Skip to content

Instantly share code, notes, and snippets.

@SlowestTimelord

SlowestTimelord/chia-gpu-cpu-plotting-options.md

Last active April 3, 2023 17:34
Download ZIP

Revisions

  1. SlowestTimelord revised this gist Apr 3, 2023. 1 changed file with 1 addition and 61 deletions.
    62 changes: 1 addition & 61 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -1,61 +1 @@
    # Available Plotters
    - Chiapos (included in [Chia full node](https://github.com/Chia-Network/chia-blockchain))
    - madMAx chia-plotter (included, [source](https://github.com/madMAx43v3r/chia-plotter))
    - Bladebit ramplot/diskplot (included, [source](https://github.com/Chia-Network/bladebit))
    - Bladebit CUDA ([alpha](https://github.com/Chia-Network/bladebit/tree/cuda-compression))
    - Gigahorse CUDA ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cuda-plotter))
    - Gigahorse CPU ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cpu-plotter))

    # Hardware requirements for different plotters (K32)

    | Plotter | GPU (min VRAM) | RAM (GiB) | SSD (GiB) | Plot Size (GiB) | Plot Time |
    |----------------------------------|----------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 78.1-101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 78.1-101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 78.1-101.3 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (1/2 partial RAM) | Nvidia (4GB) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CUDA (1/4 partial RAM) | Nvidia (4GB) | 64 | 250 | 71.3-84.2 | 8-20 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Chiapos: most dependent on SSD sustained write speed
    - madMAx, Bladebit ramplot/diskplot, Gigahorse CPU: most dependent on CPU and SSD sustained write speed
    - Bladebit CUDA, Gigahorse CUDA: most dependent on PCIe generation and GPU (e.g. PCIe3 is maxed out with a 3060ti)
    - Gigahorse CUDA 1/2 and 1/4 partial RAM: most dependent on RAM speed and SSD(s) sustained write speed. 1/4 partial RAM can have SSD load split across two devices.
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - madMAx and gigahorse CPU plotter scales with more CPU threads
    - Supported Nvidia GPUs: GTX 1000/1600 series, RTX 2000/3000/4000 series
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
    - GPU plotters will require a fast SSD buffer drive to write plots to (they generate plots faster than a single HDD can write)
    - Chiapos is slow but is well suited for plotting directly to HDDs, especially in parallel due to low CPU and RAM requirements
    - Farming Gigahorse compressed plots is only compatible with the Gigahorse farmer and has an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)

    ## Larger K-sizes
    - As a rule of thumb, K33 will have double the RAM and SSD requirements as K32.
    - There are no foreseeable plans to move to K33 minimum plot size.
    - Most plotters support K32, K33, K34
    - Currently, Bladebit GPU does not support K33+

    ## Expected future developments
    - AMD GPU support
    - Larger k-size support across all plotters
    - Bladebit CUDA integration into Chia GUI
    - Bladebit CUDA (partial RAM) RAM requirement reduced to 128GB
    - Gigahorse CUDA (partial RAM) RAM requirement reduced to 64GB
    - Harvesting support for compressed plots (no GPU needed on harvesters)
    - Bladebit ramplot RAM requirement reduced to 256GB

    # Additional Links
    - [Chia Blog - GPU performance benchmarks](https://www.chia.net/2023/01/20/gpu-plotting-is-real-and-very-fast/)
    - [Chia Blog - Plot compression overview](https://www.chia.net/2023/01/20/plot-compression-is-here/)
    - [madMAx Gigahorse compression levels and requirements](https://github.com/madMAx43v3r/chia-gigahorse)
    - [chia-plot-sink](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/plot-sink) - tool to move plots to destination drives in parallel to be used with Gigahorse
    - [Plow](https://github.com/lmacken/plow) - tool to watch for new Chia plots and move to destination drives in parallel (local or network).
    This page has moved to https://xch.farm/plotting/
  2. SlowestTimelord revised this gist Feb 15, 2023. 1 changed file with 15 additions and 13 deletions.
    28 changes: 15 additions & 13 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -9,25 +9,27 @@
    # Hardware requirements for different plotters (K32)

