Thread by @bramcohen

People are asking/speculating about the new Chia plotter. It's better but the details are complicated (thread)

What it does is make better use of available cores for multithreading. This results in a big headline speedup on SSD in terms of the minimum number of seconds to finish a whole plot

But it isn't nearly as big an improvement to overall rate of plotting if you compare to running multiple plots on multiple drives at once. It also probably makes almost no difference writing to HD because that was nearly I/O bound already

Overall throughput changes people are reporting on their own hardware setups are between 50% better on a probably not already well optimized system to 20% worse on a super well optimized system done by a professional

TBW is completely unchanged. The previous 3x speedups we implemented were almost entirely in reducing writes so that's carried over

About as apples-to-apples as you can get is that a setup with the new plotter with one computer connected to a bunch of SSDs will get about the same performance as the old plotter on multiple computers connected to the same SSDs, so there's hardware cost savings

The minimum setup with the new plotter will be able to do more and it will lose less data if interrupted in the middle of a plot. It's also a lot more brainless to configure properly so comparing perfectly optimized systems is probably unfair

The new plotter doesn't yet produce byte-for-byte identical outputs. It's dropping a fairly trivial amount of data which makes the final output file smaller but reduces the proof of space hit rate by a much larger but probably still small amount. We haven't measured it yet

It's probably possible to fix it to drop nothing without a meaningful reduction in speed but we'll see how that goes.

We knew before this that better pipelining was possible but had decided to stop working on plotting in favor of launching our network, getting transactions started, and building pooling protocol.

Right now the whole team is heads down on pooling protocol, it will be out soon.

There are one or two TBW-reducing optimizations which can be done and also a change to format which reduces plot size by about 2%. We'll get to these less soon than pool protocol.

A heads up to anyone working on optimized plotters: We're making some changes to the format of plots which give a pool puzzle hash instead of a pool key so please make sure to update that before we launch

If you aren't working on that don't worry, unless you really went out of your way to do otherwise all your plots are using pool keys and can remain unchanged