I’m using xgboost in R/Rstudio desktop on windows 10, 4 cores 8 logical processor. Since I updated xgboost to 0.71.2, it suddenly became very slow. After investigation, it appears it wasn’t using multithreading/CPU properly: all cores were used, but it wouldn’t go beyond 50-60% CPU usage, and the nthread input in params for xgb.train (for example) had close to no impact on the performance. I rolled back to 0.71.1, it’s working fine again (for nthread=8, all cores being used, 100% CPU usage.
Did anybody noticed this issue as well? Any solution/fix for latest version of xgboost?