This version is not performance comparable to v0.8.5 due to new optimizations.
General New Features:
- General performance changes due to a completely new training model.
- Tuning Profiles: Each binary now contains various preset tuning profiles optimized for different CPUs.
- The stress-tester will now display the elapsed time when an error occurs.
- Benchmark Pi now has a 2nd multi-threaded option that disables SMT. (uses no more than 1 vcore per physical core)
- Custom Compute now has an option to load defaults that disables SMT.
- The refresh rate of status prints has been increased.
- New and faster secondary formulas for Log(2) and Log(3) thanks to Jorge Zuniga.
- Over a dozen new custom formulas have been added.
Parallel Framework Improvements:
- The Task Queue and Push Pool frameworks now support core affinities.
- The Task Queue and Push Pool frameworks now have better load balance for imbalanced processor groups. (untested)
- The TBB parallel framework now supports limiting the # of threads.
Integration/Wrapper Improvements:
- New command line options have been added to better control console printing. You can choose beween WinAPI and Virtual Terminal escape sequences for how y-cruncher should do console manipulation such as colors and refreshing status lines.
- Added a command line option to log all console output to a file. This is a better alternative to piping that will strip out all escape sequences.
Other:
- Removed FFTv2, FFTv3, SFTv2, and SFTv3 from the stress tester.