Plotting speeds just fell off a cliff... help!

Hey there,

Pretty new to Chia, could use some guidance. First day after getting rig going I started off by plotting 2 plots in parallel with no delay. I did this a few times just to get a good base speed. The parallel plots were completing in about 6.5 hrs with the 1st phase completing in avg 130 minutes.

I then setup 4 plots in parallel with a delay of 150 seconds to ensure only 1 plot is in phase 1 at a time. From my understanding this should have been no problem with 12 threads.

When I did this I came back after about 10 hrs and the plots were as follows:
P1 - 17%
P2 - 8%
P3 - 2%
P4 - 1%

I waited a couple more hours and the plots were not really moving so I deleted the bottom two and the top two started progressing again but still pretty slowly.

Once the staggered plots finished I deleted the stranded temp files from the ssd and went back to plotting 2 at a time in parallel only the speeds have not recovered. It’s now taking 10+ hrs to plot what used to take only 6.5 hrs.

Is there some cache/virtual mem somewhere I need to clear to get speeds back up to fresh out of the box speeds?

Ryzen 5 3600
32GB DDR4
2TB Teamgroup MP33 NVME SSD gen3x4
Ubuntu 20.04

Any help would be appreciated
Thanks!!

Make sure that the SSD is being trimmed regularly - not sure how that works on Ubuntu, but if you search on the forum there has been discussion about it.

Also, look at using the madmax plotter - it will blow the original speeds you were seeing out of the water.

This show the extended write performance of the 1tb version of the drive. There seems to be some inconsistencies, perhaps not the best for plotting. Four plots doesn’t seem like much, but… still 1+tb/plot created might be too much for a DRAMless ssd.

Is this meant to be 150 minutes?

Yeah thats weird… that you can get these kinda speeds with this SSD. Feel like I wasted my money on my MP600 Pro (also plotting at 6.5hr/plot). Either yours is really fast, or mine is really slow…

Anyway, its hard to diagnose this problem, something must’ve changed with your setup. Did you move the SSD to a different port? Did you add a new device? Want to consider a fresh install of Ubuntu (don’t forget to save your keys though)?

Only other things that come to mind is:

  • Did you turn off swap? Apparently a swap file or partition can cause a slow down.
  • You sure you deleted all the .tmp files from your plotting drive?
  • Is your SSD sufficiently cooled?
  • Are you sure you’re plotting with the SSD as temp?

Trim means to run

“Sudo fstrim -av”

Every 6 hours, everytime a plot finishes, then run the trim

Why in the hell are u using ‘chia plot’? Mad-max makes plots in 20 minutes on my machine, I used to do chia-plot and got 16/day, staggered by 4-5 1 hr interval, now I run mad-max linear, and I get 3*24, e.g. 72 plots per day, because all is done in memory & cpu, there is less wear&tear on the NVME’s

There are even GPU versions that are faster, but for me the cpu is just fine, I have a 32 core, but I only use 16 threads when I run;
But it sounds certainly that your problem is you don’t ‘trim’, if you don’t run that every few hours your plotting will grind to a halt.

fstrim -av, av means all ssds on your system, to trim

Yeah 150 minutes, not seconds.

And nothing changed in setup, I literally let the remaining staggered plots finish, then cleared the ssd, and started 2 simultaneous plots again.

UPDATE

Hey Everyone,

Trimmed SSD, shut everything down, rebooted PC, and started plotting again fresh.

Started again two plots simultaneously in parallel while watching HTOP to monitor the system.

I allocated 4GB per plot and allcoated the full 12 threads 6 per plot.

My RAM usage has been steadily going up throughout the plotting, it’s 4 hours in and RAM usage is up to 15.7GB out of 32.

The SWAP usage has been slowly climbing as well. It’s up to 134MB usage.

The only other thing I’m doing is farming. There are no other heavy processes running in the background.

I can’t imagine this is normal, what could be causing the RAM usage to keep climbing like this??

Thanks!!

It hit 16GB of RAM usage for 2 simultaneous plots? Thats not normal… either you have a memory leak from another program… or you have the commands executing incorrectly.

What operating system are you using? and are you using GUI or command line to create plots? Running a script?

Ubuntu 20.04, fresh install besides chia and a couple forks that I installed. (Flax and Spare)

I’m currently just plotting with the GUI. What I don’t get is that the first couple times plotting I got great times with CPU usage in the 200’s but now I’m getting this memory usage and PC usage has dropped to 50% for the same stages in plotting.

I’m outta ideas, I’m not a Gui user. I do everything from the command line. My guess, its a bug within the Gui program. Try updating the Chia installation and see if that fixes it.

IMO, Use swar plotter to help you stagger better.

I just swapped over to linux myself, personally, I notice quite a few bugs / memory leaks with linux that I’ve spent quite a lot more time troubleshooting than my windows setup (it was pretty much just up and running once installed).

also, /var/log/syslog – I would keep a tail on that and see if anything comes up while it’s plotting. due to my usb drives, I noticed that the way I had /etc/fstab setup, there was a usb drive constantly mounting/unmounting every few seconds as it went idle and when harvester would query the plots on it. kernel.log is also a good idea to keep an eye on.