So, what I would suggest is starting 10 separate queues,
not concurrent, with 2 threads per queue (the default). Your temp plotting directory will be on the 4TB SSD, and the final destination folder will be to one of the HDDs. I'm assuming for the purposes of this post that you have the following drive letters:
C: = OS (not used for Chia, other than for holding the main app and blockchain -- you really should get a 256GB or larger SSD IMO)
D: = Plotting SSD (make sure this is actually in the primary M.2 slot that has the direct link to the CPU for performance reasons)
E: = SATA 4TB HDD1
F: = SATA 4TB HDD2
G: = USB 16TB HDD3
H: = USB 16TB HDD4
- Start the first queue, give it a name (like "D1toE1" or whatever your drive letters are), and put maybe 10 plots into this queue. Start it.
- Wait about 1-2 minutes, then repeat the above with a new name ("D2toF1"), start it.
- If you have the two external drives, do two more queus spaced about a minute apart ("D3toG1" and "D4toH1"), each with 10 plots.
- You now have four queues that are each 10 plots deep running at the same time, using about 8 of the CPU threads.
- Now, because you only have 4 HDDs, I'd wait 10-15 minutes and then repeat the above, once again for each HDD.
- That brings you up to eight concurrent queues that are ten deep (16 threads in use, about 800-1000GB/s of writes to the SSD).
- Wait 15-20 more minutes and start the final two queus ("D9toG3" and "D10toH3" or whatever -- not that the queue names really matter, but it will help you figure out what's going where in the GUI).
You could start these other queues whenever, but the main issue is when the plots are done, the PC needs to copy them all to the HDD. 101.4GiB of data at a rate of maybe 100-150 MB/s will take 12–18 minutes, and if you have multiple queues ending at the same time for the same drive, you'll increase the bottleneck on the HDD. There are ways to do all of this from the command line as well, but that's more complex so I'm not going to try to figure that out.
How long will it take to complete all those plots? I'm not entirely sure, as it will depend on the final SSD speed, CPU speed, RAM speed, and HDD speed during the process. Most likely it will take less than 8 hours per cycle through the queues, meaning you'd do more than 30 plots per day, but it's possible the SSD when tasked with ten separate plots will slow down a bit (should be around 1GB/s of writes, which it can handle ... but can it sustain that over the long haul? I don't know). If you get it all up and running per these instructions, I'd love to hear your feedback on final plotting times the system sustains. We also have some people (our RAM and SSD reviewers) working to determine how much speed you need on each of those and how it scales. But that will take some testing so we'll have future articles on those subjects.