Is my setup for 970 evo plus with madmax correct?

Hi everyone , i have been running my plotting since yesterday and getting result of 92minutes each plot.
I wonder if there is anything i can do to:

  1. Fine tuning it with current rig i have
  2. Add another NVME Slot onto the PCIE Lane and insert one more 970 evo plus

My rig as actually a server rig that i used to host some Virtual Machines . But currently only 1 VM are running which is for my Smart Home Project. It given 2 Cores & 6GB of RAM . (I wouldn’t mind to turn this off if need when doing the peformance testing)

Here is my rig setup
CPU: Intel i9-10900
SSD: Patriot P210 512GB (Windows OS)
NVME (Temp): 1x Samsung Evo Plus 2TB
HDD (Final): 5x WD- 18TB

Here are my config on MadMax
.\chia_plot -n 164 -c xxxx -f xxx -t D:\ -d E:\madmaxplot\ -r 20 -u 256

and below are the result of my test run for 1 Plot:

Network Port: 8444
Final Directory: E:\madmaxplot
Number of Plots: 1
Crafting plot 1 out of 1
Process ID: 5940
Number of Threads: 20
Number of Buckets P1: 2^8 (256)
Number of Buckets P3+P4: 2^8 (256)
Pool Puzzle Hash: e8d39a837aecfbb9a485a727e4b7c2c3abdf6d8446e35a48b737205d1c47523e
Farmer Public Key: 930b066d5bc72ba97fca339f8a6afa1afb5fd83c8c3f47f2672abbc96641561b1d9e5eb062c495a7f2f877fa4bf4b890
Working Directory: D:
Working Directory 2: D:
Plot Name: plot-k32-2021-09-28-21-20-0ce65f51ef161550e6b906aebce8a6a5d77bf8068c0181fdcd7a658fed51fd97
[P1] Table 1 took 43.7247 sec
[P1] Table 2 took 258.049 sec, found 4295034718 matches
[P1] Table 3 took 383.835 sec, found 4295001338 matches
[P1] Table 4 took 464.33 sec, found 4294974792 matches
[P1] Table 5 took 457.086 sec, found 4294922766 matches
[P1] Table 6 took 430.034 sec, found 4294751956 matches
[P1] Table 7 took 293.076 sec, found 4294563221 matches
Phase 1 took 2330.31 sec
[P2] max_table_size = 4295034718
[P2] Table 7 scan took 13.2005 sec
[P2] Table 7 rewrite took 43.2904 sec, dropped 0 entries (0 %)
[P2] Table 6 scan took 107.713 sec
[P2] Table 6 rewrite took 145.594 sec, dropped 581249405 entries (13.5339 %)
[P2] Table 5 scan took 103.419 sec
[P2] Table 5 rewrite took 139.57 sec, dropped 761980488 entries (17.7414 %)
[P2] Table 4 scan took 101.846 sec
[P2] Table 4 rewrite took 138.012 sec, dropped 828871507 entries (19.2986 %)
[P2] Table 3 scan took 93.8416 sec
[P2] Table 3 rewrite took 136.227 sec, dropped 855076842 entries (19.9087 %)
[P2] Table 2 scan took 90.9767 sec
[P2] Table 2 rewrite took 135.74 sec, dropped 865628559 entries (20.1542 %)
Phase 2 took 1253.42 sec
Wrote plot header with 252 bytes
[P3-1] Table 2 took 131.225 sec, wrote 3429406159 right entries
[P3-2] Table 2 took 85.5854 sec, wrote 3429406159 left entries, 3429406159 final
[P3-1] Table 3 took 144.668 sec, wrote 3439924496 right entries
[P3-2] Table 3 took 85.0203 sec, wrote 3439924496 left entries, 3439924496 final
[P3-1] Table 4 took 146.547 sec, wrote 3466103285 right entries
[P3-2] Table 4 took 87.6563 sec, wrote 3466103285 left entries, 3466103285 final
[P3-1] Table 5 took 149.842 sec, wrote 3532942278 right entries
[P3-2] Table 5 took 86.9655 sec, wrote 3532942278 left entries, 3532942278 final
[P3-1] Table 6 took 157.329 sec, wrote 3713502551 right entries
[P3-2] Table 6 took 93.9979 sec, wrote 3713502551 left entries, 3713502551 final
[P3-1] Table 7 took 137.83 sec, wrote 4294563221 right entries
[P3-2] Table 7 took 111.312 sec, wrote 4294563221 left entries, 4294563221 final
Phase 3 took 1419.27 sec, wrote 21876441990 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 536.188 sec, final plot size is 108830855830 bytes
Total plot creation time was 5539.3 sec (92.3217 min)
Started copy to E:\madmaxplot\plot-k32-2021-09-28-21-20-0ce65f51ef161550e6b906aebce8a6a5d77bf8068c0181fdcd7a658fed51fd97.plot
Copy to E:\madmaxplot\plot-k32-2021-09-28-21-20-0ce65f51ef161550e6b906aebce8a6a5d77bf8068c0181fdcd7a658fed51fd97.plot finished, took 389.272 sec, 266.624 MB/s avg.

with current setup that i have it took about 92 Minutes for 1 Plot and 165plot to fill up the 18TB HDD . Which will take 10 Days for 1 HDD , i have about 5 of them . So technically it will took 50 Days to done plotting all of them . But if in future i would want to add more HDD , it might take time for it to get done. So hopefully you guys can recommend me on how to get full potential out of my rig or maybe setup another rig specifically for plotting?

Thank you!

I dont think you should have -r set to 20, you only have 20 threads and your using some of them for your other project.
That said MM normally just uses all cores no matter what you specify in that command.

Another nvme could help, personally it made no difference on my machine.

I have an 11900k, I would have thought you would have speeds closer to mine ( 35 mins ) as you have more threads than I but slighy slower I believe.

I’d think stopping your other project while plotting would speed things up.

I dont bother setting -u and just let it use defaults.
It looks like your plotting OG plots, you might want to consider creating a nft and plotting poolable plots.

You didn’t say what os but your “DRIVE LETTERS” give it away. People say you can get higher speed in linux.
Your 256 is pushing it towards cpu bound I think. Did you try 64?

I have the same CPU, although I bumped RAM to 128GB, and am using 1 Samsung 970 1TB. My plotting times are ~30 mins on Win 10 Pro, ~27 on Ubuntu 21.04.

Also, in BIOS there was a selection for what type of cooling I have. If you don’t do anything, it will give just 65W to your CPU, and it will throttle a bit. If you set it for max, it gives up to ~200W (you get higher CPU freq), although you need to monitor your temps. I used some old case that I had, and it only had one 140mm exhaust, therefore, I didn’t use liquid cooler (what I would prefer - cooler/quieter). Instead I used Noctua, and it still runs not that hot. This CPU is non “K” version, so is rather stubborn with trying to push it more.

You need to be careful with how you plug your NVMes. You need to check your motherboard diagram, as on mine (MSI z490 a-pro) looks like only one NVMe connector is separate, everything else (the second NVMe connector and PCIe slots) looks like are shared, or I missed something in the BIOS. Otherwise, they will start sharing bandwidth, basically killing plotting times (what was my problem).

As far as your VM, CPUs are shared resources, so as long as that VM is underutilized, it will not interfere too much with your plotting times. Neither will it suffocate, as OS will try to do proper allocations. The problem is not so much with leaving a couple of threads free (as MM is really greedy, and will ignore that), as with CPU affinities / priorities. You can do that through PowerShell, but it is a major pain (would rather avoid doing it). (Yeah, it would be nice if MM had CPU priority selection.)

I am using RAM for tmp2, so the original post was wrong (I started with two NVMes)
So, that ramble about those two NVMes only apply, if you don’t upgrade your RAM.

I did mention my OS if i’m not mistaken on the SSD with Windows OS in it . it is saperated from the Tempo NVME

Update: Sorry just saw that you were asking about the 256, I’m thinking of switching it to default. When i was doing the setup i was following youtube instruction to just get it work yesterday hahaha. So before trying other thing would love to know others opinion on this first

Wow, you get about 35 mins per plot? is it with 1 NVME or 2?
May i know your full spec and your settings? did u use madmax too?

30-31 mins with 2x 1TB Samsung 970.
Sorry, I take it back. This is how I started. Once I updated to 128GB RAM, I am using just one NVMe.

I did try to RAID0 those two NVMes, but it didn’t work that well (basically no speed gains).

Thanks for the input
I might be need to reconfigure my setting.
Is there any issue if stop it plotting on middle phase? current on 2nd Phase.

