Radxa Cubie A7Z #100
Description
Basic information
- Board URL (official): https://docs.radxa.com/cubie/a7z
- Board purchased from: TODO
- Board purchase date: TODO
- Board specs (as tested): TODO
- Board price (as tested): TODO
Linux/system information
# output of `neofetch`
_,met$$$$$gg. radxa@radxa-cubie-a7z
,g$$$$$$$$$$$$$$$P. ---------------------
,g$$P" """Y$$.". OS: Debian GNU/Linux 11 (bullseye) aarch64
,$$P' `$$$. Host: sun60iw2
',$$P ,ggs. `$$b: Kernel: 5.15.147-12-a733
`d$$' ,$P"' . $$$ Uptime: 1 hour, 30 mins
$$P d$' , $$P Packages: 1816 (dpkg)
$$: $$. - ,d$$' Shell: bash 5.1.4
$$; Y$b._ _,d$P' Resolution: 3456x1080
Y$$. `.`"Y$$$$P"' Theme: Adwaita [GTK3]
`$$b "-.__ Icons: Adwaita [GTK3]
`Y$$ Terminal: /dev/pts/1
`Y$$. CPU: (8) @ 1.794GHz
`$$b. Memory: 1230MiB / 7934MiB
`Y$$b.
`"Y$b._
`"""
# output of `uname -a`
Linux radxa-cubie-a7z 5.15.147-12-a733 #12 SMP PREEMPT Tue Dec 9 07:24:08 UTC 2025 aarch64 GNU/Linux
System topology
Benchmark results
CPU
- Geekbench 6: (TODO single / TODO multi - PASTE_URL)
- TODO Gflops (geerlingguy/top500-benchmark HPL result)
Power
- Idle power draw (at wall): TODO W
With WiFi: 1.6W - Maximum simulated power draw (
stress-ng --matrix 0): 5.6 W - During Geekbench multicore benchmark: 4.3 W
- During
top500HPL benchmark: TODO W
Disk
MANUFACTURER_AND_MODEL_OF_DISK_HERE
Kingston SD Canvas Go! Plus(64G)
| Benchmark | Result |
|---|---|
| iozone 4K random read | 11.06 MB/s |
| iozone 4K random write | 11.10 MB/s |
| iozone 1M random read | 78.74 MB/s |
| iozone 1M random write | 46.45 MB/s |
| iozone 1M sequential read | 78.78 MB/s |
| iozone 1M sequential write | 60.03 MB/s |
wget https://raw.githubusercontent.com/geerlingguy/pi-cluster/master/benchmarks/disk-benchmark.sh
chmod +x disk-benchmark.sh
sudo MOUNT_PATH=/ TEST_SIZE=1g ./disk-benchmark.sh
Run benchmark on any attached storage device (e.g. eMMC, microSD, NVMe, SATA) and add results under an additional heading.
Network
iperf3 results:
iperf3 -c $SERVER_IP: 231 Mbpsiperf3 -c $SERVER_IP --reverse: 297 Mbpsiperf3 -c $SERVER_IP --bidir: 234 Mbps up, 196 Mbps down
(Be sure to test all interfaces, noting any that are non-functional.)
GPU
glmark2
glmark2-es2 / glmark2-es2-wayland results:
=======================================================
glmark2 2023.01
=======================================================
OpenGL Information
GL_VENDOR: Imagination Technologies
GL_RENDERER: PowerVR B-Series BXM-4-64
GL_VERSION: OpenGL ES 3.2 build 24.2@6603887
Surface Config: buf=32 r=8 g=8 b=8 a=8 depth=24 stencil=0 samples=0
Surface Size: 800x600 windowed
=======================================================
[build] use-vbo=false: FPS: 457 FrameTime: 2.189 ms
[build] use-vbo=true: FPS: 514 FrameTime: 1.947 ms
[texture] texture-filter=nearest: FPS: 564 FrameTime: 1.774 ms
[texture] texture-filter=linear: FPS: 539 FrameTime: 1.856 ms
[texture] texture-filter=mipmap: FPS: 565 FrameTime: 1.772 ms
[shading] shading=gouraud: FPS: 432 FrameTime: 2.319 ms
[shading] shading=blinn-phong-inf: FPS: 436 FrameTime: 2.298 ms
[shading] shading=phong: FPS: 392 FrameTime: 2.556 ms
[shading] shading=cel: FPS: 381 FrameTime: 2.630 ms
[bump] bump-render=high-poly: FPS: 260 FrameTime: 3.854 ms
[bump] bump-render=normals: FPS: 588 FrameTime: 1.703 ms
[bump] bump-render=height: FPS: 566 FrameTime: 1.768 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 386 FrameTime: 2.592 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 207 FrameTime: 4.836 ms
[pulsar] light=false:quads=5:texture=false: FPS: 659 FrameTime: 1.519 ms
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: FPS: 181 FrameTime: 5.528 ms
[desktop] effect=shadow:windows=4: FPS: 461 FrameTime: 2.172 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 161 FrameTime: 6.214 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 162 FrameTime: 6.192 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 262 FrameTime: 3.825 ms
[ideas] speed=duration: FPS: 568 FrameTime: 1.762 ms
[jellyfish] <default>: FPS: 275 FrameTime: 3.641 ms
[terrain] <default>: FPS: 28 FrameTime: 37.021 ms
[shadow] <default>: FPS: 301 FrameTime: 3.327 ms
[refract] <default>: FPS: 53 FrameTime: 18.995 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 652 FrameTime: 1.535 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 448 FrameTime: 2.234 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 621 FrameTime: 1.611 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 518 FrameTime: 1.931 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 391 FrameTime: 2.562 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 530 FrameTime: 1.888 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 537 FrameTime: 1.865 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 510 FrameTime: 1.961 ms
=======================================================
glmark2 Score: 411
=======================================================
vkmark
vkmark results:
