Odd setup (lotsa RAM), looking for plotting optimization advice

I like upcycling salvage, so I have a freaky setup leaning against a wall in the garage. :slight_smile:

  • 700gb of SSD from 4 individual drives, RAID-0 (not quite 3x 300, dangit!)
  • 14 1TB drives for farming (also RAID-0)
  • lotsa cores.
  • 250GB of ram (!!!)
  • All stuffed inside a big yellow Dell PowerEdge.

I’d love to make use of all that ram. I’ve been playing with

  1. Various file formats on the SSDs → Not a ton of difference.
  2. Using the SSDs as a swap space, and tmpfs to make it to 900gb → goes from 9 hrs/plot when I do 2 in parallel to 15hrs/plot for 3 in parallel, so it falls off the cliff.
  3. Maybe using the spinning rust to plot because parallel access is fast because of so many drives → no clear outcome yet.

If you have 14 cores and 28 threads, I would plot directly to the 14x1 TB Raid0 drives and not use the SSD at all.

If you’re using Ubuntu, disable swap.

Each plotting process should have the following:
Ram Usage (-b) = 3400
Number of threads (-r) = 2
Number of buckets = 128

I am guessing you will produce 14 plots every 16 hours, which is much more than your current attempts.


I’ll give it a shot, thanks! (I’m gonna fill up those spinning rust soon, so at that point I have to slow down or buy more drives).
I just wish there was some good way of making use of all that ram.

Just a heads up, the number of cores isn’t the only consideration, it’s the speed and modern architecture. There’s old xeon multi cpu servers floating around with many many cores but run dog slow as plotters. just something to keep in mind. The fastest plots come in around 4-5hours each and are all from AMD Ryzen systems.

1 Like

Oh ya, no where near 5 hours. I’m clocking in to 10 hours on the SSD, 13 on the spinning rust.

3400 is the “medium” suggestion from this February blog post, which also says the maximum usable -b (for k=32, 128 buckets, bitfield on) is 6750. With how much RAM OP has, I see no reason to go below that. (No idea if versions after the post changed anything on this front, though)

This isn’t the case any more with version 1.0.4 and up. Anything more than 3389 with two threads with bitfield enabled will not be used.


In my testing adding memory “improved” plot time from 6:22 to 6:18 … so I wouldn’t even bother adding memory at all unless you have literally no other use for it. It’s like 1-2 percentage point faster plotting at best.


OK, good to know, thank you!