Custom RAM-only, k32-only Chia plotter running on a 64-core ARM processor. Full, valid plot performed in 6:11 min (370.87 seconds).

Because consecutive pots can be performed, technically the last portion writing to disk can be discarded from the timings as this operation is performed in the background whilst the next plot starts, with very little CPU interference (only on a single thread, at that).

In fairness, the initial part where the buffers are pre-allocated was not counted. This warm-start is performed for making consecutive plots as well, so I did not count it, but otherwise, it would add a few seconds to the final time. I may show another benchmark later without warm-starting.

If interested in using this to fill unprovisioned data centers, feel free to contact me.