Radxa Cubie A5E Hands-On: The Budget NVMe SBC—Full Review & Results

Table of Contents

Technical Specification
My setup & first impressions (OS situation)
Thermals Performance
Benchmarks Results
1) CPU: Sysbench Test (Calculates prime numbers upto 20,000 for every 10k requests)
2) Memory Tests
3) Geekbench (CPU)
Storage & networking
Whisper on Home Assistant (Docker)
So… should you get one?
Support the Channel

In this video/article I’m sharing my experience with the Radxa Cubie A5E—a tiny SBC that surprised me with how much it packs into such a small footprint, and how it stacks up against boards like the Raspberry Pi 4 and Pi 5.

front

Buy Radxa Cubie A5E:

AliExpress Arace

Technical Specification

Category	Spec
SoC	Allwinner A527
CPU	8× Arm Cortex-A55: 4× @ 1.8 GHz + 4× @ 1.4 GHz
GPU	Mali-G57 MC1; OpenGL ES 3.2, Vulkan 1.1–1.3, OpenCL 2.2
Memory	4GB LPDDR4
External storage	microSD slot M.2 M-key (NVMe PCIe Gen 2 x1 lane, 2230 modules)
Display	HDMI 2.0 up to 4K@60fps 4-lane MIPI-DSI
Camera	MIPI-CSI: 1× 4-lane or 2× 2-lane
Networking	2× Gigabit Ethernet (one PoE-capable with add-on HAT) Wi-Fi 6 (802.11ax) Bluetooth 5.4
USB	1× USB-A 3.0 (Host) 1× USB-C 2.0 OTG (data + power)
Expansion	40-pin 2.54 mm header (UART/I²C/SPI/GPIO/PWM)
Power	5 V via USB-C or GPIO. Recommended ≥10 W baseline; ≥15 W when fully loading USB3/PCIe
OS	Debian Linux Android 13 Armbian OS

(Specs as per Radxa’s product page.) Radxa Cubie A5E

My setup & first impressions (OS situation)

I started with Armbian. It booted fine over LAN, but at the time there were a couple of missing pieces for my testing: no HDMI output and no exposed temperature sensors, which I needed for thermal checks. On the plus side, it shipped with a Linux 6.16 kernel, so I’m optimistic about where community support is heading.

Next I tried Radxa’s own Linux image (they also offer Android). The image was labeled Debian Bullseye, but the system actually identified itself as Debian 13 “Trixie” during my checks—which was pre-release at the time. Despite that, it worked surprisingly well and gave me the sensors and display I needed.

Thermals Performance

With no heatsink, the A5E idled around ~70 °C. On this board the sensors present as two groups: cpul (the 1.4 GHz cluster) and cpub (the 1.8 GHz cluster).

Since I planned to run benchmarks, I installed the metal case (with a thermal pad). Assembly took a minute—remove the side struts, slide the board in, refit the I/O side panels (Wi-Fi antenna + LAN line up nicely), then screw it all back together.

Result: idle temps dropped to ~43 °C—about a 40% reduction—and with this, I continued my tests.

Benchmarks Results

All tests were run on the 4 GB variant, fresh boot, same desk ambient, desktop still running in the background as it came built in the debian Image, accessed via SSH.

1) CPU: Sysbench Test (Calculates prime numbers upto 20,000 for every 10k requests)

Total time: ~39 s
Throughput: ~2,500 events/s
Peak temp: ~58 °C (with the case)

A touch slower than Raspberry Pi 4 in this specific test (by ~3 s), but close.

router

2) Memory Tests

MEMCPY: ~2,900 MiB/s
Block Copy (1 KiB): ~4,000 MiB/s

Better than Raspberry Pi 4 in my runs. The Raspberry Pi 5 and Cubie A5E both have a 32-bit memory bus, but Pi 5’s LPDDR4X tends to push higher transfers than LPDDR4 that the A5E and Pi 4 has.

router

3) Geekbench (CPU)

Single-core: 241
Multi-core: 1005
Power draw: ~2 W idle; up to ~6 W during multi-core phases
Thermals: peaked around ~75 °C ( with the case )

The Cubie A5E did perform better than the Raspberry Pi 4 in the multicore performance test.

Geekbench Score

router

Storage & networking

NVMe over M.2 M-key (PCIe Gen2 x1): lspci confirmed Gen2 x1; hdparm reads landed around ~366 MB/s, which is right in line with PCIe 2.0 x1 limits.
Dual Gigabit Ethernet: in repeated iperf runs, one port consistently tested a bit slower than the other. I am not sure why.

router

Whisper on Home Assistant (Docker)

One of my standard checks: Home Assistant + Whisper + Piper in Docker.
With Whisper small-int8, speech-to-text took about ~11 seconds for my sample—faster than my Raspberry Pi 4 with the same model. For voice-friendly HA automations at the edge, that’s a nice win at this size and power.

router

So… should you get one?

If you want a low-power, compact box with Wi-Fi 6, two GbE ports, and NVMe, the Cubie A5E is great for things like:

Pi-hole / AdGuard Home
Home Assistant (light-to-moderate setups)
DIY routers / multi-NIC tinkering

Just remember:

Passive cooling with the metal case is highly recommended.
PCIe is Gen2 x1—perfectly fine for snappy apps and Docker volumes, but not a storage monster.

Finally I would be still monitoring the Armbian community build for a more stable release.

router