Running four AKD1500 accelerators on a Raspberry Pi 5

A real DMA failure became the multi-card recovery design.

20 September 2026

0001:06:00.0 produced DMA wait completion timed out roughly every five seconds. The link remained up; user-space learning failed with errno(110).

sudo systemctl stop vorsa
echo 1 | sudo tee /sys/bus/pci/devices/0001:06:00.0/remove
sleep 2
echo 1 | sudo tee /sys/bus/pci/rescan
sudo systemctl start vorsa

What changed afterwards

Build 1.9.36 added current task, last compute time and calls/s per card. A failed card leaves the group, three continue, an alarm is recorded, a 60 s rejoin attempt follows, and reset reloads the learned state.