I am running Gigahorse GPU plotter on a Linux machine with 32GB of RAM utilizing -3 tmp dir.
RAM stays practically unused during GPU plotting.
Is there any clever way to utilize some of that RAM in order to speed up tmp dir?
Maybe on the system level?
Some kind of caching of tmp device?
Maybe JBOD made of RAM disk and tmp device?
How much memory can your mother board hold? What size gigahorse plots you trying to make? Both my window and linux machine plotters have 256gb of system memory. Sorry I can’t help you, but tell you to look on ebay for a birthday gift to yourself!!!
As drhicom said. You wanna get gpu plotting going ,ram is the answer .
From what I have seen of Max’s programming prowess ,he didn’t leave any performance on the table with what you have .
Overclock the gpu ,make sure its stable . Resizeable bar if your system is capable .
Ram using a xmp profile in the bios ?
Overclock the cpu if its capable .
Most of all, you need more memory if you want 2 minute plots.
What are you using for a temp drive ? M.2 or U.2 are the kings there . U.2 being pricey ,but with a lot longer life span .
Hope some of this helps .
Thx for suggestions!
I do not need any really fast/massive plotting.
I have a need to replot once (or twice) ~3k plots. We are not talking about continuous PB-scale workloads.
I can live with ~10mins plots, but I just wanted to utilize RAM which is hanging there a bit better…
Least you are on Linux. Windows would be 30 minute plots lol. I learned the hard way 
. Perhaps setting swapiness to zero?
Set the swappiness parameter to a low value or 0.
Option A: Open the file /etc/sysctl. conf in a text editor and enter the vm. swappiness parameter of your choice. Example: vm.swappiness = 0. ...
Option B: To change the value without restarting the operating system, run the following command: sysctl -w vm. swappiness=0.