[crazypython]

Is it possible to optimize plotting to use I/O less and RAM more? Or is it an inherent boundary of the protocol like energy use is in PoW?

[r1ch]

The reference implementation is Python, so there's lots of room to optimize. I wonder if you can eventually build an ASIC that replaces the disk based lookup with in memory calculations of the same data.

[crazypython]

Can you send a link? I'd like to help.

[nessex]

Most of the work happens in C++[1]

[1] https://github.com/Chia-Network/chiapos

[wmf]

If you have ~300 GB of RAM you should be able to plot directly in RAM.

[crazypython]

Imagine a cloud provider has 300GB RAM. What if they plotted and let other people later download the plot?

[wmf]

I think bandwidth charges kill that idea.

[crazypython]

On AWS, yes.

On Scaleway and Hetzner, there is unmetered 100Mbps bandwidth.

On Linode, Vultr, and DigitalOcean, there is a 1-month 1TB limit.