First before we get into this, I want to say how good FlexFarmer is as a piece of software. There are some security issues with it, as with all Chia software, but it is young and I have faith that FlexPool will continue to improve it. I understand, and mostly agree with, Chia Network’s arguments about it. But it works, it works well and it does significantly reduce the resource requirements for farming. This is an issue that Chia will have to wrestle with, and likely create a halfway solution for, since the requirement differences are very real. The Chia client is keeping two separate databases up to date (with all the issues related to that) and handling a lot of internet traffic in order to maintain both Full Node and Wallet sync. This is a cost borne by the farmers in order to keep the network secure and decentralized, but it is a cost.
FlexFarmer basically eliminates this overhead, with a very lightweight package that can run on a properly wired potato. It connects to a FlexPool-managed Full Node over the internet and the FlexFarmer client signs the blocks but the FlexPool central node actually farms them. You still get your 0.25xch reward as the block farmer and FlexPool gets the 1.75xch pool reward portion, exactly as normal.
The main issue with FlexFarmer is the claims they have made about it. On numerous occasions they claimed that farmers on their network using FlexFarmer see higher rewards and that their client will check signage points missed by the Chia client. They even claimed that users on their pool (not just FlexFarmer users) would farm blocks more often because of their efficient infrastructure. That last one is clearly BS but the others are worth a test, so that’s just what I did.
I do not have a huge farm, and the farm I do have is all on a single NFT plot so it is not the ideal situation for A/B testing. So instead I spent a few weeks going back and forth and back and forth for similar amounts of time, sometimes a week sometimes a day, and analyzing the results. And the honest truth is that there isn’t really a difference, at least not one that becomes statistically significant at 454 plots.
Now, my setup is pretty well optimized and I do not have an issue with stale shares using the standard Chia GUI. I also have an Intel Rocket Lake CPU with AVX512 and very good single threaded performance, so that might help with not having problems. I think users who have underpowered systems who struggle to keep up with the Full Node and Wallet sync, like Raspberry Pi users, will probably see significant improvements here as their CPU is doing less. But as you can see in the following 24 hour snapshots, they are within error of margin to each other.
Over the 6 weeks or so I have been doing this experiment the results have swung a little bit in either direction, showing that there will be differences no matter what but that they are almost certainly caused by external factors like pool luck. It is possible with a bigger farm, or a better test that the differences would come out of the statistical noise. Or if I was having a lot of trouble with stale shares with the Chia client then FlexFarmer might fix that. But barring those two things, for my setup at least, the two are neck and neck even.