Unless you have ramdrive or very fast nvme it won’t save you much/any time, you can only max out the CPU if the disk can keep up. Running this on my laptop to a USB drive, it maxed out the USB and took all night to get to P1 Table 3 but on my server its 300 seconds (ramdrive and SAS HDD). If I were you I would use the regular plotter and parallel a few plots unless you want to buy a lot of ram or a few nvmes which you might significantly wear out in the process.
Love the table! Working on something similar myself, will share tomorrow after more testing. For your 3900 system, have you tested the 2T NVMe SSDs not in Raid0 (aka one for temp1, one for temp2)? I’d love to see that compared to using them in raid0. Raid0 helps sustained transfer rates which is great for bigger files, but sometimes adds latency to smaller transactions. Could be an interesting study.
Hi, please share,
Do the plots created with MM Plotter are accepted on HPOOL?
They are accepted by the official chia. Anything else is based on the official chia.
Info: All the value needed for proofs are created in “phase 1”, the rest is wrapping up/compression. If the files created in “phase 1” are identical, they would do the same thing for proofs.
Now comes this… there will be more of these types soon…
Tennessee National Laboratory Super Computer with 192 cores (maybe 1TB RAM at least?)
All my plots created by MM or official CHIA plotter works in HPOOL without needing to replot anything.
Thanks for giving me the motivation to try this out. With similar specs to you (5950X, 64GB 3600MHz RAM) and different drives (2x P3600 and 2x 970 PRO) with 20 threads and 512 buckets I achieved 1066.49 sec plots. This equates to 81 plots per day. Very impressive!
Phase 1 took 466.472 sec
Phase 2 took 314.143 sec
Phase 3 took 258.27 sec, wrote 21877287047 entries to final plot
Phase 4 took 27.5885 sec, final plot size is 108836088690 bytes
Total plot creation time was 1066.49 sec
5 Likes
Amazing! need more details on how you setup your temps etc.
OK pulled myself together and got this in the name of cheap HDD SAS ( can it be done faster ? )
Dual Xeon E5-2420 0 @ 1.90GH / 32Gb RAM / -t Raid0 -d Raid0
Phase 1 took 3533.29 sec
Phase 2 took 1481.73 sec
Phase 3 took 3057.47 sec
Phase 4 took 133.717 sec
Total plot creation time was 8206.39 sec
How are your SSDs configured? as in Temp 1 and 2 or RAID.
Your best bet SSD as “temp2”, without RAID.
It’s two Btrfs RAID0 volumes with the options in this post: (Optimizing plotting on an AMD Ryzen 5950x (or any other 16c/32t CPU) - #108 by Harris)
2x 1.2TB P3600 - temp1 (25% of the writes)
2x 1TB 970 PRO - temp2 (75% of the writes)
The 970 PROs perform slightly better (for these workloads) but it is wiser to swap the temps around because the P3600 has 6 times the endurance of the 970 PRO so the P3600 should be doing 75% of the writes. This increases the time to 1083.08 secs which equates to ~80 plots a day (yes, it’s crazy that a 16 second difference produces an extra plot a day because 16*81=1296 seconds which is enough extra time for one full plot!).
There is probably room for improvement here as I have simply shoehorned madMAx into my classic “parallel” chiapos setup with no additional tweaking.
2 Likes
have nearly exact same setup- just 128gb … but never mind . My plotter ends after P1 Table 6 … do you have any idea?
Thank you!
Temperature throttling. Observe with “core temp” application.
1 Like
constant around 70-75 C … too hot?
Is that processor temp or nnme temp. Either way it is too hot. Way too hot for an nvme, on the high end for the processor. My nvme’s run at about 40 and my processors around 50-65.
1 Like
NVME´s around 40 … CPU is P1 around 75 … just changed dir 1 & 2 to my nvme and now its running.-- seems to be the RAM …
1 Like
may I ask how you share just one PCIe 3.0 Gen 3 x4 bus with both your p3600 and 970 pro? I assume they are u.2 and m.2? or?
I think you’re confusing “bus” with “slot”. In addition to the lanes provided by the CPU itself, the B550 chipset provides 4 PCIe 3.0 lanes which is a bus with 4GB/s of bandwidth. Depending on the motherboard, this bandwidth can be shared with multiple slots.
On my ASUS board, two slots share this bus - an M.2 slot (970 PRO) and a PCIe x16 slot (P3600). If both slots are used then they share the 4GB/s of bandwidth, which can limit the speed of these drives if they are both operating at full speed simultaneously, because they are each capable of using 4 lanes (4 GB/s) each.
This bandwidth sharing is rarely an issue in Chia plotting unless both drives are simultaneously saturating the bus with large sequential workloads at all times which doesn’t seem to be case (at least in my experience).
1 Like
How are you only getting 80 plots? I’ve been waiting for someone with 5950x/3950x to show that 800 second plot time! On swars manager I calculated 5960x should do 90+ plots a day in parallel and MM should do better!