We have flags ergonomic tuning (number of GC threads and other) for T-series SPARCs. VM checks kstat data to recognize T-series but it does not check for M-series for which the same tuning should be done.
M-series has a lot more threads then T-series: ParallelGCThreads=ncpus*5/16 (1488*5/16=465). Currently M-series CPU is not recognized as T-series and not niagara_plus and as result the number of GC threads even bigger (933).

Here is code in vm_version_sparc.cpp:

unsigned int VM_Version::calc_parallel_worker_threads() {
  unsigned int result;
  if (is_niagara_plus()) {
    result = nof_parallel_worker_threads(5, 16, 8);
  } else {
    result = nof_parallel_worker_threads(5, 8, 8);
  }
  return result;
}

Notice that even more serious than configuring the wrong number of GC-Threads is the fact the without the proper detection of M-Series CPUS the "UseMemSetInBOT" option will not be properly unset on these CPUs. From vm_version.cpp: "When using CMS or G1, we cannot use memset() in BOT updates because the sun4v/CMT version in libc_psr uses BIS which exposes 'phantom zeros' to concurrent readers - see https://bugs.openjdk.java.net/browse/JDK-6948537.