How to build a fast Chia Plotter that hits 4 TB in plots per day for under $2000
One of the challenges in being a miner on the Chia Network is the generation of plots — a way to force a time and space based proof of work. If you read my previous introduction for getting started with linux and getting started with the Chia Network you will understand its not always easy to just start. I will walk through one of my builds that achieves 4 TB in plots per day ( 50 plots per day). This build will assume that you will be plotting in parallel to maximize the number of plots using a bash script, Plotman, Swar, etc.
First, as a rule of thumb when getting started with specing out a build is figuring out a general idea of our resource constrains.
- Cpu limitations: In general you want to determine the number of cores, since each plot instance uses 2 threads: # cores +# threads / 2
- Memory limitations: Each plot uses around 3389 MB in memory. Total memory in GB/ 3.4 GB
- Storage limitations: Each plot uses a maximum of disk space of around 250 GB. Total Storage in GB / 250 GB
Great, so now we have a general idea of which will be bottlenecks for our build depending on the components we pick. I will now go through my build that hits 4 TB/day in plots as an example. Keep in mind, you really need to read the reviews and make sure you are getting parts that can withstand heavy usage especially on the read/writes for our NVMes, check the TBW — terrabytes written.
Build:
- Intel Core i9–10850K 3.6 GHz 10-Core Processor
- Cooler Master Hyper 212 EVO 82.9 CFM Sleeve Bearing CPU Cooler
- MSI Z490-A PRO ATX LGA1200 Motherboard
- G.Skill Trident Z RGB 64 GB (2 x 32 GB) DDR4–2666 CL18 Memory
- 2x Western Digital WD BLACK SN750 NVMe M.2 2280 2TB (SAMSUNG 980 PRO M.2 2280 2TB good alternative)
- MSI GeForce GT 710 1 GB Video Card
- Metallic Gear Neo Silent ATX Mid Tower Case
- EVGA GD (2019) 500 W 80+ Gold Certified ATX Power Supply
- SSD Hard drive for OS (Any will work)
- SSD Hard drive for staging plots before moving to main farming HDDs (Any will work)
So, based on the previous metrics that we looked at above this particular build should be able to do.
- CPU: 10 Cores + 20 threads = 30/2 = 15 plots
- Memory: 64 GB/3.4GB = 18 plots
- Storage: 4000GB/250GB = 16 plots
This is hitting a bottle neck based on the selected CPU can be running on a max of 15 plots at a given time.
So all these numbers and estimates are great but how does it work in the real world? Very well. I am averaging around 7 hours per plot when running up 15 plots in parallel which works out to be 41 plots per day (4.1 TB in plots). The total build cost was around $1,800 depending on if you have some spare parts that can be repurposed. The $/plot is $35 — not bad but maybe we can do better, this metric a great way to calculate against different builds. If you were to download plots from AWS it would cost around $7 per plot so this build would pay for itself in little over a week.
I am still messing around with the settings for how long each plot should be delayed. Be on the look out for my next posts about setting up a node properly (lots of people are doing this wrong!), another budget build, and more.
Anyway, happy farming! 🚜
Update: Looks like the NVMes have gotten worse, performance has decayed down to just 4 TB per day and sometimes even below that if the cpu gets backed up.
