https://github.com/dkull/rpt/releases/tag/v0.0.3
RPT can now choose the best threadcount for you. Just give 0 as the threadcount.

This information should be transferrable to LLR64, eg. you can test the best threadcount and use that in LLR64 with the same k,n

Currently the best one has to be at least 5% better than the last best, this ensures we don't waste cores for negligible improvements. Also this minimizes the effects of jitter.

Code:

./rpt_release_linux64_0_0_3 39547695 3664175 0
=== RPT - Riesel Prime Tester v0.0.3 [GWNUM: 29.8 GMP: 6.2.0] ===
LLR testing: 39547695*2^3664175-1 [1103035 digits] on 0 threads
step #1 find U0 ...
found V1 [11] using Jacobi Symbols in 1ms
found U0 using Lucas Sequence in 390ms
step #1.5 benchmark threadcount ...
threads 1 took 1319ms for 1000 iterations
threads 2 took 770ms for 1000 iterations
threads 3 took 552ms for 1000 iterations
threads 4 took 489ms for 1000 iterations
threads 5 took 514ms for 1000 iterations
threads 6 took 543ms for 1000 iterations
threads 7 took 513ms for 1000 iterations
threads 8 took 466ms for 1000 iterations
using fastest threadcount 4
FFT size 384KB
step #2 LLR test ...
./rpt_release_linux64_0_0_3 39547695 506636 0
=== RPT - Riesel Prime Tester v0.0.3 [GWNUM: 29.8 GMP: 6.2.0] ===
LLR testing: 39547695*2^506636-1 [152521 digits] on 0 threads
step #1 find U0 ...
found V1 [11] using Jacobi Symbols in 0ms
found U0 using Lucas Sequence in 65ms
step #1.5 benchmark threadcount ...
threads 1 took 159ms for 1000 iterations
threads 2 took 155ms for 1000 iterations
threads 3 took 149ms for 1000 iterations
threads 4 took 164ms for 1000 iterations
threads 5 took 162ms for 1000 iterations
threads 6 took 169ms for 1000 iterations
threads 7 took 174ms for 1000 iterations
threads 8 took 192ms for 1000 iterations
using fastest threadcount 3
FFT size 50KB
step #2 LLR test ...