Gigahorse Ubuntu Simple Plotting Guide


I just started to replot all my disk. My plotter and farmer are the same I have all in one.

Each plot take arround 5 or 6 min to be created but the problem is the plot transfer to the final destination take arround 20 min for each plot to be done, for example the plotter give me some messages like " Waitinf for 901112 MiB available space on: /media/xxxx" or “All destinations are busy or full. waiting…” in 24h I made only 50 plots.

sudo ./cuda_plot_k32 -C 7 -n 1 -r 1 -t /media/xxx/TEMP/ -2 /media/xxx/TEMP2/ -d /media/xxx/HDDEXT23/ -c xchxxxx -f acbxxx

What I missing? I don’t have any clue about how used Plot_sink because I don’t have any network and I don’t know how to build it exactly, also I read something about SSD Buffering but I’m not sure how to configure it. I have 3 nvme high speed available 2tb each and an Asus Hyper M.2 X16 card, Could be enough to get a good configuration.

Hardware available:

  • i9-10850K

  • 128G RAM

  • RTX 4070 ti 12G

  • 3 x Samsung 990 pro 2tb

  • 1 x Samsung 980 pro 1tb

  • Asus Hyper M.2 X16 card

Could you please in your experience recommend me what to do to improve my performance? Thanks in advance!!!

Have you tried

Multiple destinations can be specified by repeating -d:

cuda_plot_kxx -d /mnt/hdd0/ -d /mnt/

this is about my setup only with other hardware components.

R9 5950x
Raid0 3x2TB NVME
RTX 4070 TI

  • The graphics card in the first PCIe slot as x8
  • the raid adapter with 2x2TB into the 2nd PCIe slot as x4x4
  • 1x2TB NVME in the first mainboard NVME slot together with the raid adapter as Raid0 (3x2TB) as storage for the hybridplot.
  • then the 1TB NVME into the slot which is connected to the chipset of the mainboard as TEMP


./cuda_plot_k32 -C 7 -n -1 -t /media/xxx/TEMP/ -2 /media/xxx/Raid0/ -d /media/xxx/HDDEXT01/ -d /media/xxx/HDDEXT02/ -d /media/xxx/HDDEXT03/ -d /media/xxx/HDDEXT04/ -c xchxxxx -f acbxxx

results in plot times of 210 seconds with simultaneous copying and farming.
pure plotting without farming and copying is about 180 seconds.

1 Like

Thank you great explanation!!! I’ll try it and let you know.

Now everything works just fine, making every plot between 4.45 min and 4.78 min without any issue or delay with final destinations, I’m very satisfied with the results, thank you very much!!!

i am glad to hear that the times are ok. if you feel like it, you can check a few things to see if everything is really optimal.

  • sometimes I have the problem that after a reboot the RAM runs with low speed (only 2133 MHz instead of 3333 MHz).
    then the plot times are about 20-30 seconds longer
    to check this: sudo dmidecode -t memory

  • on some mainboards not all NVME slots are connected with x4 sometimes some are only connected with x1, just check the manual to make sure the right slot is used

  • if your plot times get worse after some time (30-60 minutes), a regular trim of -t /dev/TEMP is necessary

  • and last but not least a good cooling of the NVMEs is important because they are under permanent load normally from 50°C on they clock down and that costs a lot of time.
    to check this sudo smartctl -a /dev/"YourNVME"

happy replotting :slight_smile:

1 Like

I still don’t get the harvester power thing.

Can a fairly good CPU like 20000 benchmark, i7-11700, do 500TB C7 compressed plots?

What compression ratio/total plot vs CPU power/how much memory?

There are GPU VRAM and CPU RAM requirements in the link. The “harvester power thing” is that a CPU is not as efficient as a GPU in finishing the ‘compressed’ plots during farming. GPUs are not general purpose computing engines, but what they can do, they do efficiently, with less power than the same thing done on a CPU (if even possible on a CPU).

So with your CPU it can do the work, up to a point, but not power efficiently, on a certain number of requests to finish plots. There are formula and there is also a plot check tool to tell you how much can be farmed, but generally, almost any GPU is going to be better than any CPU.

For more detail search and/or visit the relevant Discords, etc.


RAM GiB require for CPU, is it regardless of farm size? The RAM size requirement is small.

As I understand, yes, it is farm size independent, as is the GPU VRAM requirement. What is important is C & K level and so determines what is needed.

GPUs can usually do what CPU do, but more efficiently because of their architecture.

Maybe for smaller farms, CPU can still be better, but the logic is that GPUs (and maybe in the future of Chia even FPGA/Asics) should always be faster and more efficient, unless the design doesn’t allow for parallelization.

At least you’re not using 10-Base-5 AKA ThickNet.
Type N or Type C were such a HUGE improvement over the original “vampire taps” (even if not originally part of the spec)!

how much does it cost per plot with this setup?

It looks like over 1$/plot to me

i can’t say exactly, because it’s not really important for me, considering my monthly energei costs, which are caused by the pure farming operation, 1-2 times replotting doesn’t matter.

my system (1.45 PiB) consumes before without GPU about 600 watt complete with router and UPS.
when plotting with simultaneous farming and plot check it fluctuated between 800-900 watts. the replott took about 6 weeks.
now i have effectively 2.12 Pib and my whole system needs around 700 watt.

from this i think you can calculate about costs and increased efficiencies if you feel like it :wink:

the 100-200 watt difference to plotting is not only the GPU. when i’m not plotting the system is undervolted & frequency downgraded, one chiplet is completely disabled. furthermore, except for the cpu fan, all other fans i used for plotting have been removed. the nvme raid card is also no longer in the system. all this adds up to about 100 watt.


you went from .4w/tib to .3w/tib that’s great

1 Like

How do you know ROI for plots?

The most efficient 0.05$/plot takes around 2 months to repay. If you plot for 0.5$, it would take 20 months. At current prices, just HDD has got 6 years ROI. Are you sure HDD survives 8 years?

I do compressed plots for other for 0.1eur per K32 plot

Hi @Chris22
i plotted some HDDs using the flexpool pool contract address (basically using your example) and C5 compression
I started farming today and I see between 1-4% of staled partials. So I planned to increase (partial) difficulty for my farm (since madmax recommends > 18 for compressed plots). Is it possible to increase difficulty? I don’t have the plotNFT login link so I cannot do it via flexpool dashboard. Any other option? Via config.yml like gigahorse?

Good question I didn’t consider this when I wrote the guide. PM me your wallet address and I’ll get it changed manually.

Hi, while creating table p1 will end sigkill. What is it?