Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Radxa X4 low-cost, credit card-sized Intel N100 SBC goes for $60 and up (cnx-software.com)
86 points by todsacerdoti 11 months ago | hide | past | favorite | 55 comments



> 2.5GbE and WiFi 6, M.2 SSD support is built-in and four to eight times faster compared to PCIe HAT for the Pi 5, and the USB 3.2 ports are capable of 10 Gbps speed.

For $60 that seems pretty killer. And I assume x86 means less "idiosyncratic" software. I wonder if its WiFi supports AP mode; this might be a nice little WAP.

Edit: I really want to see pricing for 16GB RAM; this could be a good desktop!

Edit 2:

    WiFi 5 & BT 5 for 4GB RAM model
    WiFi 6 & BT 5.2 for 8GB RAM or more models
Small caveat. Still great.


>And I assume x86 means less "idiosyncratic" software.

It does. My N100 mini PC runs plain ol' LMDE. Intel chipset/GPU/etc drivers are and have been rock solid on Linux since forever. Very much a "just works" situation


This looks like my next router with OpenWRT. I've had a Pi4 running OpenWRT great, but I love how things just work on x86, and an M2 SSD would be a big upgrade from microSD. The 10gbps USB ports are especially good for USB ethernet adapters >1gbps. I wouldn't use the wifi as I use 'dump APs' for the wireless part of my network.

I'm curious how the RP2040 shows up when running Windows, even if that isn't what I'd do.


But it's not $60, It's $74 + $78 in shipping + taxes and fees.


That appears reasonably good. It's a shame ARM SBCs main Achilles heel has always been storage ( and networking a bit, and definitely bootloaders and OS support., )

Word of caution: These things never get acquired for that price. When everything is said and done, the real total price is closer to double. I'm talking about supporting knick-knacks, import fees, shipping fees.

I recently ordered a ODROID-H4 PLUS ( N97, not N100 ) and it's looking good, but it's more upmarket than this SBC.


N97 is better than N100 (it is faster and it supports in-band ECC), but it has a higher power consumption.

ODROID H4+ is faster than Radxa X4, due to the higher clock frequency and due to having a 64-bit DRAM interface, while Radxa X4 has a 32-bit DRAM interface.

Radxa X4 is cheaper, when considering the price including the DRAM & WiFi (under $100 with 8 GB DRAM), and it is much smaller, having the credit card form factor. An advantage of Radxa X4 over almost all other small SBCs is that its USB 3 ports are 10 Gb/s ports, allowing fast links to USB hubs or external SSDs or USB Ethernet interfaces.

When compared to the SBCs using RK3588 or to Radxa NIO 12L, the only cheap SBC with a CPU including a quadruple Cortex-A78 (from Mediatek; comparable SBCs from Qualcomm or NVIDIA are very expensive, starting around $500), Radxa X4 is cheaper than most of them and whether it is faster depends on the application. The N100 CPU will be faster in single thread, but it can be slower in multiple-thread applications, when, depending on cooling, its clock frequency may drop to lower values than for the SBCs using Cortex-A76 or Cortex-A78. Many of those SBCs use 64-bit DRAM interfaces, so they may also be faster for programs that happen to be limited by the memory bandwidth (vs. Radxa X4 with a 32-bit DRAM interface). The GPU of N100 is also much faster and it has much better software support than the Arm GPUs. USB and PCIe should also work better on N100.

There is also AAEON UP 7000, which is a very similar credit-card-sized SBC with Alder Lake N, but it is 2 to 3 times more expensive.

This year, Intel has introduced a refresh of Alder Lake N, with the code name "Amston Lake". The Amston Lake CPUs are branded in the Atom x7000 series and they are intended for embedded applications. Radxa has said that there will be future variants of the Radxa X4 which will use some Amston Lake models instead of N100, including 8-core models.


It's been quite a while since I saw the announcement of a new Linux-compatible SBC and thought, "this is really cool."

This is really cool.


In the schematic on Radxas site[1] there is no connection between the RP2040 and the power circuit, I would love to be able to shutdown the Intel chip from the Raspberry Pi microchip. Seems far fetched that it would be hackable if this is not a feature, I'm guessing the power chips are a lot more complicated and undocumented than a RP2040.

[1] page 22 https://dl.radxa.com/x/x4/radxa_x4_v1.11_schematic.pdf


You do not need a hardware means for the RP2040 to shutdown the SBC.