    | Plotter | GPU (min VRAM) | RAM (GiB) | SSD (GiB) | Plot Size (GiB) | Plot Time |
    |------------------------------|----------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 78.1-101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 78.1-101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 78.1-101.3 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |
    |----------------------------------|----------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 78.1-101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 78.1-101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 78.1-101.3 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (1/2 partial RAM) | Nvidia (4GB) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CUDA (1/4 partial RAM) | Nvidia (4GB) | 64 | 250 | 71.3-84.2 | 8-20 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Chiapos: most dependent on SSD sustained write speed
    - madMAx, Bladebit ramplot/diskplot, Gigahorse CPU: most dependent on CPU and SSD sustained write speed
    - Bladebit CUDA, Gigahorse CUDA: most dependent on PCIe generation and GPU (e.g. PCIe3 is maxed out with a 3060ti)
    - Gigahorse CUDA 1/2 and 1/4 partial RAM: most dependent on RAM speed and SSD(s) sustained write speed. 1/4 partial RAM can have SSD load split across two devices.
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - madMAx and gigahorse CPU plotter scales with more CPU threads
    - Supported Nvidia GPUs: GTX 1000/1600 series, RTX 2000/3000/4000 series
  3. SlowestTimelord revised this gist Feb 10, 2023. 1 changed file with 4 additions and 4 deletions.
    8 changes: 4 additions & 4 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -14,10 +14,9 @@
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 101.3 | 1.5-6 minutes |
    | Bladebit CUDA compressed | Nvidia (8GB) | 256 | | 78.1-87.5 | 1.5-6 minutes |
    | Bladebit ramplot | | 416 | | 78.1-101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 78.1-101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 78.1-101.3 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    @@ -38,6 +37,7 @@
    - Farming Gigahorse compressed plots is only compatible with the Gigahorse farmer and has an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)

    ## Larger K-sizes
    - As a rule of thumb, K33 will have double the RAM and SSD requirements as K32.
    - There are no foreseeable plans to move to K33 minimum plot size.
    - Most plotters support K32, K33, K34
    - Currently, Bladebit GPU does not support K33+
  4. SlowestTimelord revised this gist Feb 10, 2023. 1 changed file with 1 addition and 1 deletion.
    2 changes: 1 addition & 1 deletion chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -35,7 +35,7 @@
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
    - GPU plotters will require a fast SSD buffer drive to write plots to (they generate plots faster than a single HDD can write)
    - Chiapos is slow but is well suited for plotting directly to HDDs, especially in parallel due to low CPU and RAM requirements
    - Farming Gigahorse compressed plots have an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)
    - Farming Gigahorse compressed plots is only compatible with the Gigahorse farmer and has an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)

    ## Larger K-sizes
    - There are no foreseeable plans to move to K33 minimum plot size.
  5. SlowestTimelord revised this gist Feb 10, 2023. 1 changed file with 3 additions and 2 deletions.
    5 changes: 3 additions & 2 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -46,9 +46,10 @@
    - AMD GPU support
    - Larger k-size support across all plotters
    - Bladebit CUDA integration into Chia GUI
    - Bladebit CUDA (partial RAM) with 128GB RAM requirement
    - Gigahorse CUDA (partial RAM) with reduced 64GB RAM requirement
    - Bladebit CUDA (partial RAM) RAM requirement reduced to 128GB
    - Gigahorse CUDA (partial RAM) RAM requirement reduced to 64GB
    - Harvesting support for compressed plots (no GPU needed on harvesters)
    - Bladebit ramplot RAM requirement reduced to 256GB

    # Additional Links
    - [Chia Blog - GPU performance benchmarks](https://www.chia.net/2023/01/20/gpu-plotting-is-real-and-very-fast/)
  6. SlowestTimelord revised this gist Feb 10, 2023. 1 changed file with 17 additions and 16 deletions.
    33 changes: 17 additions & 16 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -1,28 +1,28 @@
    # Available Plotters
    - Chiapos (included in [Chia full node](https://github.com/Chia-Network/chia-blockchain))
    - madMAx chia-plotter (included, [source](https://github.com/madMAx43v3r/chia-plotter))
    - Bladebit RAM/disk (included, [source](https://github.com/Chia-Network/bladebit))
    - Bladebit ramplot/diskplot (included, [source](https://github.com/Chia-Network/bladebit))
    - Bladebit CUDA ([alpha](https://github.com/Chia-Network/bladebit/tree/cuda-compression))
    - Gigahorse CUDA ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cuda-plotter))
    - Gigahorse CPU ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cpu-plotter))

