One very popular account that you probably have already heard is here: https://jeena.net/lemmy-switch-to-piefed.

Another informative discussion relates to the upcoming (in 2026) switch of slrpnk.net from Lemmy to PieFed, see e.g. https://slrpnk.net/comment/18799445. Some highlighted nuggets from that:

the main bottleneck on performance is the database itself, and at that point the language the frontend is written in doesn’t seem to make much of a practical difference

the core issue with Lemmy is really that it is very annoying to run and maintain, has huge memory issues (ironically given that it is written in Rust) that the devs ignore since years, and the image integration is a stuff of nightmares. In addition, the upstream devs are often actively hostile to sensible suggestions how to improve things and the proposed solutions by them often make things actively worse (latest case in point: the next version will hardcode lemmy.ml as a source to pre-fetch popular communities). After nearly 5 years of running Lemmy, I am ready to cut my losses and rather give Piefed a try, and so far the devs and community around it has been very welcoming and actually have lots of sensible ideas.

(Note that the proposed hard-coding issue has been somewhat walked back, as in it will still be hardcoded to some instance but it only remains lemmy.ml by default yet can be changed. Using a single instance as the ultimate source of truth though, it will still be subject to issues of defederation.)

The Lemmy backend causes the Postgres database to use more and more RAM, to the point that it crashes with out of memory issues randomly and causes other processes to go down with it. I have reported the issue multiple times and I am not the only one with the problem since many years,

There are also some further links there to older discussions and additional blog posts, such as https://join.piefed.social/2024/02/13/technical-performance-of-each-fediverse-platform/.

See also notes for developers at https://join.piefed.social/docs/developers/, e.g. it mentions PieFed relying upon the Flask framework, and the code repository at https://codeberg.org/rimu/pyfedi, which reportedly the Docker containerization makes it fairly straightforward to install? (I have no personal experience with that though, or Docker containers at all myself.)

In my mind, PieFed is running circles around Lemmy and has been for like a year now. Nobody knows how scalable any of these approaches would be to handle like a million of people, but on the other hand the entire Threadiverse has only ~35k active users currently (last I checked) and that is already down from our peak at 55k just after the Rexodus. i.e., scalability is the least of our concerns right now, and can be postponed for another day, in lieu of aspects such as features proferred to users and ease of use to instance admins.

Then again, FOSS is FOSS, so I wish both Lemmy and PieFed (and Mbin, nodeBB, etc.) the absolute best of success - when one is improved, we all benefit due to the federated nature of content shared via ActivityPub Protocol. I just think that Lemmy has little hope for the future, while PieFed continually impresses me. Nothing is perfect, but on the whole I hear the best things about it, and I have little doubt you'll enjoy having delved deeper into learning about it, based on so many stories shared in e.g. !piefed_meta@piefed.social that have said exactly that.