Radxa X4 review – An Intel N100 alternative to Raspberry Pi 5 tested with Ubuntu 24.04

W’vee already looked at the Radxa X4 kit featuring an Intel N100 SBC with a design similar to the Raspberry Pi 5 and accessories including a Radxa Power PD 30W power adapter, an NVMe SSD, and a USB-C to USB-C cable in the first part of the review, before installing Ubuntu 24.04 on the board.

In the second part of the Radxa X4 review, we will test Ubuntu 24.04 in more detail with some benchmarks and power consumption measurements to show how well it works (or not) compared to a Raspberry Pi 5. We will also test the 40-pin GPIO header on the Radxa X4 controlled through a Raspberry Pi RP2040 microcontroller.

Basic system information

Ubuntu 24.04 with Linux 6.8 was installed on a 128GB SSD. We then run the inxi command to check other devices in the system.

Everything seems to be detected, and we can see that the temperature at idle is only 42°C. The system used 2.48 GB out of 8GB, and the installation of Ubuntu 24.04 required only 11.67 GB of storage.

Radxa X4 benchmarks on Ubuntu 24.04

sbc-bench.sh benchmark

Let’s start the benchmarks with Thomas Kaiser’s sbc-bench.sh script:

The full results can be found on the link just above, but a summary is available below:

As we can see below, the temperature peaked at 89°C in the single-core OpenSSL benchmark and was unusually lower in multi-core benchmarks such as 7-zip and cpuminer at around 77-78°C.

When running the same test on the UP 7000 credit card-sized fanless SBC with the same Intel N100, we found out the temperature was up to 99°C with cpuminer, so Radxa must have adjusted the system’s parameters to avoid overheating during multi-core workloads. It does show in benchmark results too, since the Radxa X4 delivered 8,120 MIPS in 7-zip, while the UP 7000 managed 12,890 MIPS on average. Quite a difference! We had similar results to the UP 7000 with the Youyeetoo X1 SBC and LattePanda Mu (best of the four in 7-zip), but those two systems run cooler as actively cooled boards. The Radxa X4 also has an external fan on top of the case, but it does not seem enough.

This is also evident from the CPU frequency set to 2400 MHz under multi-core benchmarks, but the UP 7000 was running at up to 2,900 MHz with some short drops to 1,800 MHz. One reason for the discrepancies can be power limits configuration:

The PL1 power limit (long duration) is set to 6W and the value of PL2 (short duration) is at 25W. PL2 was also set to 25W on the UP 7000, but PL1 was set to 9.125W (by default).

We can change the PL1 limit to 15W with the following command:

We test again the CPU temperature is very high, although the CPU frequency is somewhat higher at least at the beginning. That’s why PL1 is set to 6W by default.

The fan on the top of the case ran at full speed, so it looks like the cooling solution needs to be improved unlessour system had some issues.

Now that we understand the reason, let’s compare the results of memory bandwidth and 7-zip against the Raspberry Pi 5 and a Rockchip RK3588 module (Mixtile Core 3588E). The values from the Radxa X4 are memcpy: 8128.3 MB/s, memchr: 13245.3 MB/s, memset: 7960.9 MB/s, which is compared to the Raspberry Pi 5 which has memcpy: 5158.3 MB/s, memchr: 13463.2 MB/s, memset: 11671.4 MB/s. It can be seen that only the memset values are lower than the Raspberry Pi 5, but in all cases, the values are lower than the Rockchip RK3588 platform.

NVMe SSD performance

Let’s check the NVMe SSD provided by Radxa works to expectations using the iozone3 utility:

The Samsung PM991 NVMe SSD (PCIe 3.0 x4) delivered up to around 1,700 MB/s read speeds and 946 MB/s write speeds. The latter is close to the 1200MB/s theoretical write speed of the SSD, but still a way off to the 3100 MB/s read speed in this test.

We can get more information about the SSD using smartctl:

3D graphics with Unigine Heaven

We tested the internal Intel graphics with Unigine Heaven benchmark 4.0. Radxa X4 managed to render the scene at 12.4 FPS on average with a score of 311 points at a resolution of 1920×1080.

Networking testing (2.5GbE and WiFi 5)

We’ll test networking with iperf3 starting with 2.5GbE (eth0) using the LattePanda Mu board on the other side.

Radxa X4 as client:

Radxa X4 as server:

No problem in either direction with 2.35 Gbps tansfer speeds.

We’ll now switch to WiFi 5 (5GHz) testing using the pre-installed M.2 AzureWave module based on RealTek RTL8852BE using Xiaomi Router AX3200.

Client mode:

Server mode:

578 Mbps and 500 Mbps are acceptable transfer rates for WiFi 5.

Web browsing – Speedometer 2.0 and WebGL Aquarium demo

We’ll evaluate web browsing performance using Speedometer 2.0, although it’s now recommended to test with Speedometer 3.0, so that we can compare it to the Raspberry Pi 5’s results. Tests were performed on both Firefox (pre-installed) and Chromium which was manually installed.

Speedometer 2.0 shows that Firefox is slower than Chromium, with a score of 175 on Chromium, which is around 12% higher than Firefox. Firefox is 153, and when compared to the Raspberry Pi 5 board that scored 56, the Radxa X4 is around 3 times faster.

The WebGL Aquarium demo was tested on Chromium and the system managed to render 15,000 fish at 32 FPS. It’s quite better than equivalent Arm solutions, for example, an RK3588S platform could only render about 5,000 fish at a similar frame rate (31 FPS).