    # Hardware requirements for different plotters (K32)

    | Plotter | GPU | RAM (GiB) | SSD (GiB) | Plot Size (GiB) | Plot Time |
    |------------------------------|-------------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 101.3 | ~4 minutes |
    | Bladebit diskplot | | 4 | 480 | 101.3 | ~10 minutes |
    | Bladebit CUDA | Nvidia (8GB VRAM) | 256 | | 101.3 | 1.5-6 minutes |
    | Bladebit CUDA compressed | Nvidia (8GB VRAM) | 256 | | 78.1-87.5 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB VRAM) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB VRAM) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |
    | Plotter | GPU (min VRAM) | RAM (GiB) | SSD (GiB) | Plot Size (GiB) | Plot Time |
    |------------------------------|----------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit ramplot | | 416 | | 101.3 | 4-10 minutes |
    | Bladebit diskplot | | 4 | 480 | 101.3 | 10-30 minutes |
    | Bladebit CUDA | Nvidia (8GB) | 256 | | 101.3 | 1.5-6 minutes |
    | Bladebit CUDA compressed | Nvidia (8GB) | 256 | | 78.1-87.5 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    @@ -52,6 +52,7 @@

    # Additional Links
    - [Chia Blog - GPU performance benchmarks](https://www.chia.net/2023/01/20/gpu-plotting-is-real-and-very-fast/)
    - [Chia Blog - Plot compression overview](https://www.chia.net/2023/01/20/plot-compression-is-here/)
    - [madMAx Gigahorse compression levels and requirements](https://github.com/madMAx43v3r/chia-gigahorse)
    - [chia-plot-sink](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/plot-sink) - tool to move plots to destination drives in parallel to be used with Gigahorse
    - [Plow](https://github.com/lmacken/plow) - tool to watch for new Chia plots and move to destination drives in parallel (local or network).
  7. SlowestTimelord revised this gist Feb 9, 2023. 1 changed file with 5 additions and 6 deletions.
    11 changes: 5 additions & 6 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -12,25 +12,24 @@
    |------------------------------|-------------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 128 | 220 | 101.3 | 20-45 minutes |
    | madMAx (partial RAM) | | 110 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit RAM | | 512 | | 101.3 | ~4 minutes |
    | Bladebit Disk | | 4 | 480 | 101.3 | ~10 minutes |
    | Bladebit ramplot | | 416 | | 101.3 | ~4 minutes |
    | Bladebit diskplot | | 4 | 480 | 101.3 | ~10 minutes |
    | Bladebit CUDA | Nvidia (8GB VRAM) | 256 | | 101.3 | 1.5-6 minutes |
    | Bladebit CUDA compressed | Nvidia (8GB VRAM) | 256 | | 78.1-87.5 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB VRAM) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB VRAM) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 128 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (partial RAM) | | 110 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Chiapos: most dependent on SSD sustained write speed
    - madMAx, Bladebit RAM/disk, Gigahorse CPU: most dependent on CPU and SSD sustained write speed
    - madMAx, Bladebit ramplot/diskplot, Gigahorse CPU: most dependent on CPU and SSD sustained write speed
    - Bladebit CUDA, Gigahorse CUDA: most dependent on PCIe generation and GPU (e.g. PCIe3 is maxed out with a 3060ti)
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - RAM requirement rounded up to nearest common system configuration size
    - madMAx and gigahorse CPU plotter scales with more CPU threads
    - Supported Nvidia GPUs: GTX 1000/1600 series, RTX 2000/3000/4000 series
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
  8. SlowestTimelord revised this gist Feb 9, 2023. 1 changed file with 3 additions and 0 deletions.
    3 changes: 3 additions & 0 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -26,6 +26,9 @@

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Chiapos: most dependent on SSD sustained write speed
    - madMAx, Bladebit RAM/disk, Gigahorse CPU: most dependent on CPU and SSD sustained write speed
    - Bladebit CUDA, Gigahorse CUDA: most dependent on PCIe generation and GPU (e.g. PCIe3 is maxed out with a 3060ti)
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - RAM requirement rounded up to nearest common system configuration size
    - madMAx and gigahorse CPU plotter scales with more CPU threads
  9. SlowestTimelord revised this gist Feb 9, 2023. 1 changed file with 1 addition and 1 deletion.
    2 changes: 1 addition & 1 deletion chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -33,7 +33,7 @@
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
    - GPU plotters will require a fast SSD buffer drive to write plots to (they generate plots faster than a single HDD can write)
    - Chiapos is slow but is well suited for plotting directly to HDDs, especially in parallel due to low CPU and RAM requirements
    - Farming madMAx compressed plots has an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)
    - Farming Gigahorse compressed plots have an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)

