DC performance issue
 

I noticed something strange running Prime95 on my MacBook Pro.

What's "wrong" with this output from Prime95?

[CODE]
[Jul 5 06:47] Iteration: 41220000 / 41307899 [99.78%], ms/iter: 17.007, ETA: 00:24:54
[Jul 5 06:52] Iteration: 41230000 / 41307899 [99.81%], ms/iter: 17.073, ETA: 00:22:09
[Jul 5 06:56] Iteration: 41240000 / 41307899 [99.83%], ms/iter: 17.184, ETA: 00:19:26
[Jul 5 07:01] Iteration: 41250000 / 41307899 [99.85%], ms/iter: 17.346, ETA: 00:16:44
[Jul 5 07:06] Iteration: 41260000 / 41307899 [99.88%], ms/iter: 17.378, ETA: 00:13:52
[Jul 5 07:11] Iteration: 41270000 / 41307899 [99.90%], ms/iter: 17.043, ETA: 00:10:45
[Jul 5 07:16] Iteration: 41280000 / 41307899 [99.93%], ms/iter: 17.214, ETA: 00:08:00
[Jul 5 07:21] Iteration: 41290000 / 41307899 [99.95%], ms/iter: 17.082, ETA: 00:05:05
[Jul 5 07:26] Iteration: 41300000 / 41307899 [99.98%], ms/iter: 17.397, ETA: 00:02:17
[Jul 5 07:29] M41307899 is not prime. Res64: DE205ECAF0B96EB3. We10: 498D7D0A,12943239,00000000
[Jul 5 07:29] Setting affinity to run helper thread 1 on any logical CPU.
[Jul 5 07:29] Setting affinity to run helper thread 2 on any logical CPU.
[Jul 5 07:29] Setting affinity to run helper thread 3 on any logical CPU.
[Jul 5 07:29] Starting primality test of M77552149 using AVX FFT length 4M, Pass1=256, Pass2=16K, 4 threads
[Jul 5 07:34] Iteration: 10000 / 77552149 [0.01%], ms/iter: 16.830, ETA: 15d 02:30
[Jul 5 07:39] Iteration: 20000 / 77552149 [0.02%], ms/iter: 17.149, ETA: 15d 09:20
[Jul 5 07:44] Iteration: 30000 / 77552149 [0.03%], ms/iter: 16.648, ETA: 14d 22:30
[Jul 5 07:49] Iteration: 40000 / 77552149 [0.05%], ms/iter: 17.129, ETA: 15d 08:47
[Jul 5 07:53] Iteration: 50000 / 77552149 [0.06%], ms/iter: 16.988, ETA: 15d 05:43
[Jul 5 07:58] Iteration: 60000 / 77552149 [0.07%], ms/iter: 16.830, ETA: 15d 02:16
[Jul 5 08:03] Iteration: 70000 / 77552149 [0.09%], ms/iter: 17.024, ETA: 15d 06:24
[Jul 5 08:08] Iteration: 80000 / 77552149 [0.10%], ms/iter: 17.059, ETA: 15d 07:07
[Jul 5 08:12] Iteration: 90000 / 77552149 [0.11%], ms/iter: 16.837, ETA: 15d 02:17
[/CODE]

You (or it) wrongly used FFT of 4M for the DC, when a much shorter one should have been used. What do I get for the correct answer? :razz:

[QUOTE=LaurV;463041]You (or it) wrongly used FFT of 4M for the DC, when a much shorter one should have been used. What do I get for the correct answer? :razz:[/QUOTE]

Prime95 claimed to be using the 2304K AVX FFT for the 41M exponent. (The CPU is a Sandy Bridge.)

I've done some experimenting, and it seems that this performance issue happens for any exponent where the FFT size is either 2304K or 2880K. I guess next year I will just make sure I get an exponent with a different FFT size from those two when it comes time for my pseudo-mandatory DC assignment.