Video playback testing with YouTube and VLC

Let’s now test YouTube video playback in Chromium.

A 4Kp30 video could be played smoothly without any frame dropped at all.

The system struggles more with a 4K video at 60 FPS losing about 17% of the frames and encountering frequent freezes while loading.

Big Bug Bunny (H.264, Full HD 1920 x 1080) played just fine in VLC.

Another 4K video at 30FPSfps played without any stuttering, but looking at the statistics, we found that there was a frame drop of around 10%, which is quite different from the test results of other N100 CPU boards like the LattePanda Mu that had a frame drop of about 1% with the video.

Radxa X4 power consumption on Ubuntu 24.04

We measured the power consumption of Radxa X4 using a USB power meter as shown in the photo below.

Here are the results:

• Power off – 0.2 Watt
• Booting – About 12 Watts
• Idle – 6 Watts (fan active)
• 4Kp60 YouTube video playback – 12 Watts
• 7-Zip Benchmark – 12 Watts

We’ll provide a summary comparison table with benchmarks and power consumption, but we can already say the Radxa X4 consumes less than other N100 boards such as the Lattepanda Mu.

Testing the GPIO pins on the Radxa X4

The GPIO pins on the Radxa X4 are not directly connected to the N100 CPU instead we have a Raspberry Pi RP2040 MCU that controls the GPIOs and is itself connected to the Processor N100 through USB and UART. So it’s very much like having a Raspberry Pi Pico connected to a computer.

Radxa provides a pinout diagram for the 40-pin GPIO header reproduced below with each pin having multiple functions, except for the power signals.

We tested the GPIO with a Beitian BN-180 GPS receiver module connected over UART, and we opted for GPIO pins 20,21, which in “Function3” of the table above correspond to UART1 Tx and Rx pins.

Radxa recommends users program the RP2040 with the PICO C/C++ SDK, but it’s up to the user to select his preferred programming language, and in this review, we will use PlatformIO IDE and Arduino programming. We’ll start by installing VSCODE on Ubuntu and then finish installing the PlatformIO extension. We can now create a project by selecting the Raspberry Pi Pico board and then try to write a simple test code to work as a serial bridge.

We rely on three serial ports in the program:

• “Serial” for debug messages from the RP0240
• “Serial2” to connect to the GPS receiver.
• “Serialx” for the Intel CPU’s ttyS4

After that, we can build and upload files without having to press the BOOTSEL button to drag the .uf2 file. We will set the platformio.ini file as follows to use picotool as the “upload protocol” for the firmware:

After the setup is complete, we can build/upload conveniently without having to keep pressing the BOOTSEL button each time we want to update the firmware. The button location on the Radxa X4 is somewhat inconvenient because it is placed on the bottom of the case. Therefore, using this method is more suitable.

Now that the Arduino firmware is uploaded to the board, we will test whether the communication between RP2040 and N100 is correct by using a minicom to monitor the data sent between the CPU and MCU.

This will display the data read from the GPS receiver on the terminal. We can connect various devices to the Radxa X4 board easily thanks to the ability to develop code with Arduino using either PlatformIO or the Arduino IDE. The data transmission between the CPU and MCU has some limitations since it only relies on UART for communication.

Radxa X4 on Ubuntu 24.04 conclusion and Raspberry Pi 5 comparison

After having tested the Radxa X4 SBC on Ubuntu 24.04 we can compare it to the Raspberry Pi 5. There are many interesting aspects, such as Raspberry Pi 5 scoring higher in the benchmarks in memset, 7-zip, and OpenSSL AES-256 16K, which scored slightly higher. However, the web browsing experience is better on x86 than on Arm as usual. Networking performance is also better on the Radxa X4 both for Ethernet and WiFi, and the power consumption is usually lower on the Raspberry Pi 5.

Radxa X4 also scored three times higher in Unigine Heaven 3D graphics benchmarks and it supports much higher SSD read speeds.

When looking at device connectivity and general usage, these two boards are not that different. However, in terms of GPIO usage, the Raspberry Pi 5 is still more convenient because it is directly connected to the CPU without having to go through the co-processor like the Radxa X4 chose to do through the RP2040.

In terms of OS support, Windows on Arm can run on the Raspberry Pi 5 but with serious limitations (some features are not working), while it will just work right after installing on the Radxa X4. So if you have to use Windows, the Radxa X4 is the answer. But if you plan on running Linux, then it depends on your use case. The Radxa X4 is better if you need higher graphics performance and/or faster Ethernet and WiFi. Another advantage is that you can expand the board with its built-in M.2 socket without having to rely on additional HAT+ expansion boards.

While everything basically works on the Radxa X4, the main disappointment is the performance compared to other Intel N100 SBCs we’ve tested. This can be mitigated by adjusting the PL1 power limit to some extent, but then high CPU temperatures cause thermal throttling since the cooling solution does not seem to be adequate for higher power limits. [Update: based on the comments below, this seems related to a dried-up thermal pad that Radxa sent with a few boards. We’ll try to test again with another thermal pad].

We’d like to thank Radxa for sending a kit with the X4 SBC and accessories for review. The Radxa X4 is currently out of stock on AliExpress, but it can be purchased on Arace for $80 and up, and we can also see some models with 64GB eMMC flash.

CNXSoft: This review is a translation – with some additional insights – of the article on CNX Software Thailand by Arnon ThongTem and edited by Suthinee Kerdkaew.

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress