Mad Max Plotter, some test results, some general info

@PAVoutsinas

Can you post your shellscript? Id like to see the start/delay/number per drive etc ?

Thanks

1 Like

not for me but thanks, I don’t think I have enough overhead to do more than 1 at a time and it’s running well for me just with the powershell.

Just that damn phase 3 stuck sometimes…still no final solution on that.
Have had it on windows, ubuntu, original plotter, Swar and Madmax

Just throwing my results in here. Ryzen 7 2700x, 128GB ram (using ~118GB For ram drive), Windows 10 (one of the newer flavors), with a Samsung 2TB 970 pro (?) NVME. 3100-3200s to complete the plot. I did notice the newer version of Stotiks dropped it by a tiny bit. May try Ubuntu/Linux Mint later this weekend.

Anyone tried this on a vm with esxi and just allocating most of the resources to the VM?

This should not be possible. Even with default settings, you should get at least 55 plots++ with your configuration.

Mine, 5950X (no PBO) on a single SSD (Corsair MP600 Pro - formatted as F2FS) as -t, 110G DDR4-2666 as -2, WD HDD Elements as -d. I am getting around 62 plots per day.

-r 16 -u 512 I am plotting on Ubuntu 21.04.

yes agreed.

I’m plotting on Ubuntu 20.04

3900X @4.3GHz, 32GB ram, 2x 1TB WD SN750 nvme (ext4)
Straight installation of Ubuntu nothing changed.
using Madmax I get sub 1800 sec plots, so 48+ p/day

-r 30 -u 256 -v 128 -G

@Cyrus you using exfat on the nvme?

Do you think the F2FS is improving your plot speed?

Also, have you tried greater values for -r?

I believe that F2FS should improve plotting, I didn’t test my speed with other filesystems though since the performance as appears on many benchmarks are a lot better than the default ext4 (it’s not even close). It’s comparable to XFS but without having to specify any mount option to make your SSD performs fast. It’s the filesystem that’s built specifically for SSD from the ground up.

I have tried a greater value of -r, but I found that -r 16 gives me the best result.

1 Like

I am using btrfs and it is really fast. Hmm wonder which would be better…

I’ve been playing with the new bucket options on Madmax. I found a good improvement combining two different bucket sizes.

I’m running 5950x , 128GB Ramdisk with a Sabrent Rocket 4 NVMe.

Setting the first two stages as 512 and the later stages as 256 lowers my time per plot by 7 per cent.

Now, right now I’m running Windows on my rig. I have flip flop between windows and Linux due to a network bottleneck I get in Linux that’s makes it hard to offload plots to my farming machine. But my Windows speeds went down from about 1650 to 1500 seconds with this change. In Linux, my best time previously was 1165 seconds. So - this might be enough to get down to 1080 seconds ok Linux?

Would love someone to test how it plays.

My current config is with 16 threads only on Windows, using 512/256 buckets.

My previous best on Linux was using 30 threads and 256 buckets.

Ive been doing some testing all in ubuntu. I have pretty much the same rig as you 5950x, 128gb ddr4 3333, sabrent 4 rocket plus. I’m struggling to beat 1200s. Once or twice on the first run I’ve gotten down to 1160s (when no copy file operations were ever running, and the ssd was freshly trimmed). That was with -R 16, -U 512 (and before the -V option). So far I’ve not found any improvement with -V. Still playing around tho. Let me know what you find in Linux. Windows is so bad I’ve totally given up on it.

Very interesting. Why do you need more RAM with 32 cores?
I got 2x E5-2695v4 with 36 cores and wanted to try 128gb RAM. What will happen?

Went from 13 plots/day on swar to 60minute plots on Madmax windows. My drives are full so haven’t run a full day of madmax to see what total plots/day are with destination copy. This is a huge performance increase for plotting times even without ramdisk. If I was plotting thousands of plots the cost of 128GB of ram would make sense, but I’m not.

R5 2600x (6core 12thread)
went from 32GB ram to 16GB ram after switching to madmax
2TB tforce z440 pcie4.0 nvme temp drive
Asrock B450m Steel Legend (so pcie 3.0 m.2 connection)
The destination is temp drive then robocopy separately after plot is done.

I really wish Madmax was available before making my plotting hardware upgrade. I would have gone with a pair of sas 15k drives in raid 0 over the cost of the SSD. Could have upgraded my storage that much more. This is a plex server/ file sharing box that also farms eth and was already online 24/7 before chia.

If I understand right, that means this isn’t good enough.

4 x DDR4 DIMM sockets supporting up to 128 GB of system memory

I won’t be able to make a RAM disk big enough for a madmax tmp dir with that motherboard. Is that right?

Hi.
Can you help me? I want to by 5800x /32gb for chia.
Now i have 2pct 2Tb NVME
Can you explain me settings?
Thanks

Since it has 4 x DDR4 memory slots, you can put 32G modules and get 128G for ramdisk

That’s a great rig, you can even go for 256GB ram and use ramdisk for temp1 as well

Hey guys,
Im new to forum, or at least new at signing up and commenting,
I’ve been following all the discussion and tried 12321321321321321 things, but still struggling with my plotter.
I’ve got a
TR 2970WX 24c 48t @ 4.0GHz,
Quadchannel 4x16 3200 Mhz (total 64 GB ram)
1x2TB FireCuda 520 Red
Windows 10
Madmax last version 0.0.5
I never get less than 2800 secs, whatever the buckets, threads and everything I try.
I am plotting on NVME not on RAM since I dont have enough. If I open 2 windows in parallel it goes to hell 7k secs

When It’s copying the plot to the HDD it takes 100% of my SSD and my CPU goes to 0 for all the time that is copying…
I already manually trimmed, optimized, with cmdlet, I don’t know what else to do, HEEEEEEEEEELP :frowning:

im doing buckets on 256 (standard last madmax version) and threads tried all the combinations you can imagine… 12 14 16 26 30 34 48… ALL, as well as buckets, tried 128 512 256…

If you need more information, please let me know

Can u paste here there command you run? -t etc etc ?
Thanks!!!

If you are using that one nvme drive for both temp1&2 I think that is your bottleneck.
Add a small but fast(sustained write) nvme as temp2.

I run everything on default, no special params apart from the number of threads,
./chia_plot -n -1 -r 24 -t /media/mehdi/chia_980_pro/ -d /media/mehdi/chia_dst_expansion_5/
followed by the farmer and pool keys.

2 Likes