Probably i’ll try to spend sometime on benchmarking this rig first before considering adding another NVME. I’m using B450-I an ITX board which only have 1 NVME Slot with PCIE 1 lane. Which made me abit unhappy but well didn’t plan to do chia when i was purchasing this rig hahaha

By the way, i do wanted to go for Linux OS but would want to try on Windows first.
i have lots of setting done on the VM so it would be hassle for me to move it around without backups.
If i install the Linux OS is it the desktop version or CLI version one ? or both of them give same peformance?

Not at all, you just need to clean up the folder, as MM just barfs.

I got about 3 mins faster time, comparing to Windows, so potentially not worth it. Although, if you run VBox, you should be able to just export those machines, and all settings would be preserved (I think). I don’t have experience with any other VM platform.

For that box, I used desktop with Gnome. For my servers, I just run CLI, but those are wimps (both CPU and RAM wise)

Yes, that is also super important, and is not set by default. Good catch.

1 Like

1 nvme is 1tb firecuda 520, i9 11900k , 64gb ram in xmp iirc 3200. Yes with madmax.

.\chia_plot -n 1 -t “D:\temp” -d “E:plot folder” -p -f -r 16

1 Like

You can get a converter so that you can plug your nvme in a GPU slot. Maybe that is an option.
A Google search says that pcie x1 is still quite fast

I am getting confused. I did set it up few months ago, after that I was setting up another box.

I remember, that that one NVMe slot was sharing the GPU slot, and was directly connected to the CPU, where everything else was sitting on the PCH chip (although, I lost that booklet somewhere). However, just looking at the PDF manual, looks like that is not true.

Yes, I got that PCIe-to-NVMe card, but again, no matter what I did, that one NVMe slot was somehow preferred, and everything else looks like was shared. So, I gave up on that (I just keep that NVMe in that GPU shared slot).

With those 30 mins plots that I have right now, I think it would be difficult to push that box even more.

Also, seeing that my box is still faster from what @Bones has, I am really happy with what i have, as he is really tough to beat :slight_smile:

Did you try that Primo Cache? It may not speed too much, but will save about 50% of writes on your tmp2 drive.

Sorry, I didn’t notice that you only use -t. Still, with Primo, you could do -t d:\tmp1 -2 d:\tmp2, and that would let you save 50% of temp2 writes (about 600GB, as t2 is writing about 1.2 TB). If you have both pointed to the same folder, then some writes from temp1 potentially would interfere, as those are bigger, and kept for longer.

I am using this one which seems to top out at 500Mbytes per second and 50minutes per plot.
It is in a x16 slot but I doubt it uses it, especially as it looks like the extra pcie lanes can be snapped off!

Mine is below. It gives me full speed (PCI3 - that is what my CPU does):

No, I’ve yet to play with primo, don’t want to slow my plotting, I will do at some point, if I haven’t fixed my system issues on my farmer I’ll probably slow my plotting and try it out.

Keen to hit 14 days expected time to win.

So you are saying right now that you only use 1 NVME Samsung 970 Pro Plus with a 128GB of RAM. for the RAM part did you use RamDisk method like you were saying " RAM for tmp2 " do we have anything to worry if we make our RAM into tmp2?

I’m thinking on not going to PCie-to-NVMe road. Might as well just go with the RAMdisk method like you did. Probably i’ll just take sometime to move my VM to another machine. But i dont think the VM affecting lots because it doesn’t use much resource for the VM as the VM doesn’t continously using full resource.

I do also seen on youtube they made on Samsung 570 with 1TB RADI0 , but in my case i couldn’t do any RAID as i only have 1 NVME Slot and the PCIE lanes eventhough i’m adding another NVME that might not wil be they way i wanted (like u are saying it doesn’t even speed up) and will cost more than adding 128GB of RAM on the rig

I’m sorry i dont quite get it what you mean by OG Plots i do also seen they saying NFT Plot etc but didn’t have time yet to get into until you mentioned it again

I’m abit confused is there any reason to include the t2 also while u only use 1 NVME at a time isn’t t1 would be sufficient?

Since you are using just one NVMe, I don’t think that Primo Cache will slow you down.

Thanks! , if i’m not mistaken the PLOTTING is actually bind on our wallet right? So it doesn’t matter what machines we plot in as long as we use same wallet then we dont need to plot a new one entirely . Is it correct?