The only workload I've done lately that buried all the cores was a significantly parallel CSV import into clickhouse from compressed zips of multiple CSVs, whil

The only workload I've done lately that buried all the cores was a significantly parallel CSV import into clickhouse from compressed zips of multiple CSVs, while filtering the CSVs with python. Running nproc=cores was 5% slower than nproc=cores+smt in overall throughput.