    ## Larger K-sizes
    - There are no foreseeable plans to move to K33 minimum plot size.
  10. SlowestTimelord revised this gist Feb 9, 2023. 1 changed file with 1 addition and 0 deletions.
    1 change: 1 addition & 0 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -27,6 +27,7 @@
    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - RAM requirement rounded up to nearest common system configuration size
    - madMAx and gigahorse CPU plotter scales with more CPU threads
    - Supported Nvidia GPUs: GTX 1000/1600 series, RTX 2000/3000/4000 series
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
  11. SlowestTimelord created this gist Feb 9, 2023.
    54 changes: 54 additions & 0 deletions chia-gpu-cpu-plotting-options.md
    View file
    Original file line number Diff line number Diff line change
    @@ -0,0 +1,54 @@
    # Available Plotters
    - Chiapos (included in [Chia full node](https://github.com/Chia-Network/chia-blockchain))
    - madMAx chia-plotter (included, [source](https://github.com/madMAx43v3r/chia-plotter))
    - Bladebit RAM/disk (included, [source](https://github.com/Chia-Network/bladebit))
    - Bladebit CUDA ([alpha](https://github.com/Chia-Network/bladebit/tree/cuda-compression))
    - Gigahorse CUDA ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cuda-plotter))
    - Gigahorse CPU ([source](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/cpu-plotter))

    # Hardware requirements for different plotters (K32)

    | Plotter | GPU | RAM (GiB) | SSD (GiB) | Plot Size (GiB) | Plot Time |
    |------------------------------|-------------------|-----------|-----------|-----------------|---------------|
    | Chiapos | | 4 | 240 | 101.3 | 1-4 hours |
    | madMAx (all RAM) | | 256 | | 101.3 | 15-30 minutes |
    | madMAx (partial RAM) | | 128 | 220 | 101.3 | 20-45 minutes |
    | madMAx (all disk) | | 4 | 256 | 101.3 | 30-60 minutes |
    | Bladebit RAM | | 512 | | 101.3 | ~4 minutes |
    | Bladebit Disk | | 4 | 480 | 101.3 | ~10 minutes |
    | Bladebit CUDA | Nvidia (8GB VRAM) | 256 | | 101.3 | 1.5-6 minutes |
    | Bladebit CUDA compressed | Nvidia (8GB VRAM) | 256 | | 78.1-87.5 | 1.5-6 minutes |
    | Gigahorse CUDA (all RAM) | Nvidia (4GB VRAM) | 256 | | 71.3-84.2 | 1.5-6 minutes |
    | Gigahorse CUDA (partial RAM) | Nvidia (4GB VRAM) | 128 | 150-180 | 71.3-84.2 | 4-10 minutes |
    | Gigahorse CPU (all RAM) | | 256 | | 71.3-84.2 | 15-30 minutes |
    | Gigahorse CPU (partial RAM) | | 128 | 220 | 71.3-84.2 | 20-45 minutes |
    | Gigahorse CPU (all disk) | | 4 | 256 | 71.3-84.2 | 30-60 minutes |

    ## Notes
    - Plot times depend on specific hardware and compression level (the higher the compression the shorter the plot time)
    - Higher compression (smaller plot files) also require additional consideration for farming/harvesting workload
    - madMAx and gigahorse CPU plotter scales with more CPU threads
    - Supported Nvidia GPUs: GTX 1000/1600 series, RTX 2000/3000/4000 series
    - Gigahorse GPU plotter supports cards with only 4GB of VRAM but 8GB is recommended
    - GPU plotters will require a fast SSD buffer drive to write plots to (they generate plots faster than a single HDD can write)
    - Chiapos is slow but is well suited for plotting directly to HDDs, especially in parallel due to low CPU and RAM requirements
    - Farming madMAx compressed plots has an [additional dev fee](https://github.com/madMAx43v3r/chia-gigahorse#chia-gigahorse-node--farmer--harvester)

    ## Larger K-sizes
    - There are no foreseeable plans to move to K33 minimum plot size.
    - Most plotters support K32, K33, K34
    - Currently, Bladebit GPU does not support K33+

    ## Expected future developments
    - AMD GPU support
    - Larger k-size support across all plotters
    - Bladebit CUDA integration into Chia GUI
    - Bladebit CUDA (partial RAM) with 128GB RAM requirement
    - Gigahorse CUDA (partial RAM) with reduced 64GB RAM requirement
    - Harvesting support for compressed plots (no GPU needed on harvesters)

    # Additional Links
    - [Chia Blog - GPU performance benchmarks](https://www.chia.net/2023/01/20/gpu-plotting-is-real-and-very-fast/)
    - [madMAx Gigahorse compression levels and requirements](https://github.com/madMAx43v3r/chia-gigahorse)
    - [chia-plot-sink](https://github.com/madMAx43v3r/chia-gigahorse/tree/master/plot-sink) - tool to move plots to destination drives in parallel to be used with Gigahorse
    - [Plow](https://github.com/lmacken/plow) - tool to watch for new Chia plots and move to destination drives in parallel (local or network).