My current set up 24 core threadripper 2970 with 6tb of m.2 ssd with a really high read and write speed and 64g of dd4 3600 ram. But for some reason I cannot plot more then 18 plots in a day. I see these builds online with half the equipment I have doing 50+ what am I doing wrong?
Share more details! What is OS, how it all configured, what do you mean “the day” (12/24Hrs)
I’ve tried the following.
2 threads 4000 ram and 12 plots both in parallel and staggered 60mins apart it took almost 18 hours
4 threads 5000 ram 6 plots bot staggered and in parallel. Not any faster
8 threads 10000 ram 6 plots in parallel and staggered not much difference
Os is windows 10 and a day is 24hrs. If I used a 12 hour day I would be 70 years old. I don’t want to be 70 so I’ll stick with 24 hrs😜
Make sure your parallel plots are divided over your NVMe, e.g. 3 on the first NVMe drive, and 3 on the second NVMe drive. Also, with chiaplotgraph you can quickly see which phase is slower than expected. Or alternatively, grep the logs on timing info of the phases.
I have 4tb in a raid 0 and the other 2 separate. I see the 2 tab slightly faster but not by much
So the first: Linux gives you +10%…
It also matters quite a lot what ssd’s they are.
The advertised speeds don’t matter much for plotting because you swamp the cache anyway. And is it 3x2TB or?
Not if he’s bottlenecked on the storage
I have 3x 2Tb wdsn850
I benchmarked my ssd and my read and write speeds were over 3k. But when I check the task manager it showed a write speed of only 250. Also another thing to note is while I look at the task manager it showed my cpu usage at 60-70 ish but the logs on chia say something like 300%
I would remove the raid 0 and have 3x3 plots going 4T 6780 MiB (assuming you are not bottlenecked on the copy phase)
Maybe a dumb thing, but you did install the chipset drivers?
To get a good look under the hood I also can recommend the sysinternals tools, process explorer and process monitor.
If the benchmark runs for more than 10 minutes, you will see the speed drop down to about 1000 MB/s on the nvme. But in any case those drives should do fine I think for plotting unless they are completely maxed out to the rim with plots.
As Yae suggested, just run the drives separately and run x plots on each. Start with 3 on each drive and see how it goes.
Still strange though, whatever the setup you have now should do a lot better than 18hours running just 6 plots…
My mobo decided to reset the bios which pushed the ram bus down to 2600 instead of 3600. This caused k34s to take 30 hours, up from 20. With that said, check your bios and make sure your bus speeds are correct. Throughput is key.
I’ve ran into a very similar issue.
Sitting on i9 10900f (10 cores/20 threads), 128 ram and 2x 970 evo plus nvmes (this is my home build, not specifically built for chia, have 2 more setups builts for it, but waiting on hdds).
So, Crystal Mark shows proper speeds, as if my NVMEs are both properly connected via PCIe, not SATA.
But during plotting speeds in Task Manager show 100-350 mb/s, basically - SATA speeds. I know SSD speeds somewhat slows down after start, but not tenfold, I guess?
Moreso, tried plotting on my other SATA SSDs (870 evos 4TB) and basically difference is every minor:
(p1- p2 - p3 - p4 - total - total+copying - copying)
3.31 hrs 1.26 hrs 3.8 hrs 19.72 mins 8.7 hrs 8.89 hrs 11.48 mins
4.77 hrs 1.59 hrs 3.22 hrs 19.92 mins 9.92 hrs 10.82 hrs 53.9 mins
first nvme to hdd, second line sata ssd to hdd
both examples are plotting 5 simultaneously, no delays.
doing 3 plots at a time gives a bit better picture:
2.96 hrs 1.03 hrs 2.21 hrs 11.63 mins 6.4 hrs 6.98 hrs 34.7 mins
but still, something just does not feel right here
You should also have a look in resource monitor. I get very different reading then what task manager is telling me. Task manager says 0-500 MB/s
In resource monitor I see a consistent 800-1000 MB/s write going. That is with 8 plots running with 45 min stagger.
Just running one plot was giving 200-250 MB/s in phase 1