Chia Plot 2.0 Hardware Overview

Chia Farming has gotten a wild new facelift with the introduction of Chia plot compression and the ramifications surrounding this. The end result is a seemingly much more complex set of variables and options for the average Chia farmer to wade through. The benefits of doing so cannot be ignored lightly with the benefits and tradeoffs outweighing inaction. Whether you are using the official version or the madMAx Gigahorse Chia farming version, the need for speed and technical execution is high.

Chia now uses GPUs to achieve an optimal level of tables and bits being able to be dropped (vs. stored) and this changes everything. We currently have one end-to-end solution that is capable of generation and farming of compressed plots and that is the madMAx Gigahorse. This document will be updated over time to reflect the software changes that impact hardware selection, benchmarking and utilization. This current document is a living piece and is neither close to finished nor authoritative. It is, however, what I have had time to check into so far.

Chia plotting 2.0 ROT (Rules of Thumb)

You want to have at least a PCIe 3 16x wide bus to connect your GPU(s) to
You want to have 256GB RAM in your system
- You cannot in most AMD and INTEL desktop system get to 256GB RAM
The higher the speed of your ram, the better. As long as its stable and not stupid expensive
Workstations are a great way to get lots of DIMMS into a system, and the ram they run is cheaper ECC
You will be doing great with a 3060 or 3060ti in a PCIe-3 generation 16 wide slot with 256GB DDR3 ecc.
- around ~210 seconds per plot based off
You will need some combination of these storage and networking things to handle these plots that are flying out super fast.
- Super fast networking
- Directly connect storage
- Software raid for final storage at a stripe size that is rather fast
- A pretty fast buffer of storage if you use one of the plot orchestration/mover scripts
Learning to use Linux is a great idea. Now more then ever
PCIe-4 system with enough RAM and multiple lanes are pretty *EPYC*