"btw, all computers are set to work up to 85% of Tj!"

Really 'all'? (Mine does not until hitting 95C, where Tj is set 100C per CoreTemp, and there's absolutely nothing that allows changing these values.)

TThrottle overacts on HT CPU's, simply because it's not in sync with what the OS-CPU combo does in allocating unused cycles. Using TM, reducing all BOINC processes to 80% I'm getting a 10C drop, but if using TT that is not achieved until it's cut BOINC by 75% [down to 25%!]. TM works at the millisecond level... real smooth computing. TThrottle has a much coarses control from observing Task Manager. BOINC itself does not sample more frequent than every second **, which is why the CPU time control works in this -raw- fashion... 50% is running 1 second, pausing 1 second.

** There was a brief test and a few client builds, on request of WCG, to set it to 0.1 second intervals [meaning 60% would run 0.6 seconds, pause 0.4 seconds], but then the Dr. remembered the science wrappers sample every second as well, so that feature was reversed (Projects would have to recompile with a new wrapper, apparently a transition bridge too far). Tested it and it worked actually quite well for WCG sciences, so a shame it was not pursued further. As the driver left, knreed, doubt we'll ever going to see smooth BOINC integrated throttling.

"btw, all computers are set to work up to 85% of Tj!"

Not getting you wrong... the 'my' is missing in your comment. ;>)