Benchmarking Wild vs. Mold
3 hours ago
- Mold's recent benchmark updates include Wild for the first time, showing Wild slower than previously published by Wild.
- Key benchmark configuration differences: Mold deletes output files and uses ext4; Wild historically uses tmpfs and does not delete output files.
- Mold passes --no-fork to override default forking behavior; Wild leaves forking enabled when measuring time.
- When replicating Mold's configuration (ext4, delete, no-fork) on a 16-core Apple M1, results for most benchmarks were close to Mold's published results.
- Further testing on clang-release revealed that Wild performs best with tmpfs, no file deletion, and forking—opposite to Mold's config.
- Wild lacks OS-specific tweaks like fallocate and hugepages that Mold uses; these are already implemented in Wild for the next release.
- Recent Mold versions (2.42.0, 2.42.1) have become considerably faster, partly explaining the discrepancy between Wild's August 4th and Mold's August 28th benchmarks.
- The author could not reproduce the Threadripper results due to hardware limitations; speculates thread count differences may contribute.