madMAx43v3r/chia-gigahorse

Gigahorse Compressed Plots

Since Chia still has a plot filter of 512 until some time in 2024, the farming capacity is twice that until then.

MMX testnet10 and mainnet will have a plot filter of 256.

Join the Discord for support: https://discord.gg/BswFhNkMzY

RAM / VRAM requirements to farm

When you mix different K size and C levels, only the higest RAM / VRAM requirement will apply.

FlexFarmer

Using FlexFarmer is an alternative to running Gigahorse Chia node / farmer / harvester. It does not require running a Node, but you have to switch your NFT to flexpool.

See here for more info on how to use FlexFarmer: https://www.reddit.com/r/Flexpool/comments/11a2mqe/flexfarmer_v230_gigahorse_madmax43v3rs_compressed/

Right now only CPU and Nvidia GPU farming is supported on Linux, no Windows support yet.

Note: The fee is taken from the partials with FlexFarmer, instead of the 0.25 XCH farmer block reward, so there is no fear of paying too much fee if unlucky.

Chia Gigahorse Node / Farmer / Harvester

In the release section you can find Chia Blockchain binaries to farm compressed plots created with the new plotters provided in this repository.

The compressed plot harvester and farmer are not compatible with the official Chia node, it only works together with the Gigahorse node. However it's possible to use a wallet from the official Chia repository, instead of the Gigahorse binary wallet.

Both NFT and OG plots are supported, as well as solo and pool farming (via the official pool protocol). Regular uncompressed plots are supported as well, so you can use the Gigahorse version while re-plotting your farm.

The dev fee is as follows:

• 3.125 % when using GPU(s) to farm compressed plots
• 1.562 % when using CPU(s) to farm compressed plots
• 0 % for regular uncompressed plots

When you find a block there's a chance the 0.25 XCH farmer reward is used as fee, this is a random process. In case of CPU farming it's 1 out of 8 blocks on average, and for GPU farming it's 1 out of 4 blocks on average.

When the fee is paid from a block, you will see a log entry like this:

full_node: WARNING  Used farmer reward of block 2187769 as dev fee (3.125 % on average)

It will show the block height as well as the average fee that applies, depending on if the proof was computed via CPU or GPU.

Pool Partial Difficulty

When farming NFT plots on a pool it is recommended to set the partial difficulty to 20 or more, otherwise your harvester will be overloaded with computing full proofs.

The chance of having to compute a full proof is roughly 1 / (2 * difficulty). The cost of computing a full proof is 8 (for C6+) or 16 (for C5 and lower) times that of a quality lookup.

For example:

• Difficulty 20 (at C6+): 8 / 40 = 20 % compute overhead
• Difficulty 100 (at C6+): 8 / 200 = 4 % compute overhead
• Difficulty 1000 (at C6+): 8 / 2000 = 0.4 % compute overhead

Plot Reload Interval

It is recommended to increase your plot reload interval to at least 3600 seconds in config.yaml:

harvester:
  plots_refresh_parameter:
    interval_seconds: 3600

The default value of 120 sec will cause too much CPU load with large plot counts.

Usage Linux

Make sure to close any other instances first:

chia stop all -d

Or close the Chia GUI if you are running it. Otherwise you cannot start the Gigahorse version.

Using the Gigahorse binaries is pretty much the same as with a normal Chia installation:

cd chia-gigahorse-farmer
./chia.bin start farmer (full node + farmer + harvester)
./chia.bin start harvester (remote harvester)
./chia.bin show -s
./chia.bin farm summary
./chia.bin plotnft show
./chia.bin wallet show
./chia.bin stop all -d

Note the usage of ./chia.bin ... instead of just chia ..., this is the only difference in usage with Gigahorse.

Alternatively, you can . ./activate.sh in chia-gigahorse-farmer to be able to use chia ... commands instead of ./chia.bin ....

Usage Windows

Make sure to close any running Chia GUI first, otherwise you cannot start the Gigahorse version.

To start the farmer double click start_farmer.cmd in chia-gigahorse-farmer, this will open a terminal where you can continue to issue commands. To only open a terminal without starting anything you can use chia.cmd. To stop everything you can use stop_all.cmd.

The usage in general is the same as normal chia:

chia.exe start farmer (not: chia start farmer)
chia.exe start harvester (not: chia start harvester)
chia show -s
chia farm summary
chia plotnft show
chia wallet show
chia stop all -d

Official GUI + Gigahorse

You can start the official Chia GUI after starting Gigahorse in a terminal, however it needs to be the same version. It will still complain about version mismatch but when the base version (like 1.6.2) is the same then it works.

When you close the GUI everything will be stopped, so you need to restart Gigahorse in the terminal again if so desired.

Installation

Note: There is no need to re-sync the blockchain, Gigahorse node will re-use your existing DB and config. Even the old v1 DB format still works.

Linux

sudo apt install libgomp1 ocl-icd-libopencl1
tar xf chia-gigahorse-farmer-*.tar.gz

Windows

Just unzip the chia-gigahorse-farmer-*.zip somewhere.

