Gigahorse GPU Plotter, GPU Farmer, Plot Sink: Boost Your Chia Farm by 47.2%! Plot in <2 Minutes

Chia Farming & Harvesting
776

Well there is a Windows ZIP file here, I do hope its native Windows as I plan to install soon.

777

Yes it’s native windows, been available for a long time now.

2 Likes
778

Thanks for the super quick reply. Installed my GPU last night.

1 Like
779

For GPU farming should some environment variable ( ```
CHIAPOS_MAX_CUDA_DEVICES, CUDA_VISIBLE_DEVICES, CHIAPOS_MAX_GPU_DEVICES…) be set, otherwise gigahorse start always CPU farming?

780

How can I know, if I habe GPU farming, or CPU farming? I have set log in cinfig.yaml to INFO, but I can’t see any info about this point.

1 Like
781

I just start the farmer and it automatically uses the GPU, I don’t know if I understood your question though.

782

The point is, how do you know that the farmer uses GPU? I have 3 gigahorse farmers, and the behavior are the same: in the log I can’t find any info if the farmer use GPU or CPU, and When I observe the GPU usage in taskmanager, the GPU is not used at all.

783

Update

784

Yes, it works fine with version 1.8.1.giga12. Although this point isn’t mentioned in the release (local GPU, not remote compute), it is solved in the new version.

785

I’d like to separate my farming and plotting machines. Right now the same machine is simultaneously plotting two plots in parallel with two GPUs and farming.

Which single GPU can I use for ~5 PB C7 farming?

786

A 3060ti or higher. You may need to overclock it slightly.

787

When you find a partial there’s a INFO message like this:

2023-05-20T04:04:17.580 harvester chia.harvester.harvester: INFO     Found proof: K = 32, C = 7, is_compressed = True, used_gpu = True, fee_rate = 3.125 %

cat ~/.chia/mainnet/log/debug.log | grep "Found proof"

Otherwise look at nvidia-smi or VRAM usage in general.

Until plot filter 256 a 3060 will work fine.

788

Hi, anyone run into a problem with Linux Ubuntu desktop where the window where I’m plotting just seems to close down?
Is there some sort of logging for the gigahorse plotter?

I’m also farming on the same machine, so maybe could be an issue?
But sometimes it closes and has done about 10 plots, and sometimes 30, 40+ plots.
So seems to be a bit random

789

Running out of RAM ?

790

Plotting with two GPUs with two instances of Gigahorse in parallel and simultaneously farming with Flexfarmer slow down the plotting process. Without Flexfarmer, I get two plots in 180-210 secs each. With Flexfarmer running, this time increases as Flexfarmer is running, eventually reaching 700 secs per GPU. The GPUs don’t reach critical temperatures, so it’s definitely not an instance of thermal throttling.

I do as follows:

  1. Run Flexfarmer
  2. Start plotting on GPU0 on CPU0 with RAM of CPU0 using numactl -N 0 -m 0 ./gpu0.sh
  3. Wait 400 secs and start plotting on GPU1 on CPU1 with RAM of CPU1 using numactl -N 1 -m 1 ./gpu1.sh so that plots don’t offload simultaneously to the SSD

Why does that happen? How can I troubleshoot it?

791

Usually the answer is “bandwith of the MB chipset”,

What kind of chipset and motherboard you have?

792

Supermicro X9DRi-LN4F+ on Intel C602.

793

Expansion Options

  • PCI SupportYes

  • PCI Express Revision2.0

  • PCI Express Configurations 2x4, 4x2, 8x1

  • Max # of PCI Express Lanes8

X8 @ Pci 2.0 = 4 Gbps/s
X4 @ Pci 2.0 = 2 Gbps/s.

I Think thats a chipset bandwith limitation, meaning you are using all the bandwith from different lanes ( sata pci, nvme slot / nvme on pci slot).

My best 2 cents.

794

I think you’re mistaken somewhere, because the motherboard has 3 x16 3.0 slots, 2 x8 3.0 slots and a 1 x4 3.0 slot.
And nvtop definitely shows that both GPUs are running at 3.0x16.

I think what you should be looking at is the amount of PCI-E lanes supported by CPU, because the PCI-E slots are handled by the CPUs and each of my 2630v2 have 40 lanes of PCIe 3.0

And it doesn’t explain why the performance is reduced gradually as plotting continues.

What I’ve found is that it seems that Phase 4 begins to take more and more time.

795

Can you share out more info, perhaps make a spreadsheet with the issue? Plot, plot times per phases - to see what’s what. Also, understand that variation in time can/will be observed ( as there’s no indication of the increase mentioned ) getting more data would help. I.e.: operational definitions need to be established first.

Also, you’re exact setup would be helpful. It looks like you have a target of a 5 PB farm, but I don’t see ( beyond the MB and such ) the setup. I.e.: OS, Memory, SSD(s), and are you trying to plot/farm at the same time ( on the same machine ), etc.

Thanks.

1 Like