1. Install vkmark with `sudo apt install -y vkmark`
2. Run `vkmark` (with `DISPLAY=:0` prepended if running over SSH)
3. Replace this block of text with the results.
Note:
vkmarkneeds to be compiled from source on Debian 12 and earlier.
GravityMark
GravityMark results:
1. Download the latest version of GravityMark: https://gravitymark.tellusim.com
2. Run `chmod +x [downloaded_filename].run`
3. Run `sudo ./[downloaded_filename].run` and press `y` to accept the terms.
4. Open the link it prints, and run the Benchmark defaults, changing to 720p resolution and 50,000 asteroids.
Note: These benchmarks require an active display on the device. Not all devices may be able to run glmark2-es2, so in that case, make a note and move on!
AI / LLM Inference
Basic AI inference results:
# Install ollama
curl -fsSL https://ollama.com/install.sh | sh
# Download some models
ollama pull llama3.2:3b \
&& ollama pull llama3.1:8b \
&& ollama pull llama2:13b
# Download the benchmarking script
git clone https://github.com/geerlingguy/ai-benchmarks.git
cd ai-benchmarks
# Run benchmark on multiple models
declare -a models=("llama3.2:3b" "llama3.1:8b" "llama2:13b")
for i in "${models[@]}"; do ./obench.sh -m "$i" -c 3 --markdown; done
Note that Ollama will run on the CPU if no valid GPU / drivers are present. Be sure to note whether Ollama runs on the CPU, GPU, or a dedicated NPU.
Memory
tinymembench results:
Click to expand memory benchmark result
tinymembench v0.4.10 (simple benchmark for memory throughput and latency)
==========================================================================
== Memory bandwidth tests ==
== ==
== Note 1: 1MB = 1000000 bytes ==
== Note 2: Results for 'copy' tests show how many bytes can be ==
== copied per second (adding together read and writen ==
== bytes would have provided twice higher numbers) ==
== Note 3: 2-pass copy means that we are using a small temporary buffer ==
== to first fetch data into it, and only then write it to the ==
== destination (source -> L1 cache, L1 cache -> destination) ==
== Note 4: If sample standard deviation exceeds 0.1%, it is shown in ==
== brackets ==
==========================================================================
C copy backwards : 1930.4 MB/s (1.0%)
C copy backwards (32 byte blocks) : 1927.6 MB/s (1.5%)
C copy backwards (64 byte blocks) : 1938.7 MB/s (0.7%)
C copy : 3735.1 MB/s (0.5%)
C copy prefetched (32 bytes step) : 1760.6 MB/s (1.0%)
C copy prefetched (64 bytes step) : 3687.9 MB/s (1.0%)
C 2-pass copy : 2119.8 MB/s (0.3%)
C 2-pass copy prefetched (32 bytes step) : 1361.4 MB/s (0.8%)
C 2-pass copy prefetched (64 bytes step) : 5372.3 MB/s (34.5%)
C fill : 8459.6 MB/s (0.5%)
C fill (shuffle within 16 byte blocks) : 8457.8 MB/s (0.5%)
C fill (shuffle within 32 byte blocks) : 8463.3 MB/s (0.7%)
C fill (shuffle within 64 byte blocks) : 8063.6 MB/s (0.8%)
NEON 64x2 COPY : 5274.0 MB/s (0.9%)
NEON 64x2x4 COPY : 5277.1 MB/s (0.6%)
NEON 64x1x4_x2 COPY : 1853.4 MB/s (0.4%)
NEON 64x2 COPY prefetch x2 : 4529.7 MB/s (0.7%)
NEON 64x2x4 COPY prefetch x1 : 4817.2 MB/s (0.5%)
NEON 64x2 COPY prefetch x1 : 4551.9 MB/s (0.4%)
NEON 64x2x4 COPY prefetch x1 : 4822.6 MB/s (0.5%)
---
standard memcpy : 5283.0 MB/s (0.5%)
standard memset : 8449.4 MB/s (0.7%)
---
NEON LDP/STP copy : 5271.6 MB/s (0.9%)
NEON LDP/STP copy pldl2strm (32 bytes step) : 5230.8 MB/s (0.5%)
NEON LDP/STP copy pldl2strm (64 bytes step) : 5246.1 MB/s (0.7%)
NEON LDP/STP copy pldl1keep (32 bytes step) : 5237.8 MB/s (0.7%)
NEON LDP/STP copy pldl1keep (64 bytes step) : 5249.1 MB/s (0.5%)
NEON LD1/ST1 copy : 5274.1 MB/s (0.5%)
NEON STP fill : 8472.4 MB/s (0.5%)
NEON STNP fill : 8462.2 MB/s (1.0%)
ARM LDP/STP copy : 5274.9 MB/s (0.5%)
ARM STP fill : 8461.0 MB/s (0.6%)
ARM STNP fill : 8453.6 MB/s (0.7%)
