Optimizing plotting on an AMD Ryzen 5950x (or any other 16c/32t CPU)

Chia Plotting
170

The average %util is 75%, it occasionally spikes to 85%. Here is the full output (note this is just a snapshot and some of the other values vary quite a lot).

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0.03   49.56    5.95    1.87    0.00   42.59

Device            r/s     rkB/s   rrqm/s  %rrqm r_await rareq-sz     w/s     wkB/s   wrqm/s  %wrqm w_await wareq-sz     d/s     dkB/s   drqm/s  %drqm d_await dareq-sz     f/s f_await  aqu-sz  %util
nvme0n1        410.00 192224.00  2639.00  86.55   10.16   468.84  775.00 398332.00  5514.00  87.68   27.27   513.98    7.00  72084.00     0.00   0.00   15.00 10297.71    0.00    0.00   25.41  73.60
nvme1n1       1640.00 192192.00  1439.00  46.74    1.60   117.19 2502.00 309508.00  2333.00  48.25   24.57   123.70    9.00  34924.00     0.00   0.00    2.00  3880.44    0.00    0.00   64.11  71.60
nvme2n1        411.00 192288.00  2690.00  86.75   12.52   467.85  776.00 398324.00  5512.00  87.66   34.24   513.30    7.00  72052.00     0.00   0.00   15.43 10293.14    0.00    0.00   31.82  72.80
nvme3n1       1639.00 192168.00  1453.00  46.99    1.74   117.25 2652.00 310064.00  2195.00  45.29   25.01   116.92    9.00  34992.00     0.00   0.00    1.67  3888.00    0.00    0.00   69.19  71.20
171

I know its not a 59xx but I have a TR Pro 16/32, WRX80 Sage with 64gb ram. I am running windows and have 7 Firecuda 520s 1tb and 2 tb. I have not been able to reach more than 40 plots per day and that is using 6 threads/4gb ram with 24 parallell process spread evenly across the temp drives(my own script that launches a process every 1200s across each drive until done) At 24 I get 52000s per plot but is the most I have been able to get in a 24h period

Any suggestions?

172

Firecuda slow. Bursts to 4500mbs then throttles down to 500mbs for extended writes (anything over a minute.) I had Firecudas and turned them in for 970 pros.

173

i can’t find any solid info on the 2tb NVME sabrent rockets. from my findings they throttle down to 1000mb/s after cache?

174

540 MBps after cache I’m afraid

175

Hi parsifal, can you tell me which software did you use to lock the core speed at 4.7ghz at most time like you said last time? I’m currently using linux ubuntu focal atm, but can’t find a software that can actually lock the core speed at that exact number i want. (I also use 5950x)

176

I tried it without RAID and mounted them w/o discard and set the default trimming of linux to hourly, but still getting IO Waits up to 10% with those Firecudas.

I am not quite sure if i should return them and get some 980/pm9a1 something like this instead. But i am a little worried about TBW and warranty… I still have a lot free HDDs which need to be filled up. Or if i should slap them additional drives in and keep those Firecudas…

I have some pm9a3 in another system and i really like them… but no chance to get one of those atm :frowning:

177

Mine are blazingly fast (980 Pro 1TB) and regarding TBW I think you need to be much less concerned then marketing has led you to believe. :wink:

Mine are now at 18% life according to the NVMe drives themselves with 301TB written. Extrapolated that points towards ~1700TBW as the real value for these drives (performing a Chia workload only!) and then it’s everyone’s guess how much longer it will actually last.

image

178

New here, and to this whole chia world, so apologies for sounding so naïve, but is there something similar that can be done on windows?

179

Hi wwwroth,

Try this powershell. save it as .ps1 and run it.

# Egrep phase of log file
# 

while ($True) {

	$LogFiles = Get-ChildItem -Path "C:\Log\2021*.log"
	ForEach ($File in $LogFiles) {
	 "" 
	 $File.FullName
	 Get-Content $File.FullName -tail 5000 | Select-String -Pattern "Total time " | Select-Object -Last 1
	 Get-Content $File.FullName -tail 5000 | Select-String -Pattern "Starting phase " | Select-Object -Last 1
	}

    Start-Sleep -Seconds 60
    clear
 }
180

Hi DougieC,

Simple powershell script. save as .ps1 and run it.

while ($True) {
robocopy E:\CHIA\PLOT\ D:\FARM\ *.plot /R:0 /W:0 /J /MOV /XX  /TEE 
Start-Sleep -Seconds 60
}
1 Like
181

Superb, thanks for that, appreciated!

182

Hey man could you share how you installed chia in clear linux? I have trouble installing chia cli and chia gui.

183

Firstly if you’re going to the effort of installing Clear Linux (presumably for improved performance) then you should install the Server version to avoid the overheads of the desktop/GUI.

To install Chia CLI it would be something like this:

sudo swupd bundle-add git
sudo swupd bundle-add python3-basic
git clone https://github.com/Chia-Network/chia-blockchain.git -b 1.1.6
cd chia-blockchain
sh install.sh
. ./activate
chia init
184

I see. Are you referring to the server version of Chia not the server version of Clear Linux right?

185

I also have the AMD 5900X (12c/24t), RAM (64GB), SSD (2TB*2, M.2, RAID0)

Now I’m doing 11 plots in parallel with settings (2 threads, 3390 MiB, 0 delay time) on Windows 10, using Chia CLI programs by opening 11 command line windows to achieve 11 plots parallel.

But comparing to your results, my averaging time is over 12 hours per plot, almost your 2x time.

Am I doing something wrong about the settings?

186

Yes, not telling us your detailed hardware config and yaml job config

187

Hey, Newbie here.
I’ve skimmed this thread but thought Id add my efforts.
I’m running a 5950x and 64gb Ram. 2 x 2tb NVME and 1 x 1tb NVME. 2 threads per plot
the two 2tb NVME’s are max 7 concurrent with a 5min stagger
the 1tb NVME is 3 concurrent with 5 min stagger and for the last 3 days I’ve been getting exactly 40 plots per day. I still have some headroom (not much) but I’m happy to keep at 40.

This rig is also mining ETH with a 3090 at the same time.

This on Win10 running SWAR manager

188

hi there, may i know your nvme brand and all the details please?

im also running a 5950 with 64gb Ram.
1x 2tb nvme seagate barracuda q5 pcie g3
2x 1tb nvme xpg
using 2 threads per plot

and using SWAR manager too

i try 2tb nvme 6 max concurrent with 30min stagger
2x 1tb nvme 3 max concurrent with 30 min stagger

the plotting is not working well, im finding my 2tb nvme is having 100% almost all the time, even when the first plot is running

do you have any idea what the problem is? or anyone can help to give any recommendations…

thanks

190

Hi

I’m using Sabrent 2tb NVME

https://www.amazon.co.uk/dp/B07MTQTNVR

If your filling up your NVME have you checked to see if you have any old Temp files still on it