What is the best setting for madmax

Bladebit with MMX v2.0.0-beta1 mmx Latest

Works but, for me, has some stability issues over longer periods. Also try --cache 116G to 120G more is better (as long as u have it free).

1 Like

On my box, Bladebit with cache 116G took 73 min/ plot. So Madmax is a better choice for me with 47 min/ plot, 30 plots/day.
I think, to get better results with bladebit, I need 512 GB ram at least.

1 Like

Well, I just fired up my i5-10500 and, for my 1st run ever on it, got 39.9 min. Looks as if something in your setup is off a bit.

Questions>

  • can you show the plotting cmd line you used?
  • how is you nvme formated?
  • What is your memory speed?
  • Which of your nvmes are you using?
1 Like

./bladebit -t 14 -f <farmer_pub_key> -c <pool_contract_address> diskplot -cache 116G -t1 <500 GB evo 980Pro> <output_dir>
nvme is NTFS formatted
Memory speed 2933 (intel core i7 10700) DDR4
Samsung evo 980 Pro 500 GB.

I open command line with admin and change directory to bladebit. Then input the BB settings line.
Am I missing something?
Will you please write your setting line?

Does taskmgr show that your CPU is being hammered / pinned?
Or are you seeing idle time?

CPU performance was %100.

Madmax finishes a plot in 47 min with simultaneous copying of the file to HDD.

My procedure>

  1. Format your plotting nvme with 64K allocation size < important :stuck_out_tongue_winking_eye:
  2. rt-clk on start > ā€œWindows Powershe4ll (admin)ā€
  3. enter the change dir cmd (below)
  4. enter the plot cmd (below - choose your ā€˜nā€™)
  5. Iā€™ve found just less one thread from the total number available works best

Iā€™ve fine tuned the amt of cache that should be set after reading a bit more. So, using approximate amts, starting with 128GB > less 8GB for the OS, less 5GB for the program itself, less 4Gb for the ā€˜heap sizeā€™ (shown when you 1st start the program and is different if you use a different bucket size other than 256)ā€¦now weā€™re at 111GB > so I set 110GB for the cache. Perhaps previous advice was too aggressive using too much resource.

For this test, I just had the plots left in the plot ssd, but normally I send them to a separate larger ssd to get rid of them, then copy from that ssd to my HD later as that temp ssd space gets fuller.

Commands at the Powershell prompt>

cd ā€œc:\bbā€ (where I have the bladebit .exe file)

.\bladebit -f xxxxx -p yyyyyy --mmx --threads 11 -n 3 diskplot -a --cache 110G -t1 d:\ d:\

2 Likes

Thank you for your help. I will try that ASAP.

Remember, Iā€™m plotting mmx, not Chiaā€¦but should be :rofl: similar!

1 Like

Note that bb doesnā€™t work that same way as MM. It seems to use the destination as the build area for the plot, meaning an ssd destination is essential for fast plotting. Using a HD slows those last plot construction plotting stages to slow down something considerable, and ruining plot times.

2 Likes

OK, I have 2 TB evo 970 Plus m2. I will set that as destinationšŸ˜‰ It will hold 18 plots.

1 Like

When your making k32ā€™s correct?

Taking your advices, now I like that Bladebit. :+1:
* I reformatted 500 GB m2 ssd with 64 K allocation.
* added -a after diskplot
* changed destination to fast nvme ssd.
Result: phase 1: 15 min.
phase 2: 3 min.
phase 3: 16.8 min.
Total: 35.3 minute. (40 plots/day) With madmax it was 47 minutes. :slightly_smiling_face:

2 Likes

Sorry what do you mean?

Fuzeguy

Thank you so much. I would not try BB again, if you would not insist on me trying again. :pray::slightly_smiling_face:

1 Like

ā€œTo Boldly Go Where Few Men Have Gone Beforeā€¦BladeBit Disk V2ā€ :rofl: :rofl:

1 Like

Only 18 plots fit onto a 1.8tb (2tb unformated) driveā€¦

Correct. You should make 18 plots, than stop BB and copy the files to HDD. So, no difference with MM, even takes more time. Madmax finishes a plot and copies one plot simultaneously to hdd in 47 minutes? If I use BB one plot takes 35 minutes and seperate copy to HDD takes 20 minutes total 55 minutes.

Sorry, I didnā€™t use BB, but I think that the final building process is similar to MM. What I mean is that while plot is being assembled (final stage), it doesnā€™t have ā€œ,plotā€ extension but rather something like ā€œ.tmpā€. If this is the case, you can have a batch file that looks for files with plot extension (say every minute or so), if it finds one, it can move it safely off of the final NVMe to whatever final destination you have (over the network, or to a local HD).

With such a simple batch program, you really donā€™t need to stop the plotting at all.

1 Like

Hi, didnā€™t know that processor existed.
Nevertheless, with my 8C/16T i7-10700, 2TB SSD and 128GB RAM Iā€™m using Madmax (standalone) plotter with:

chia_plot -n -1 -r 12 -t /mnt/firecuda-2TB/tempdir/ -2 /tmp/ramdisk/ -s /mnt/firecuda-2TB/stagedir/ -d /mnt/hdd/ -c pool-key -f farmer-key

Other parameters left to default.

The /mnt/firecuda-2TB/ā€¦ are directories on the SSD, /mnt/HDD is the drive the plots need to land on.
/tmp/ramdisk isā€¦ wellā€¦a 110G ramdisk.

The -s parameter moves finished plots to SSD space instantly, releasing the ramdisk, after which a new plot is started right away and the finished plot is automatically moved to HDD while the next one is running.
So no need to use -w or plot in batches of 18 with copying the plots to HDD in between. That adds up plotting and copying time, so not very attractive.

Plot times are around 36 minutes including the copy to HDD and non-stop;-)

I donā€™t know bladebit v2, will look into that sometime. It does not have a similar -s staging directory option as Madmaxā€™s plotter to release RAM asap?