How much did you manage to squeeze out of the 5900x?

Chia Plotting
1

Hi everyone,
I recently started plotting with different configurations, from 4 parallel up to 16. I am currently running 16 parallel on 4 1tb 980 pro staggered by 20 min (enough time to copy to destination). I allow up to 4 jobs per ssd and up to 2 phase 1 jobs per ssd. 2 threads per job and 4gb of RAM.
My logic is that having 8 phase 1 plots + 8 plots on other phases that need only 1 thread, I would be exactly using 8*2 + 8 threads = 24 → what my 5900x has to offer.

I am seeing plot times slightly below 9 hours for a single plot. This would mean around 43 plots a day.

When I ran one single plot for benchmarking, it finished in 4 hours! So definitely losing on performance here, but I am not sure where the sweet spot is.

Have you managed to plot more than that on a similar setup?

2

48-49 plots per day with:

  • 5900X
  • 64GB 3600
  • 4x 2TB 970 evo plus - XFS+mdadm raid0

Using GitHub - ericaltendorf/plotman: Chia plotting manager with the following relevant settings:

  • tmpdir_stagger_phase_limit: 8
  • tmpdir_max_jobs: 12
  • global_stagger_m: 25
  • n_threads: 4
  • job_buffer: 6000

Also see the thread about the 5950x (plotting performance is the same/similar to the 5900x) Optimizing plotting on an AMD Ryzen 5950x (or any other 16c/32t CPU)

3 Likes
3

Midnight to midnight I should be right at 40 for today. Rarely will you reach your “theoretical” max. I’m also using optimized chiapos.

4

Oh yeah forgot to mention, I am also using the optimized chiapos!

5

Oh nice I just found out about the XFS filesystem. Will give it a try. Also do you have 2x 2TB paired on raid 0? so in your system what appears are 2x 4TB tmp drives?

6

I initially had a single raid0 with 4 drives, then tried 2 raid0 each with 2 drives, no difference.

Oh and I forgot to mention that I use a Kingston KC2500 as staging drive.

1 Like
7

Im using 5*6 + 6 threads = 36 threads on my 24 thread 3900x

havent run for long yet, but seems to work and saw clear improvement over 4t plots

Something you might try

8

If using swar you can use multiple destinations drives and it round robins them so you could in THEORY use 10 min stagger with 2 destination drives etc.

You can over provisions ram for sure. It balloons up and then down so is never at its full amount the whole plot process.

Also potentially over provision threads too

9

Whats your plots per day, OS and hardware. I am also using a 3900x. My best is 26 per day, but I haven’t really loaded it to max as I save some resources to game and Harvest.

10

I just managed to get 42 in a 24hr period fully parallel. 12 jobs 5 in phase 1, 20 min stagger.

2 Likes
11

Any logic behind these settings? You run 12 jobs parallel per 4Tb raid 0 instances, this doesn’t seem to fit the size. Also why do you use a staging drive since the stagger is 25 minutes anyway?
and tmpdir_stagger_phase_limit: 8 is also odd. You will never reach this value anyway unless you run 16 plots in parallel and start all of them simultaneously.

12

Right, but having a higher value is not harmful (because the 25mn stagger will help prevent too many plots in phase1 anyway).

Since I do 48 plots per day: 24h*60min / 48plots = ~30mn per plot, the 25 minutes is slightly lower but close.

Without a staging drive, once a plot is complete, it takes 100GB / 0.15GB/s = ~660s = 11min to copy to a spinning disk. During that time no new plot/process is started.
With the staging drive, it takes 100GB / 1.5GB/s = ~66s = 1min to copy.
So the staging drive helps start new plots 10 minutes quicker/sooner.

12 jobs overall. Going higher does not increase my plots/day.

Here’s the disk space left with 12 plots running:

Filesystem                         Size  Used Avail Use% Mounted on
/dev/md0                           3.7T  1.5T  2.2T  42% /mnt/970-evo-plus-r0-2x-1
/dev/md1                           3.7T  1.4T  2.4T  37% /mnt/970-evo-plus-r0-2x-2

Had I known, I would have bought 4x1TB instead of 4x2TB. 2TB still gives 2x TBW.

Screen Shot 2021-06-02 at 11.27.33
Screen Shot 2021-06-02 at 11.27.332342×718 111 KB

Screen Shot 2021-06-02 at 11.28.23
Screen Shot 2021-06-02 at 11.28.232330×704 66.2 KB

Screen Shot 2021-06-02 at 11.27.58
Screen Shot 2021-06-02 at 11.27.582352×720 129 KB

13

This isn’t the primary benefit of staggering IMHO.

See Define Parallel Plotting - #2 by aaronabrown

So if you stagger the start times, you should be able to improve your performance and not overlap bottlnecks.

14

Thanks for all your insights here Adrien! These results are amazing! I’m running very similar system, difference being Win10 OS, 32Ram and 4x 1Tb 890pro. I’ve only just broken 3tb/day with similar settings. I’m thinking it might be time to take the Ubuntu plunge.

With regards to your Ram usage, am I correct in seeing that 26GB is being used?

15

That’s how I understand those metrics as well

16

42-43 plots seems about what i get, counting empirically what i see from any moment and 24 hours back.