You might also have to install latest Microsoft Visual C++ Redistributable: https://aka.ms/vs/17/release/vc_redist.x64.exe

Limit GPU / RAM usage

Please take a look at:

• How to limit GPU usage
• How to limit RAM usage

Note: When changing environment variables you need to restart the Chia daemon for it to take effect: ./chia.bin stop all -d or chia.exe stop all -d

Remote Compute

It's possible to move the compute task to another machine or machines, in order to avoid having to install a GPU or powerful CPU in every harvester:

To use the remote compute feature:

• Start chia_recompute_server on the machine that is doing the compute (included in release).
• export CHIAPOS_RECOMPUTE_HOST=... on the harvester (replace ... with the IP address or host name of the compute machine, and make sure to restart via chia stop all -d or stop_all.cmd on windows)
• On Windows you need to set CHIAPOS_RECOMPUTE_HOST variable via system settings.
• CHIAPOS_RECOMPUTE_HOST can be a list of recompute servers, such as CHIAPOS_RECOMPUTE_HOST=192.168.0.11,192.168.0.12. A non-standard port can be specified via HOST:PORT syntax, such as localhost:12345. Multiple servers are load balanced in a fault tolerant way.
• CHIAPOS_RECOMPUTE_PORT can be set to specify a custom default port for chia_recompute_server (default = 11989).
• See chia_recompute_server --help for available options.

To use the remote compute proxy:

• Start chia_recompute_proxy -n B -n C ... on a machine A. (B, C, etc are running chia_recompute_server)
• Set CHIAPOS_RECOMPUTE_HOST on your harvester(s) to machine A.
• chia_recompute_proxy can be run on a central machine, or on each harvester itself, in which case A = localhost.
• See chia_recompute_proxy --help for available options.

When using CHIAPOS_RECOMPUTE_HOST, the local CPU and GPUs are not used, unless you run a local chia_recompute_server and CHIAPOS_RECOMPUTE_HOST includes the local machine.

CPU based Compute Servers

For CPU based compute it's important to increase CHIAPOS_MAX_CORES on the harvesters to achieve full CPU utilization on compute servers. Because CHIAPOS_MAX_CORES is the maximum parallel requests made from a harvester to recompute servers, and a request is processed on a single CPU core only. By default CHIAPOS_MAX_CORES is the number of phsical CPU cores on the harvester.

For example if you have a single compute server with 32 CPU cores, you should set CHIAPOS_MAX_CORES on the harvesters to 32. The sum of CHIAPOS_MAX_CORES accross all harvesters should be greater or equal to the sum of CPU cores on all compute servers. In case of low number of harvesters (ie. 1-3) you should set CHIAPOS_MAX_CORES to the number of CPU cores on your compute server.

Known Issues

• AMD GPU getting stuck in Linux, workaround is: watch -n 0.1 sudo cat /sys/kernel/debug/dri/0/amdgpu_pm_info

Fixed in latest version

• Harvester crashing randomly after some time
• Multiple OpenCL GPUs not working together when farming
• Multiple GPUs not being fully utilized when farming

Gigahorse GPU Plotter

You can find the GPU plotter binaries in cuda-plotter.

They support plotting for Chia as well as MMX.

CPU Plotter

You can find the CPU plotter binaries in cpu-plotter.

They support plotting for Chia as well as MMX.

Farming Benchmark

To test how many plots you can farm on a given system you can use the ProofOfSpace tool in chiapos.

Plot Sink

Plot Sink is a tool to receive plots over the network and copy them to multiple HDDs in parallel.

You can find binaries in plot-sink

See also the open source repository: https://github.com/madMAx43v3r/chia-plot-sink

Docker Usage

The Dockerfile file uses multiple build stages to support 3 different applications CPU-Only, AMD-GPU, NVIDIA-GPU.

Each image provides a volume for /data which you can override with your own volume or a mapped path to customize the storage location of the node data.

The default behavior of the container is to look in /data for an existing db/config and use it. Otherwise it will generate a fresh config and start syncing the node from scratch.

You can set which services to run with the CHIA_SERVICES environment variable.

Docker Run Examples:

-e CHIA_SERVICES="harvester"
-e CHIA_SERVICES="node farmer-only"
-e CHIA_SERVICES="node farmer-only wallet"

Docker Compose Examples:

environment:
  - CHIA_SERVICES="harvester"

environment:
  - CHIA_SERVICES="node farmer-only"
  
environment:
  - CHIA_SERVICES="node farmer-only wallet"

CPU-Only

Docker Run Example:

docker run --rm -it -v /path/to/.chia:/data chia-gigahorse

Docker Compose Example:

version: '3'
services:
  node:
    image: chia-gigahorse
    restart: unless-stopped
    volumes:
      - /path/to/.chia:/data

AMD-GPU

Docker Run Example:

docker run --rm -it --device=/dev/kfd --device=/dev/dri --group-add video --group-add render -v /path/to/.chia:/data chia-gigahorse-amd

Docker Compose Example:

version: '3'
services:
  node:
    image: chia-gigahorse-amd
    restart: unless-stopped
    group_add:
      - video
      - render
    devices:
      - /dev/dri:/dev/dri
      - /dev/kfd:/dev/kfd
    volumes:
      - /path/to/.chia:/data

Note: - render in group_add might need to be removed, depending on your system.

NVIDIA-GPU

Docker Run Example:

docker run --rm -it --runtime=nvidia -v /path/to/.chia:/data chia-gigahorse-nvidia

Docker Compose Example:

version: '3'
services:
  node:
    image: chia-gigahorse-nvidia
    restart: unless-stopped
    runtime: nvidia
    volumes:
      - /path/to/.chia:/data

Note: for nvidia you also need the NVIDIA Container Toolkit installed on the host, for more info please see: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker