Parallel plotting is not required to fully utilize your resources with this plotter.
If i want to scale it, better to buy cheap multiple old servers with 128gb ram and build a farm ?
If those are 16 core servers, 128GB RAM would be enough to do 75% of plotting in RAM.
Anyone have any idea definitively how much tmp space it needs on disk (not using ramdisk)?
Check out this thread: required space for ramdisk plotting only? Ā· Discussion #134 Ā· madMAx43v3r/chia-plotter Ā· GitHub
1 Like
Any idea why it fails after Table1 with ucrtsbase.dll error as faulting module in eventvwr? Using a multi-cpu system.
Sorry, I donāt know really. I am using it on 3 different machines, a 1 cpu server and 2 desktops. Works pretty much the same way everywhere. Try to download the newest release (assuming u havenāt already) alpha 0.4. Make sure u r running latest chia .7
They said they fixed a bug with crashing.
Also, make sure your dirs do have the minimum required amount of space.
Check out the Git /issues section for more info.
How do you run it? What parameters r u using?
I couldnāt get it to run at all until 0.0.4 now it runs. I havenāt had a crash after table 1 though, the error I had with prev verisons was it not beginning plotting at all.
Prior to 004 it crashed after the first few lines. Also referencing that dll. Now it crashes some steps later.
Can you say more about your setup?
This on Windows Server 2016. Same behaviour with various disks (Ramdisk/NVME/SAS/HDD):
I think multi-CPU is what Stotik does not like.
And thats the output of chia_plot.exe (Then it crashes):
Final Directory: F:
Number of Plots: 1
Process ID: 8820
Number of Threads: 4
Number of Buckets: 2^7 (128)
Pool Public Key: Mykey
Farmer Public Key: Mykey
Working Directory: B:
Working Directory 2: G:
Plot Name: plot-k32-2021-06-12-11-52-f3001f06cc15aedd4115fab6cf9a6113f52fe241af15f7050235d11b2a39f6ba
[P1] Table 1 took 129.234 sec
With this Error in eventLog:
Faulting application name: chia_plot.exe, version: 0.0.0.0, time stamp: 0x60c38f6a
Faulting module name: ucrtbase.dll, version: 10.0.14393.3659, time stamp: 0x5e9140a1
Exception code: 0xc0000409
Fault offset: 0x000000000006de4e
Faulting process id: 0x564
Faulting application start time: 0x01d75eea14452e6e
Faulting application path: C:\Users\Administrator\Desktop\madMAx43v3r_chia-plotter_win_build_v0.0.4\chia_plot.exe
Faulting module path: C:\Windows\System32\ucrtbase.dll
Report Id: af68249e-ef48-4804-89e0-9e91b9e7e9ee
Faulting package full name:
Faulting package-relative application ID:
Not sure how it runs on Windows Server. I have Win 10 Enterprise on all my systems.
Do you run in from the plot.cmd file or did u make a batch with preferred parameters?
I run it from a batch, here:
chia_plot.exe -p PKEY -f FKEY -n -1 -t F:\ -2 R:\ -r 24 -u 128
where F: is a 1tb 980 PRO NVMe and R: is a 112 GB ramdisk
Regardless of how I launch it, it behaves like above.
Stotik itself refenced another build that the ones with dual-xeons may try to use:
Dual Xeon on Windows version doesn't work (kills process) Ā· Issue #126 Ā· madMAx43v3r/chia-plotter Ā· GitHub (Scroll down).
Works better for me as I came past T1. However T2 took ages. Testing now with more Thread and Bucket size.
Update on what I have tried, in case it helps anyone, or in case anyone has any ideas for me.
-Machine is Dell R720 with Dual E5-2650V2 (8c/16t), 320GB of DDR3 ram, 8x 3TB SAS drives, Perc700. Operting system running on first SAS drive.
-All using windows powershell and Stotkis 0.04 madmax fork. Ram drive in use - ImDisk.
-Plotting one madmax thread to ramdrive only - 10000 seconds.
The following schemes are parallel plotting, two plots, with a stagger to keep P1 and P3, using a separate SAS drive per process as temp1 and using one ramdrive (with two folders) as temp2:
-Plotting 2 madmax processes staggered, approximately 13500 seconds (for two plots), but this waxes and wanes as its impossible to keep the P1 and P3 from crashing into each other all the time.
-Use of primocache. No benefit though seemed to help a little with copying the plots away to my farmer machine on 1gbit LAN. Possibly crashing into the plotter somewhat less, worth spending money on? Not sure.
-Use of processor affinity (which does hold so long as you set the number of plots to -1/continous, and keep that powershell window open) seemed to make no difference to plot times, maybe CPU use is lower at points when the P2/4 coincide.
-Use of 128, 256, 512 buckets made no difference. I havenāt tried 64 but I might do as I have the ram.
-I was setting threads as 20 per process, then switched to 32, it seems to use whatever youāve got, this made no difference.
I am concerned that others with similar hardware are getting well down into the 4 figures plot times so the next steps are:
-Reconfigure SAS drives from individual drives to 3 drives raid 0 striped to use as temp 1ās. Can see this helping a fair bit.
-Set up two separate ram drives rather than one ramdrive with two folders - canāt see this doing much.
Use linux within windows, or a bare-metal installation. I have been looking to relocate the op system to an ssd on the built in sata controller (I am only lacking power for it) and use all the SAS drives eventually for farming once | am all plotted out.
Switch the bios to allow turbo mode on the processors, I do not feel this currently limits me as I rarely see 100% CPU (92% maybe) so I am still being limited by disk IO to/from probably the SAS rather than the ramdisk, but then I have DDR3 ram, none of the drive test programs will test my ramdisk, it doesn;t show up as a drive to them.
I just joined the merry band of people using Madmax (Stotiks, windows version in my case)
After a long time tinkering with my system I (finally) had it running at 32 plots/day with swar on the original plotter.
(3900x, 64GB3200, 3x 1TB SN750)
First Mad plot just finished in 1807 secondā¦wow ā
Used one 1TB SN750 for each temp drive, impressed with the results, running a few plots in a row now to see how the times hold up. But this could mean an improvement of 40-50% in my total plot/day output.
Iām getting my 2nd plotter tonight with 2x 2680v2 and 256GB ram.
I want to try to make two 110GB ramdisks, one for each CPU and the use 2x 1TB nvmeās as the other temps and try running 2 madplots at a time like that.
2 Likes
Sounds good. I am ethically not using any NVMEs/SSDs but it is tempting. I have just reconfigured my SAS drives to give me 3 drives striped raid 0, which should be able to max out the controller, and I am hoping for a big improvement.
Next test, just plotting 1 plot to 2 separate ram drives to see if thats faster than 1 ramdrive with two directories.
Then 1 plot with raid 0 3xsas plus ramdrive.
Then 2 plots āā āā āā
Then linux via windows/baremetal. I am determined to join the 2000-3000 second plot brigade, but trying to remember I am using about Ā£1000 worth of recycled 7 year old server not a brand new threadripper.
A stripe with three drives is not a good idea because all possible fs block sizes are not divisible by three.
That means if your stripe size is 4 kb you have to use a block size of 12 kb, which isnāt possible with any fs i know.
So if you write in 16 kb blocks, then one hdd has to write two 4 kb blocks while the other ones only have to write one 4 kb block.
1 Like
Ok so stripe with 2 or 4 drives for best efficiency?