You can run on the N100 a program that would monitor continuously the RP2040, e.g. by reading periodically the state of a GPIO pin or of a variable in the internal memory. Such a program can shutdown all the SBC.


I suppose the usecase would be a "poor man's IPMI", a separate chip dedicated to monitoring the main system, with the ability to power if off and on.

Sadly most(?) x86 systems don't offer a non-visual way to access BIOS/UEFI settings (like ARM boards usually do with uBoot over serial), so the utility of such monitoring without video input is limited, but just being able to remotely reset a system via a built-in chip would be very useful.


While this is true, it misses most usecases, since you need cooperation from the Intel chip and you cannot start it up again after it shut down


This little PC looks awesome. I have a little Beelink N100 PC with 16GB of RAM and a 512GB NVME SSD, thing cost like $160 and sips power, even when transcoding 4K video for my Jellyfin server.

The fact that these are available with 4 and 8 GB of RAM now and with Pi-compatible GPIO could mean some excellent home project opportunities.


When you say "sips power", what exactly do you mean?

I've been watching for an SBC you could use to make something like a TRS-80 Model 100. They ran off 4-AA batteries and would get you 20 hours of continuous use. That translated to at least a week of real world use (you wouldn't want to be typing on one for 4 hours a day...)

That was in 1983. Today, it seems like you should be able to make something similar that will run for weeks or months on AA batteries.


I don't think AA batteries have improved a whole lot over the past 40 years, but power requirements have definitely gone up. The linked SBC needs a 12v/2A power supply minimum. I think the CPU in the Model 100 used something like 1 watt max and even then I recall reading somewhere that it went as far as suspending clock cycles in between key strokes to conserve energy (this may have been something else, I can't find where I read it).

So in my very-much-not-an-expert opinion, I don't think it would be feasible to run a modern computer off of AA batteries, let alone for weeks.


The minimum power supply is dimensioned to allow for USB powered peripherals.

The idle power consumption of such a SBC like Radxa X4 must be significantly below 5 W, probably around 2 to 3 W.

That is still too high for standard AA batteries, but rechargeable batteries should work.

If a long battery life would be needed, it is likely that one can shut down the N100 CPU and wake it up either periodically or on USB activity caused by the RP2040 MCU. That should reduce the average power consumption to fractions of a watt.


To run on AA batteries you'd probably have to go to microcontrollers like ESP32 which are way less powerful than that Intel device but also way more powerful than TRS-80. Assuming a power consumption of 0.25W for that SoC and basic 2000mAh AA batteries, a 4 AA battery setup could run the device for 48 hours.


There are 300mW class Linux-capable chips like SAMA5D27. 500Mhz Arm Cortex-A5.

There are all kinds of Linux Capable chips available between 0.3W to 5W, and with LPDDR1 or other older standards, you can get very good and cheap prices while still only sipping power.


My n100 system idles at 10W from the wall. It has many more components to power (2x m.2, sata, 5x nics, etc.) than this SBC though.

Much higher than a raspberry pi, but still low in comparison to any desktop server


That is quite high for an N100 system, even if it is not uncommon and much of the power may be used by the listed peripherals.

Most well designed N100 systems that have only a moderate number of peripherals have an idle power consumption under 5 W.


10W? I think that means 4 AA batteries would keep it running for about an hour.


Does it have to be an SBC?

A modern MCU will absolutely run circles around the processing and memory capabilities of the Model 100, and can probably even do that for 20 hours using four AA batteries.


There a lot of applications for which a MCU is sufficient and it will be much cheaper and it will have a much lower power consumption.

However not even the fastest existing MCUs, i.e. those with Cortex-M85 cores, can reach processing or memory speeds comparable with CPUs like the Intel Alder Lake N series or the Arm CPUs using Cortex-A78 or Cortex-A76 cores.

The fastest MCUs may have a clock frequency of up to 1 GHz, but most of them have clock frequencies many times lower (this is because the MCU cores use shorter pipelines), while the SBCs use clock frequencies between 2 and 4 GHz.

The Cortex-M7 or Cortex-M85 MCU cores have an IPC (instructions per clock cycle) that is 2 to 3 times higher than the IPC of most other MCU cores, but even their IPC is 2 to 3 times lower than of Alder Lake N, Cortex-A78 or Cortex-A76. Due to the much higher clock frequency and IPC, the SBCs are much faster than any MCU.