==========================================================================
== Framebuffer read tests. ==
== ==
== Many ARM devices use a part of the system memory as the framebuffer, ==
== typically mapped as uncached but with write-combining enabled. ==
== Writes to such framebuffers are quite fast, but reads are much ==
== slower and very sensitive to the alignment and the selection of ==
== CPU instructions which are used for accessing memory. ==
== ==
== Many x86 systems allocate the framebuffer in the GPU memory, ==
== accessible for the CPU via a relatively slow PCI-E bus. Moreover, ==
== PCI-E is asymmetric and handles reads a lot worse than writes. ==
== ==
== If uncached framebuffer reads are reasonably fast (at least 100 MB/s ==
== or preferably >300 MB/s), then using the shadow framebuffer layer ==
== is not necessary in Xorg DDX drivers, resulting in a nice overall ==
== performance improvement. For example, the xf86-video-fbturbo DDX ==
== uses this trick. ==
==========================================================================
NEON LDP/STP copy (from framebuffer) : 1224.1 MB/s (0.7%)
NEON LDP/STP 2-pass copy (from framebuffer) : 1085.2 MB/s (0.3%)
NEON LD1/ST1 copy (from framebuffer) : 1230.4 MB/s (0.6%)
NEON LD1/ST1 2-pass copy (from framebuffer) : 1097.2 MB/s (0.3%)
ARM LDP/STP copy (from framebuffer) : 1167.8 MB/s (0.5%)
ARM LDP/STP 2-pass copy (from framebuffer) : 1082.3 MB/s (0.4%)
==========================================================================
== Memory latency test ==
== ==
== Average time is measured for random memory accesses in the buffers ==
== of different sizes. The larger is the buffer, the more significant ==
== are relative contributions of TLB, L1/L2 cache misses and SDRAM ==
== accesses. For extremely large buffer sizes we are expecting to see ==
== page table walk with several requests to SDRAM for almost every ==
== memory access (though 64MiB is not nearly large enough to experience ==
== this effect to its fullest). ==
== ==
== Note 1: All the numbers are representing extra time, which needs to ==
== be added to L1 cache latency. The cycle timings for L1 cache ==
== latency can be usually found in the processor documentation. ==
== Note 2: Dual random read means that we are simultaneously performing ==
== two independent memory accesses at a time. In the case if ==
== the memory subsystem can't handle multiple outstanding ==
== requests, dual random read has the same timings as two ==
== single reads performed one after another. ==
==========================================================================
block size : single random read / dual random read
1024 : 0.0 ns / 0.0 ns
2048 : 0.0 ns / 0.0 ns
4096 : 0.0 ns / 0.0 ns
8192 : 0.0 ns / 0.0 ns
16384 : 0.0 ns / 0.0 ns
32768 : 0.0 ns / 0.0 ns
65536 : 0.0 ns / 0.0 ns
131072 : 1.3 ns / 1.8 ns
262144 : 2.6 ns / 3.3 ns
524288 : 14.5 ns / 20.4 ns
1048576 : 21.6 ns / 26.8 ns
2097152 : 82.3 ns / 122.3 ns
4194304 : 125.1 ns / 166.4 ns
8388608 : 149.5 ns / 183.6 ns
16777216 : 162.3 ns / 193.3 ns
33554432 : 171.5 ns / 199.7 ns
67108864 : 179.4 ns / 207.1 ns
Core to Core Memory Latency
TODO: If this is a new CPU/SoC, run c2clat to generate a core to core memory access latency graph, and paste it here. For instructions, see: https://gist.github.com/geerlingguy/842974c0e49c201c28f4be54a05cc89c
sbc-bench results
Run sbc-bench and paste a link to the results here:
wget https://raw.githubusercontent.com/ThomasKaiser/sbc-bench/master/sbc-bench.sh
sudo /bin/bash ./sbc-bench.sh -r
Phoronix Test Suite
Results from pi-general-benchmark.sh:
- pts/encode-mp3: TODO sec
- pts/x264 1080p: TODO fps
- pts/x264 4K: TODO fps
- pts/phpbench: TODO
- pts/build-linux-kernel (defconfig): TODO sec
Kingston SD Canvas Go! Plus