Install Omarchy (Arch + Hyprland) without touching your son's main drive, then prove the card is healthy before listing it. Tick steps off as you go; your ticks stay in this browser.
Download the ISO from omarchy.org and write it with balenaEtcher, or from a Linux terminal:
lsblk # find the USB stick, e.g. /dev/sdX (check the size!)
sudo dd if=omarchy.iso of=/dev/sdX bs=4M status=progress oflag=syncDouble-check /dev/sdX is the USB stick. dd overwrites whatever you point it at.
If you can get a shell, confirm only the second drive and the USB stick appear:
lsblk -o NAME,SIZE,MODEL,SERIAL,TYPE,MOUNTPOINTS
omarchy-update # full system + Omarchy update; reboot if it askssudo limine-scan in Omarchy adds Windows to Omarchy's boot menu. Not needed for this test.All commands run in an Omarchy terminal. Open several and let Hyprland tile them.
sudo pacman -Syu
yay -S --needed vulkan-radeon vulkan-tools lm_sensors pciutils git cmake \
base-devel shaderc vulkan-headers glmark2 vkmark amdgpu_top
mkdir -p ~/gputest && cd ~/gputestlspci -nnk | grep -A3 -i vga # expect Navi 21 ... 6900 XT, driver amdgpu GPU=$(lspci | grep -iE 'vga.*(amd|ati)' | awk '{print $1}') sudo lspci -vv -s $GPU | grep -i lnksta cat /sys/class/drm/card*/device/mem_info_vram_total # expect ~17163091968 (16 GB) vulkaninfo --summary | grep -i devicename # expect AMD Radeon RX 6900 XT (RADV NAVI21)
Pass: 16 GB VRAM, amdgpu driver, Vulkan sees the card. Screenshot this for the listing.
Terminal 1, kernel errors. Anything printed here during the tests is a red flag:
sudo journalctl -k -f | grep -iE "amdgpu|ring|timeout|reset|fault"
Terminal 2, temperatures, clocks, power and fan:
amdgpu_top # or: watch -n1 sensorsJunction (hotspot) more than about 25–30 °C above edge points to dried paste. 100–110 °C junction means it is throttling.
cd ~/gputest
URL=$(curl -s https://api.github.com/repos/GpuZelenograd/memtest_vulkan/releases/latest \
| grep browser_download_url | grep -i linux | grep -iv aarch64 | head -1 | cut -d'"' -f4)
echo $URL && curl -L -o memtest.tar.xz "$URL"
mkdir -p memtest && tar -xf memtest.tar.xz -C memtest
MT=$(find memtest -type f -name 'memtest_vulkan*' | head -1)
chmod +x "$MT" && "$MT"Pick the RX 6900 XT if asked, run 20–30 minutes, then Ctrl+C. Pass: zero errors. Any error line means faulty or unstable VRAM.
while true; do vkmark || glmark2; done
Pass: no crashes, black screens, artefacts or kernel messages; temperatures settle inside the limits above.
The test an AI buyer cares about. Builds llama.cpp, downloads a ~5 GB model, and checks the card gives bit-identical output three times. Unstable cards drift.
cd ~/gputest
git clone --depth 1 https://github.com/ggml-org/llama.cpp
cmake -S llama.cpp -B llama.cpp/build -DGGML_VULKAN=ON -DCMAKE_BUILD_TYPE=Release
cmake --build llama.cpp/build -j$(nproc)
curl -L -o qwen3-8b-q4km.gguf \
https://huggingface.co/Qwen/Qwen3-8B-GGUF/resolve/main/Qwen3-8B-Q4_K_M.gguf
BIN=llama.cpp/build/bin
for i in 1 2 3; do
$BIN/llama-simple -m qwen3-8b-q4km.gguf -ngl 99 -n 300 \
"Explain how a graphics card works, step by step." > run$i.txt 2>/dev/null
done
diff run1.txt run2.txt && diff run1.txt run3.txt && echo "IDENTICAL - PASS"Then a sustained compute load for 10–15 minutes:
while true; do $BIN/llama-bench -m qwen3-8b-q4km.gguf -ngl 99 -p 512 -n 128; done
Pass: "IDENTICAL - PASS", steady pp512 / tg128 numbers run after run, no kernel messages. Screenshot the numbers.
| Outcome | What to say | Ask (CAD) |
|---|---|---|
| All tests pass | Fully tested: VRAM, 3D and compute stress; temps OK | ~$600 |
| Passes, hotspot hot | Works; would benefit from a repaste | $500–550 |
| One BIOS stable | Works on BIOS 1; BIOS 2 unstable | $550–600 |
| VRAM errors or crashes | For parts or repair, described honestly | $150–300 |
Screenshots to keep: B2 output (16 GB, card name), memtest_vulkan after 20+ min with no errors, amdgpu_top under load, "IDENTICAL - PASS" with llama-bench numbers, plus photos of the card, its three 8-pin sockets and the BIOS switch.