GitHunt
MA

martin-kokos/sbc-overview

Overview of current interesting consumber SBCs

Curated overview of consumer SBC (single board computers) that are interesting

This is not a comprehensive list, mainly ARM based SBCs. Hopefully at some point, they will be added to this (a bit outdated) list: https://en.wikipedia.org/wiki/User:OliverGalvin/Comparison_of_single-board_computers
For many comprehensive reviews and benchmark see review by Jeff Geerling: https://sbc-reviews.jeffgeerling.com/

SBCs

Make Model Soc RAM size Storage Notes Review Price (per RAM version)
Raspberry Pi Zero 2 W BCM2710A1 512 MB microSD by Jeff Geerling 24 EUR
Luckfox Lyra Zero W RK3506B 512 MB 256 MB flash/microSD 25 EUR
Orange Pi Zero 2W H618 1/2/4 GB microSD by Jeff Geerling 32/45/. Eur
Orange Pi 3 LTS H6 2 GB eMMC 8GB/microSD 61 EUR
Orange Pi 3b RK3566 2/4/8 GB eMMC socket/microSD 51// GB
Orange Pi 4b RK3399 2/8/16 GB eMMC/microSD
Banana Pi M5 S905X3 4GB 16GB eMMC/microSD Not available
Radxa ROCK 5C RK3588S2 2/8/16 GB SDcard/eMMC socket NPU by Jeff Geerling 89/102/136 EUR
Radxa ROCK 5C Lite RK3582 2/8/16 GB SDcard/eMMC socket 85// EUR
Raspbery Pi 5 BCM2712 4/8 GB microSD by Jeff Geerling ___/80 EUR
Radxa X4 X4 N100 4/8/12/16 GB m.2(42mm) by Jeff Geerling 70/93/114/130 EUR

Processors

Name Cores RAM type Freq GPU Notes
Rockchip RK3588 (S, S2)1 4x A76, 4x A55 Mali-G610 MC4 3x core NPU2 3
Rockchip RK3399 2 x A72, 4x A53 DDR3/4 LPDDR3/4 Mali-T864
Rockchip RK3582 2x A76, 4x A55 LPDDR4/LPDDR4x/LPDDR5 Mali-G610 MC4 (no 3D)
Rockchip RK3506B 3x A7, 1x M0 DDR3 1.2 GHz Mali G52
Broadcom BCM2710A1 4x A53 LPDDR2 VideoCore IV
Broadcom BCM2712 4x A76 LPDDR4X VideoCore V3D VII
Allwinner H618 4x A53 DDR3/4 LPDDR3/4 Mali-G31 MP2
Allwinner H6 4x A53 DDR2/3 LPDDR3/4 Mali-T720 MP2
Allwinner A64 4x A53 DDR2/3 LPDDR2/3 Mali-400 MP2 Outdated
Intel N100 4x Alder Lake-N LPDDR4/5 3.4 GHz x86-64bit
NXP i.MX 8M 4x A53, 1x M4F DDR3/4 LPDDR4 Vivante GC7000-Lite
Amlogic S905X3 4x A55 DDR3/4 LPDDR3/4 Mali G31 MP2

Cores

Name ISA
Cortext M Armv6-M
Cortex A7 Armv7
Cortex A53 Armv8-A
Cortex A55 Armv8-A
Cortex A72 Armv8-A
Cortex A76 Armv8-A
Alder Lake-N x86-64bit

Storage

microSD = has a slot for microSD card, needs to be bought extra
eMMC = has on-board soldered eMMC (optionally)
eMMC socket = has a socket for external eMMC board, needs to be bought extra

Some boards have eMMC storage directly on-board. These boards look more expensive but you don't need to buy the microSD card.
Socketted eMMC is about 64GB for 20 EUR and 256 for 45 EUR
microSD Samsung EVO A2 256 is 20 EUR
m.2 (42mm) 256GB is 40 EUR

More power

For more powerful computer systems look at the Intel N100, N150, N200 line

Nvidia boards

At this moment, I do not recommend these boards for commercial use, due to their poor bringup software (Installation wizard), poor documentation, poor support on forums and poor supplly chain security.
A good overview is on Wikipedia

Generations:
TK1, TX1, Nano = long obsolete Maxwell (2014, 2015)
TX2 = obsolete Pascal (2016)
Xavier = Volta microachitecture (2017)
Orin = Ampere microachitecture (2020)
Thor = Blackwell microachitecture (2024)

Form factors:
Nano, NX = SoDIMM modules4 usable with the Dev-Kit board
AGX = SoM with 699-pin Moles connector5

Nano

The starter board everyone usually knows is the Jetson Nano (now obsolete), and its newer variants Orin Nano and Orin Nano Super. ~15W TDP
The compute module (Tegra SoC and friends) is in a form of a DIMM module that's usually on a motherboard ("devkit") with all the peripherals broken out.
Orin Nano and Orin NX and Xavier NX are mostly compatible.6
The versions with less memory may also have less CUDA and tensor cores and slower memory.7
Other funny stuff may include things such ass; the Orin Nano USB3 ports are internally connected to a USB2 hub. USB3 could theoretically be brought out via m.2 PCIe.
Orin Nano has no HW video encoder except JPEG.
GPIO pin mapping is complicated and may need a reconfiguration via device-tree overlay generated by a official Office Spreadsheet file (which hangs LibreOffice).
The USBC port acts as a Ethernet-over-USB port and works on 192.168.55.0/24 address space. Peer IPs are 1 and 100

Xavier

Xavier NX (20W TDP) and higher spec Xavier AGX (30W TDP).8

AGX

Higher power boards such as Jetson AGX Xavier have a different form factor from the Nanos.
To compare modules, view this overview or this disgusting modal.

Thor

Upcoming (2025) series. Marketing says 2000 TFLOPS * 4bit ** sparse, so divide it by a factor of two and then another factor of two, 75-130W TDP.9

Hire me

For consultations, schedule a call via Calendly link in my profile.

References:

Footnotes

  1. S and S2 variants are just binned to a lower spec: https://www.rockchips.net/rk3588-vs-rk3588s-in-depth-technical-comparison/

  2. https://clehaxze.tw/gemlog/2024/02-14-benchmarking-rk3588-npu-matrix-multiplcation-performance-ep2.gmi

  3. https://clehaxze.tw/gemlog/2023/07-13-rockchip-npus-and-deploying-scikit-learn-models-on-them.gmi

  4. https://developer.download.nvidia.com/assets/embedded/secure/jetson/docs/Ruggedized_Systems_with_Jetson_SoDIMM_Modules_TB-10728-001-v1.1.pdf

  5. https://www.molex.com/en-us/products/part-detail/2034560003

  6. https://developer.download.nvidia.com/assets/embedded/secure/jetson/orin_nano/docs/Jetson_OrinNano_OrinNX_XavierNX_Interface_Comparison_Migration_DA-11081-001_v1.1.pdf

  7. https://www.seeedstudio.com/blog/nvidia-jetson-comparison-nano-tx2-nx-xavier-nx-agx-orin/

  8. https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-xavier-series/

  9. https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/