I started ESP and found it took very long on both hosts, more than 300 000 s. What's more, the i7-4790K took 340 000 s, the i3-2120 took only 305 000 s approximately. SGS and 321 sieve, the i3 was usually 50% slower.

Unfortunately I had MT disabled and only now saw that it is reommended. What settings should I use?

I went with 50% of the processors and 2 threads, is that correct? Unfortunately these tasks take so long, you cannot really experiment.

MT all the cores you have available for crunching;
Or use Prime95 to test the FFT size ESP has.
See here for instructions.
http://www.primegrid.com/forum_thread.php?id=8240
____________
SHSIDElectronicsGroup@outlook.com

waiting for a TdP prime...
Proth "SoB": 44243*2^440969+1

Thanks for the link.

So far it seems restricting to physical cores and MT=2 has done a lot of good. Two ESP tasks are running on the i7 with one giving an estimate of 70 000 s, the other something like 130 000 s.

According to taskmanager one process utilizes 25% of cpu ressources, the other 12-13%. I changed MT to 1 to prepare for SGS tasks and it might be one ESP task is running with this setting?

I did prime95 test and 1 worker was optimal in terms of throughput. The posts said a worker was the same as a task, so it wouldn't tell about the optimal number of threads?

Anyway, after pausing/unpausing Boinc, now both ESP tasks are running at 25% CPU each. So apparently it wasn't related to the MT setting.

What are other people using on a 4 physical cores i7? How many threads?

For "big" units like this, it is generally good to use 1 task using as many threads as you have real cores with Intel CPUs. It can get more complicated on AMD side.