I believe it is a result of changes described here (the post is about thinLTO but it explains parallel codegen after the TOML config example): https://internals

I believe it is a result of changes described here (the post is about thinLTO but it explains parallel codegen after the TOML config example): https://internals.rust-lang.org/t/help-test-out-thinlto/6017 https://internals.rust-lang.org/t/help-test-out-thinlto/6017 In order to parallelize the build process at the rustc-llvm codegen boundary, Cargo has to implement some orchestration between compiler processes (using jobserver-rs [1]) so that 10 rustc processes didn't all create 10 threads and codegen units each and blow up the system. I'm guessing that memory quotas are a part of that implementation. [1] https://crates.io/crates/jobserver/reverse_dependencies https://crates.io/crates/jobserver/reverse_dependencies