K35 plotting tweaks?

Thank you, I saw this information earlier, but there are so many on the Internet that I have already got confused.

I also met earlier that the probability of finding a block on K33-35 is greater than on K32. In other projects, you need to put a collateral, block your coins on your wallet, but in Chia you need to spend time and equipment for more K … K35, for example.
But now I read the official documents and it directly says that the K33-35 … has absolutely no advantages whatsoever.
Or is there?
In general, as I understand it, there is probably no difference, the question has been raised many times. No difference(

1 Like

There will be no advantage @vovannovig, in fact you will be at a disadvantage because you won’t be able to use bladebit or madmAx plotter. This is me fooling around as I usually do when I stumble onto projects I find interesting.

A little update on the second K35 plot. Total storage free was as low as 335GB before final stages started going. This is on windows and chia staying in sync.

4TB should be considered the absolute minimum if you are attempting this madness keeping the full node running. The upside is that you might be able to parallel plot with the additional storage.

Will update on final build time with 8 threads/41000 RAM allowed via GUI.

Plot #2 completed in 40.761 hours. With 8 threads and a buffer size of 41000MiB.

The recent post regarding windows 11 limiting NVME speeds appears to make sense of my SSD low utilization rate according to the task manager.

So maybe we are leaving something on the table? I will test on windows 10 next and then on Ubuntu before making a final decision in regards to sticking with an OS.

Here are the numbers for each phase according to the log:

Phase 1 - 53249.479

Phase 2 - 29447.411

Phase 3 - 59493.240

Phase 4 - 4550.015

Let’s assume 4TB NVME storage along with 52GB of RAM is installed. What would be recommended in terms of staggering 2 plots to run in parallel? There appears to be plenty of headroom if the task manager is somewhat accurate in what is displaying. 64GB+ RAM can be installed if one feels it is absolutely necessary.

@xkredr59 or anyone else with great knowledge of this? This is not only about my experiments, but maybe helpful for the community as a whole.

I am ready to try a K32 tomorrow and thinking of going with madmax plotter & 128GB RAM. Will this cause any issues putting these K32 plots on the same HDD even though they are generated from different plotters? I would assume there will be no issues, but still getting my feet wet.

No problem at all :grinning:
Madmax’s plots are fully compliant.

1 Like

Since madmax (and bladebit) came to the market the whole staggering discussion has faded a bit.
Both use all cores assigned all the time so no use.
But in the ‘old days’ it was more or less common knowledge that getting optimum staggering was a matter of trial and error.
Only with 40+ hours trials this in not very attractive I guess.

I would suggest plot1 : 6 threads and plot2: 6 threads starting 10 minutes later.
If your NVME is fast enough to keep all threads happy you can expect both to finish P1 after roughly 54000 sec. For P2/P3 it’s a question how cpu/ram/nvme speeds work to keep both plots out of each others hair :grimacing: It will downgrade P2/P3 times certainly but just be patient and see how much time it takes for 2 K35 plots.

1 Like

Will staggering cause an issue if planning to use a 128GB RAMDisk? I am guessing it would.

I think 6 threads makes sense. 6 threads, 4x1TB SSDs in RAID 0, 128GB RAM (a little less due to iGPU of 5700G) @ 3200. I would be very interested to see what can be accomplished with this setup. I am assuming I don’t touch anything in regards to the buckets etc. Hoping for 25-30 minutes per plot if possible.

The only reason I am considering this is my longterm goals and this should in theory allow for my poor SSDs to survive a bit longer after 120+ drives are plotted.

Oh, you want to run multiple madmax instances at the same time for k.32’s… I assumed you wanted to do this with standard plotter for k.35’s.

On a single processor machine I never heard of people doing that and claiming an advantage.

For k.32 the ramdisk is very helpful, ~110GB for k.32 will do, for k.33 and k.34 with madmax you’ll need much more ram than your system can handle.

1 Like

Doing several tests. So far 8 threads appears to be the best option for this cpu as phase 1 and 2 with 6 threads is a tad slower.

@xkredr59 as far as K32 is concerned. I am planning a single plot at a time with 128GB ramdisk per system and then slowly transitioning to K35 as I reach certain internal targets I want to hit. I would assume with 4TB storage and 128GB RAM I will be able to parallel plot 35s once that time comes for the transition.

This has been a tremendous help. There is so much information out there regarding chia and it’s a bit overwhelming to keep track for a newcomer.

Final K35 test plot time was just a tad over 41 Hours.

8 threads appears to be the sweet spot for this configuration with 41000 MiB set as the max ram allowed.

RAMDisk test with madmAx is next

Alright so K32 with madmAx took 46.8409 minutes.

Any suggestions to try and get this down? I was hoping for sub 30 minutes, but I am now seeing that Windows 11 is likely what’s slowing me down with the NVME issues it currently has.

I will keep playing around with this a bit for the coming days. Very impressed overall.

Did you do this with a ramdisk voor temp2 (-2 parameter)? How did you configure the ramdisk in windows 11?

1 Like

So keeping it pretty simple @xkredr59 since the people helping me get this operation going won’t be comfortable in powershell if needed.

Made RAMDisk with imdisk, fixed state to keep the speed as fast as possible. Set it to temp2 destination.

Set thread count to 6 and let it rip. Left the multiplier at 1. 256 buckets.

Time is down to ~45 minutes per plot.

What else can I do? Next test will be on Windows 10 because the NVME utilization is still very low according to task manager.

I use -r 12 on my i7/9700 cpu which has similar number of cores/threads (8/16) and clockspeed
as yours.
On Ubuntu Linux with NVME for temp1 and 110GB ramdisk for temp2: ~37 minutes per plot.
I would expect a bit faster than 45 minutes for your setup, but think the difference is more windows/linux than windows10/windows11.
It has been reported by others that going to linux improved their plot times. Linux is said to have more efficient ramdisk.
Is it possible for you (or those helping) to give Ubuntu a try?
If your Windows install is not on the NVME but another disk maybe you can remove and save that disk and install another (ssd?) for installation of Ubuntu linux. You can easily switch back to Windows then by just plugging the windows disk back in.

1 Like

if you want to go faster, you need to plot on Linux and use EXT4, NTFS is inefficient.
For comparison, I have a 700$ server with 2x Xeon E5-2680v2 CPU’s and 128GB RAM and I get 26 minute K32 plot times with madmax on Ubuntu server. Temp drive is a enterprise SSD

1 Like

I will definitely give Windows 10 a try and then explore Linux options. Ubuntu is pretty simple these days so either that or something like elementary OS might work for the normies.

@ChiaBeast Just a warning, when changing OS make sure to backup the blockchain and wallet databases before you do.
Well, at least if the chia client is installed on your OS drive. I don’t expect you changed the location of the databases in that case, so wiping the OS or changing boot-drive will throw you back to syncing the blockchain and wallet from scratch again, costing a couple of days…

If you backup and restore the blockchain and wallet db’s you can be up and running in minutes vs days.

1 Like