Questionable Hardware (Riesel LLR)

[storm5510]

I believe it is possible that I may have an incompatibility using LLR on my older i5-3570. For those not familiar with LLR, it searches for Riesel prime sequences in the form of k*2^n-1, where k is fixed.

I have ran several sequences up to n = 1,000,000. Perhaps as many as six. In all that, the I5 has only found one prime sequence. My i7-7700 has found over a dozen, minimum. I stopped running LLR on the i5 as I considered its output as questionable. I felt it may not be catching sequences which were prime.

In some ways, the i5 and i7 are not all that dissimilar. Both have SSE, SSE2, SSE4, and AVX. Only the i7 has AVX2 and FMA. The i5 runs Windows 7 Pro x64. The i7 runs Windows 10 Pro x64. I would not think the OS of each would make any difference.

[masser]

Have you tried doublechecking the residues produced by the i5 with the i7?

[kuratkull]

Take a bunch of reasonably sized primes from Top5k or from the forums and run them.

[storm5510]

Originally Posted by masser

Have you tried doublechecking the residues produced by the i5 with the i7?

It just so happens that I keep all the residues on the i5. I tried 15. All the residues match up on the i7. I found the single sequence of the prime.

12217*2^525957-1 is prime! (158333 decimal digits)  Time : 60.053 sec.

Above is the result from the i7.

This issue may be caused by something else. I sieve an entire series in one pass. A complete sieve can have anywhere from 15,000 to 35,000 terms to test. I run two instances of LLR on the i7, and a single on the i5. I wrote a small program to split these into either two, or three, different smaller files while maintaining the workload balance. It is not an even split between three files. The i5 gets roughly one-third of the entire sieve to run. The i7 gets the remainder split into two even size files. I thought the number of primes found would be proportional to the split. Prime numbers do not have predictable patterns though. All of this may simply come down to the luck-of-the-draw during the split. I checked the sums of all the split files to see if they total to the size of the original sieve file when I finished the program They did every time, so nothing is being skipped. It simply seems strange that the i7 would get nearly all of the prime sequences and the i5, only one.