Next challenge is on llr321 (multithreaded CPU application),
starts Monday March 21, 03:21 UTC = Sunday March 20, 23:21 EDT, 20:21 PDT,
ends Saturday March 26, 03:21 UTC = Friday March 25, 23:21 EDT, 20:21 PDT.
According to
apps.php, the current application version is 9.01 of December 20, 2020. This corresponds to sllr2_1.1.0_linux64_201114 on 64 bit Linux (as opposed to sllr2_1.1.1_linux64_210204 which some of the other LLR based subprojects have adopted in the meantime).
Here is a randomly picked AVX-512 capable Xeon on Linux with a few llr321 results from single-threaded runs:
host ID 1105438/
llr 321 results
I am seeing 15 results which all had 1000K FFT length = 7.8 MiB FFT data size. The 5 among them which are from a "main task" took this computer 44…45 hours each to complete. (LLR will choose slightly different FFTs on different hardware, and one would configure a pure llr321 workload for multithreading, not singlethreaded.)
I'll check the range of FFT lengths of currently available llr321 workunits on Haswell in a bit. As you can see at Subproject status --> 321 Prime Search
Range statistics, this project looks at a single
k,¹ in contrast to some of the other LLR subprojects which search a few
k's in parallel. Maybe the llr321 workunits will be a bit more homogeneous than some others.
Edit,
the current search space is
n = 18,307,407…18,449,019 which results in 1 M FFT length = 8 MiB FFT data size on Haswell, and on Zen 2 too.
________
¹) The candidates have the form of 3*2^
n±1.
"These numbers have been studied by mathematicians since Thābit ibn Qurra in the 9th century",
said Reggie of PrimeGrid. And mankind is still not finished with them? :
-O