HP DL380 G9 , E5-2667v3, 32GHz, 256GB DDR4 2166 RAM: 45 minutes per plot?

Latest performance, 28.8 minutes, 100% plotted in RAM. I’m happy :slight_smile:
Thanks for everyone’s ideas and support!

./build/chia_plot -n -1 -r 32 -u 256 -v 128 -t /mnt/ram/ -d /mnt/firecuda/

[P1] Table 1 took 13.9157 sec
[P1] Table 2 took 133.175 sec, found 4295001222 matches
[P1] Table 3 took 139.551 sec, found 4295023393 matches
[P1] Table 4 took 154.708 sec, found 4294939025 matches
[P1] Table 5 took 153.482 sec, found 4294968865 matches
[P1] Table 6 took 153.627 sec, found 4294906483 matches
[P1] Table 7 took 124.45 sec, found 4294779981 matches
Phase 1 took 872.95 sec
[P2] max_table_size = 4295023393
[P2] Table 7 scan took 8.27616 sec
[P2] Table 7 rewrite took 34.5546 sec, dropped 0 entries (0 %)
[P2] Table 6 scan took 25.1656 sec
[P2] Table 6 rewrite took 36.9818 sec, dropped 581265744 entries (13.5338 %)
[P2] Table 5 scan took 24.5409 sec
[P2] Table 5 rewrite took 35.2735 sec, dropped 761961092 entries (17.7408 %)
[P2] Table 4 scan took 24.0875 sec
[P2] Table 4 rewrite took 35.0206 sec, dropped 828817000 entries (19.2975 %)
[P2] Table 3 scan took 23.8153 sec
[P2] Table 3 rewrite took 34.5936 sec, dropped 855115402 entries (19.9094 %)
[P2] Table 2 scan took 23.4591 sec
[P2] Table 2 rewrite took 34.6511 sec, dropped 865594479 entries (20.1535 %)
Phase 2 took 362.299 sec
Wrote plot header with 252 bytes
[P3-1] Table 2 took 37.4185 sec, wrote 3429406743 right entries
[P3-2] Table 2 took 28.2596 sec, wrote 3429406743 left entries, 3429406743 final
[P3-1] Table 3 took 39.7989 sec, wrote 3439907991 right entries
[P3-2] Table 3 took 27.9827 sec, wrote 3439907991 left entries, 3439907991 final
[P3-1] Table 4 took 40.0459 sec, wrote 3466122025 right entries
[P3-2] Table 4 took 28.3941 sec, wrote 3466122025 left entries, 3466122025 final
[P3-1] Table 5 took 40.6359 sec, wrote 3533007773 right entries
[P3-2] Table 5 took 28.7075 sec, wrote 3533007773 left entries, 3533007773 final
[P3-1] Table 6 took 42.6549 sec, wrote 3713640739 right entries
[P3-2] Table 6 took 30.7469 sec, wrote 3713640739 left entries, 3713640739 final
[P3-1] Table 7 took 37.4204 sec, wrote 4294779981 right entries
[P3-2] Table 7 took 35.8018 sec, wrote 4294779981 left entries, 4294779981 final
Phase 3 took 424.376 sec, wrote 21876865252 entries to final plot
[P4] Starting to write C1 and C3 tables
[P4] Finished writing C1 and C3 tables
[P4] Writing C2 table
[P4] Finished writing C2 table
Phase 4 took 71.473 sec, final plot size is 108833546539 bytes
Total plot creation time was 1731.17 sec (28.8528 min)

I’ve tried this. Still having issues with it. Using a simple program called chia-plot-mover to move plots from my buffer SSD to my final drive. I’m wondering if this program isn’t messing things up. I need to try copying directly to my final destination.

Yeah. Idk man. I’ve tried. It’ll plot 2 or 3 sometimes 10 plots then all of a sudden it’ll crash with mem and swap full.

Even though you are using a buffer SSD, I think it may still not be moving fast enough. I’ve tried the same thing (256GB and all Ram drive) and was running into issues with the completed plots not offloading fast enough. I gave up and went back to using an NVME.

Yup! This was the problem. Last night I added the wait for copy flag and all issues are gone. I just lose about 2 minutes per plot to wait for copying to the buffer SSD to complete.

I plot using madmax entirely in ram on Ubuntu 20.04 allocating around 270GB of ram, set wait for copy first, unless you have very fast copying away to your final drive, in which case it is not a problem.

However, you do not get a performance increase, unless your temp 1 drive you were using before was very slow. Temp 1 is just used to aggregate the plot, temp 2 is used for all the calculations.

So get a fast hard drive for temp 1 get 110GB of ram extra for temp 2 and enjoy.

Yeah. I had to use wait for copy. I also have very fast copy to temp drive. Still wasn’t enough. I wasn’t really looking for performance gains. Just trying to save some SSD wear and tear.