LLR 3.8.20 ( base3- base 9)
 

I found this little anomaly so if someone can solve mistery :)

D:\PGEN>llr64 -d -q"2976*9^1000000+1"
Base prime factor(s) taken : 3
Starting N-1 prime test of [COLOR=red]2976*9^1000000+1[/COLOR]
Using all-complex AVX FFT length 256K, Pass1=256, Pass2=1K, [COLOR=Red]4 threads[/COLOR], a = 3
2976*9^1000000+1, bit: 140000 / 3169934 [4.41%]. Time per bit: 0.364 ms.
Caught signal. Terminating.

D:\PGEN>llr64 -d -q"2976*3^2000000+1"
Base prime factor(s) taken : 3
Starting N-1 prime test of [COLOR=Magenta]2976*3^2000000+1[/COLOR]
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, [COLOR=magenta]4 threads[/COLOR], a = 3
2976*3^2000000+1, bit: 141000 / 3169935 [4.44%]. Time per bit: 0.444 ms.
Caught signal. Terminating.

So basicly, those two numbers are same, but test with base 9 is faster then test in base 3. I was expection opossite results but if someone has some to say... :)
Computer was on idle and only LLR was running while I do this test

That's too small a test for 4 threads to be efficient;if you repeat the test single-threaded, perhaps you can distinguish between one FFT using 4 threads more effectively and base-9 actually being faster.

I mean, 256k FFT may use 4 threads better than 240k. Anything smaller than 100k (128k?) per thread is mostly wasted.

1 core

D:\PGEN>llr64 -d -q"2976*9^1000000+1"
Base prime factor(s) taken : 3
Starting N-1 prime test of 2976*9^1000000+1
Using all-complex AVX FFT length 256K, Pass1=256, Pass2=1K, a = 3
2976*9^1000000+1, bit: 23000 / 3169934 [0.72%]. Time per bit: 1.235 ms.
Caught signal. Terminating.

D:\PGEN>llr64 -d -q"2976*3^2000000+1"
Base prime factor(s) taken : 3
Starting N-1 prime test of 2976*3^2000000+1
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 3
2976*3^2000000+1, bit: 20000 / 3169935 [0.63%]. Time per bit: 1.247 ms.
Caught signal. Terminating.

It looks it was main reason!
Thanks