Due to the MCU clock frequencies being under 1 GHz, not even their cache memories can have throughputs as high as the DDR memories of the SBCs, which use 4.267 to 4.8 giga transfers per second.

So any modern SBC will run circles "around the processing and memory capabilities" of any MCU. However, you are right that in many cases the speed of a SBC is not needed and a MCU is good enough.

A SBC is typically needed when you want a USB 3 interface or an SSD, because extremely few MCUs have any peripheral interface faster than 1 Gb/s Ethernet.


Correct.

But in context: It simply can't take very much grunt to provide a modern take on a ridiculously-limited portable computer from 1983.

The TRS-80 Model 100 that is the context here had an 80C85 processor running at a scalding 2.4MHz, and featured as much as 24 kilobytes of SRAM.

And that's a pretty excellent spec for a portable computer in 1983, but it's a complete snoozefest compared to a RP2040 dev kit like the Pi Pico (at a cost of around three US dollars).


I am pretty sure RP2040 is not compatible with the Raspberry pi boards, so it will not be enough to replace Raspberry PI.


RP2040 is just a microcontroller vs the "full computer" of a normal pi board. This is the distinction people are usually making when they say RP2040 isn't a pi.

This board is a full intel computer plus a rp2040 driving the pi-style GPIO pins. It should basically be able to do whatever a pi can do.


To be compatible and be able to use all the Pi Hats would be nice but my point is that I am not sure a RP2040 can replace the Pi and be called "compatible". You can do a lot with an MC on the board that is true.


And much more…


The embedded RP2040 is really interesting. I wonder what the CPU<->MCU latency looks like. You can pull off some really neat stuff with very tight timing using that chip.

Also an interesting addition for a platform that is clearly meant to compete against RPI in general.


Looks like just an UART connected between the N100 and the RP2040.


Yep. The whole point is that the GPIOs are owned by the RP2040 and you use it as a I/O coprocessor.


There is also a USB link. It is likely that this can access the debug interface of the MCU, which is needed to reprogram it.


Does anyone use a solar-powered remote device like this? What is low enough to run off battery for a while?


We live in a world where fully-featured homes full of tech are run from solar and battery power.

A person can run any computer they wish from solar and battery, given sufficient quantities of panels and cells.


Has anyone done good power measurements of similar boards?

How many watts does it draw in idle, surfing the web, and watching a YouTube video?


A typical chinese N100 minipc takes from 5w at idle to 20W at full load.


What does the cooling look like on this? Guessing it's not suitable for full passive?


You can do full passive, but you need a quite clunky heat sink.

My home router is a $80 N100 + 4x 2.5GbE box, passively cooled.


Where did you buy it? Most of the router boxes I've seen are closer to 150-200?


Ah, bought it while I'm in China. Just checked Amazon and even aliexpress and you are right, it's 150-200. Didn't realize it could differ by 2x.


The article shows a related model with a passive heatsink bolted onto it. The heatsink roughly doubles the volume of the thing, but passive seems possible. If nothing else, the spec sheet says it can go up to 60C.


My only wish is that they had put an oculink port on it.


It has PCIe 3.0 x4 on the M.2 port.

You can buy a cheap OCuLink adapter and and put it in that port.


Good point. I was thinking the N100 had more pcie lanes than it actually does.

It only has 9. shrug


AliExpress seems to be down, at least the mobile app.


Unfortunately it seems like it is already sold out.


So it even has Raspberry-Pi-like availability? They really thought of everything!


Hopefully no nasty surprises in firmware.


Already out of stock?


Link in TFA goes to AliExpress. This is the actual vendor: https://radxa.com/products/x/x4/


If you scroll down to the Buy Now you will see that Radxa also links to Aliexpress as the only distributor.


Officially that may be the case, but arace.tech claims to have 8GB versions in stock, and they seem to have been an official partner of Radxa in earlier launches.


aliexpress had some in stock during the night. It's now at "14 sold".

"there's dozens of us"


After the experience I had last month with RPi, I hope this will be the end of RPi.

Note: the problem I had, is I had a system with an old Rpi, I did made a physical clone, new board, copy the SW, but on a new OS (bookworm) and surprise! Nothing works!!! Raspian is a chaos, and using systemd in an embedded system is just Not For Me!


I do that all the time, I've ran the same image on a raspberry 1 through 5.


The problem was not the HW, but the OS update.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: