So if I got all of that right, a RTX 3060 (170W) could be sufficient for a 4PiB farm at level 7 compression. That shaves off about 25% of the plot size.
Current netspace is 21EiB = 21,504PiB
21,504PiB/4PiB = 5,376 of those cards needed (unrealistic, ideal minimum for this example).
That’d be a total power draw of about 915kW. 24/7 for a year results in around 8,000,000kWh/a or 8GWh/a.
That is the equivalent of the energy requirement of about 4,000 households.
By compressing plots that way this GPU energy consumption results in a “virtual” gain of 25% storage. It’s equivalent to adding 5.25EiB of netspace. Yet, without increasing the node count.
Assuming the 0.19TWh annual energy consumption of the whole network shown on chiapower.com is correct, adding 25% more netspace to the network increase the consumption by 47.5GWh per year or 0.0475TWh/a.
For some perspective:
- A standard on-shore wind turbine with a common 2-3MW generator can harvest around 6GWh of wind energy per year.
- A modern nuclear power plant often runs two generator of around 1.6GW each, so about 3.2GW with 93% of operation time/efficiency. That can yield around 0.26TWh/a.
- The BTC network alone is drawing around 115TWh/a (equivalent to the yearly yield of almost 20k modern on shore wind turbines or 450 nuclear power plants)
Sure, there are 100,000 Chia nodes and not just about ~5,300. Many more will “waste” energy by over-provisioning GPU power. Yet, not every node will operate said RTX 3060.
Yet, it is far away from “entirely killing the spirit of Chia” as some might have implied.
On the upside, old GPUs could be reused in smaller farms just like old server hardware (and drives) are used to farm with upcycled hardware in the first place.
Also, future GPUs (and CPUs) will become more efficient, whereas it is unreasonable to assume that the motors of spinning hard drives have a lot of untapped energy savings left. SSDs would bring significant power savings, but with just about 2.5EiB of SSD storage being produced globally per year it is unreasonable to assume the Chia network will soon switch to storing plots on SSDs.
Very important to note:
The energy required to replot is not taken into account in those back of the envelope calculations, of course.
The big question remains: Why all that?
For the network itself there is no benefit in just having more plots. It won’t affect the number of transactions and neither increase the security of the network. This can only be achieved by having more nodes, and not by this “virtual” increase of netspace.
The only benefit is on the level of an individual farmer to have a slight edge over others at the cost of an increase operating operating and the additional need to replot (cost for energy + hardware-wear).
All that brings me to:
What are the real risks when deciding to switch to those compressed plots?
What if the network is switched over to k34 or even k35 plots? What would the GPU-/hardware requirements be to operate your proposed compressed plots? Is it even achievable with today’s hardware?
And what keeps Chia from transitioning to higher k-values earlier?
The only thing that comes to mind is that the transition cost will be hard for every farmer as replotting is required, which could deter some farmers. But what if the transition to higher k-value plots would be announced ahead of time. As in “you have 6 months to transition to k34/k35”.
The incentivize replotting a new official (low-risk) plot format could help if it offers some (maybe not 25%) space savings, but enough so that it exceeds what the higher k-value plots need as excess in storage.