4060 model specifics might not align with current offerings as of my last update in April 2023) can be an excellent choice for a wide range of applications, particularly those requiring substantial graphical processing power and AI capabilities.

[@chetana9802]

any of your videos considered ollama or llamafile on these wide range of devices?

[@Electromakerio]

Thanks for the suggestion! We haven't covered Ollama or Llamafile yet, but we'll definitely consider exploring them in future videos. Appreciate your input!

[@chetana9802]

Awesome to hear that! Even ARM Ampere has Llama 3.1 demo in the works ​@@Electromakerio

[@Electromakerio]

@@chetana9802 Interesting. That suggests that high-performance ARM chips are being recognized for their ability to handle complex AI workloads. It makes you wonder how long x86 architecture and GPUs will be the most dominant in the market

[@chetana9802]

@@Electromakerio nothing beats ARM with efficiency

[@Electromakerio]

@@chetana9802 x86 processors often deliver higher raw performance and are optimized for tasks requiring heavy computation and compatibility with legacy software. The choice between ARM and other architectures often depends on the specific use case, so while ARM shines in efficiency, other factors like performance, software support, and ecosystem might make a different architecture a better choice in certain scenarios.

[@乾淨核能]

why use SBC?why not use NV GPU?

[@Electromakerio]

SBCs are great for portable, cost-effective, and energy-efficient applications with moderate computing needs. On the other hand, NVidia GPUs are better suited for high-performance computing tasks, especially those related to AI, machine learning, and advanced graphics processing. In some cases, a combination of both might be the best approach, utilizing SBCs for control and low-power tasks while leveraging NVidia GPUs for heavy lifting.

[@乾淨核能]

@@Electromakerio wow! thank you for detailed explainaton!!

[@Jp-ue8xz]

because there's an awful thing called an electricity bill

[@乾淨核能]

@@Jp-ue8xz thank you!

[@Electromakerio]

@@Jp-ue8xz stop being such a negative nancy 😅

[@malindailankoon]

❤

[@Jp-ue8xz]

Awful vid, it's just about shitty overpriced boards which don't even have competitive hardware for the task (w/ like 2 exceptions like nvidia jetson that are still grossly overpriced). You can find better arm boards with 16gb and even 32gb ddr4 ram, WITH an npu/tpu/mma/accelerator whatever, for anywhere between half and 80% the price of the stuff presented here

[@Electromakerio]

The goal of this video was to provide a broad overview of some of the popular and versatile options available today, taking into account not just hardware specs but also factors like ease of use, community support, and overall value for AI projects. While it's true that some boards might be more expensive, they often offer additional features and support that can be valuable depending on the specific use case. We always welcome suggestions for better alternatives, and if you have specific boards in mind that you believe offer superior performance at a better price, please feel free to share them!