npu-monitor: a system tray that knows what hardware each box actually has

Fri, 21 Aug 2026 00:00:00 +0000

Every multi-host monitoring tool I tried assumed homogeneous hardware: one GPU, maybe two, same on every machine. My LAN doesn’t work like that — one box has an AMD APU iGPU plus a Ryzen AI NPU, another has an AMD R9700 and an NVIDIA RTX 5060 Ti. So npu-monitor treats the device list as dynamic: each host reports its own real silicon and the tray renders whatever arrives.

The numbers come from the actual kernel interfaces, not screen-scraping: ROCm sysfs (gpu_busy_percent) for AMD cards, nvidia-smi for CUDA, /proc/stat deltas for CPU, and a raw DRM_IOCTL_AMDXDNA_GET_INFO ioctl against /dev/accel/accel0 — using the real 168-byte sensor struct from the kernel UAPI — for the NPU’s utilization columns.

csv2pg: CSV → PostgreSQL without the shell dance

Tue, 16 Jun 2026 00:00:00 +0000

Most “import this CSV” advice ends with psql invocations, \copy meta-commands, and a schema you hand-write after eyeballing the data in a spreadsheet. I got tired of that ritual, so csv2pg does the whole path in-process through psycopg3: analyze the file, infer the schema, create the table, stream the rows, verify the count, commit. No psql binary involved anywhere.

The interesting part is the type inference. It samples up to 100 random rows per column to guess a candidate PostgreSQL type, then validates that guess against every row in the file. If validation fails, the column widens to the next compatible type — up to five attempts — before falling back to text. You get a tabulated summary plus a generated .sql file before anything touches the database.

Python on AiSlackin

npu-monitor: a system tray that knows what hardware each box actually has

csv2pg: CSV → PostgreSQL without the shell dance