Prime95 version 27.1 early preview, not-even-close-to-beta release

[Brain]

Rehighlightening this thread:
DC going well, ETA Dec 14:

[Main thread Dec 8 18:22] Mersenne number primality test program version 27.1
[Main thread Dec 8 18:22] Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 256 KB, L3 cache size: 6 MB
[Work thread Dec 8 18:23] Setting affinity to run worker on any logical CPU.
[Work thread Dec 8 18:23] Setting affinity to run helper thread 1 on any logical CPU.
[Work thread Dec 8 18:23] Setting affinity to run helper thread 2 on any logical CPU.
[Work thread Dec 8 18:23] Resuming primality test of M26451023 using AVX Core2 type-3 FFT length 1536K, Pass1=384, Pass2=4K, 3 threads
[Work thread Dec 8 21:23] Iteration: 7020000 / 26451023 [26.53%].  Per iteration time: 0.019 sec.

[monst]

M27515261 was successfully verified with v27.1, build 1:

http://www.mersenne.org/report_expon...&B1=Get+status

I have 3 more in the queue.

[MrHappy]

DC of M27787667 was successful with 27.1

[LaurV]

Four expos in range 25.1M to 27.8M were:
-partially DC-ed with P95 v26
-stopped when the new version appeared, at "scores" between 30% done and 68% done
-continued with the new version P95 v27
All residues matched.
Another 4 expos are under the same test (they were all 8 stopped somewhere in the middle when the new version appeared, and I continued on them in two "lots" of 4 expos each, with the new vesion).

[Ralf Recker]

[Dec 11 11:58] Iteration: 850000 / 26154413 [3.24%].  Per iteration time: 0.005 sec.

[LaurV]

How in the seven seas did you get that 5 milisexes ??!?!?
I can't go so low even if I run 8 cores for one assignment, and I dare to say that I have a good CPU, running at 3.4 gigs...

[Dec 11 18:31:30] Resuming primality test of M25154581 using AVX Core2 type-3 FFT length 1536K, Pass1=384, Pass2=4K, 8 threads
[Dec 11 18:31:30] Iteration: 10867978 / 25154581 [43.20%].
[Dec 11 18:31:40] Iteration: 10870000 / 25154581 [43.21%].  Per iteration time: 5.177 ms.
[Dec 11 18:32:32] Iteration: 10880000 / 25154581 [43.25%].  Per iteration time: 5.215 ms.

[Ralf Recker]

Originally Posted by LaurV

How in the seven seas did you get that 5 milisexes ??!?!?

I started one DC with 4 threads after some extra torture tests to make sure that the CPU is able to handle the additional(?) load of the AVX FFTs. The extra gain from running more than 3 threads is neglectable (a.k.a. zero ) so I switched to running 2 DCs with 2 threads each to improve the throughput a little bit...

[Dec 11 13:13] Iteration: 1570000 / 26154413 [6.00%].  Per iteration time: 0.007 sec.

and

[Dec 11 13:15] Iteration: 190000 / 28706551 [0.66%].  Per iteration time: 0.007 sec.

[LaurV]

I would really like to see the microseconds on that one, could you add the TimeStamp=2 and the TimingOutput=4 switches to prime.txt and repost? That is quite interesting, with 2 workers I get 10 ms (!?) and can't go lower. What frequency is your CPU?

[Ralf Recker]

Originally Posted by LaurV

I would really like to see the microseconds on that one, could you add the TimeStamp=2 and the TimingOutput=4 switches to prime.txt and repost? That is quite interesting, with 2 workers I get 10 ms (!?) and can't go lower. What frequency is your CPU?

4.4 GHz - The multiplier is set to 45x for the first turbo mode step but it's pretty obvious why it doesn't kick in. Here are the timings:

[Dec 11 15:26:41] Iteration: 2510000 / 26154413 [9.59%].  Per iteration time: 7.413 ms.
[Dec 11 15:27:55] Iteration: 2520000 / 26154413 [9.63%].  Per iteration time: 7.352 ms.
[Dec 11 15:29:08] Iteration: 2530000 / 26154413 [9.67%].  Per iteration time: 7.318 ms.
[Dec 11 15:30:21] Iteration: 2540000 / 26154413 [9.71%].  Per iteration time: 7.327 ms.
[Dec 11 15:31:35] Iteration: 2550000 / 26154413 [9.74%].  Per iteration time: 7.337 ms.
[Dec 11 15:32:48] Iteration: 2560000 / 26154413 [9.78%].  Per iteration time: 7.360 ms.

[Dec 11 15:25:51] Iteration: 1110000 / 28706551 [3.86%].  Per iteration time: 7.566 ms.
[Dec 11 15:27:06] Iteration: 1120000 / 28706551 [3.90%].  Per iteration time: 7.482 ms.
[Dec 11 15:28:20] Iteration: 1130000 / 28706551 [3.93%].  Per iteration time: 7.358 ms.
[Dec 11 15:29:33] Iteration: 1140000 / 28706551 [3.97%].  Per iteration time: 7.297 ms.
[Dec 11 15:30:46] Iteration: 1150000 / 28706551 [4.00%].  Per iteration time: 7.326 ms.
[Dec 11 15:31:59] Iteration: 1160000 / 28706551 [4.04%].  Per iteration time: 7.352 ms.

At 3.4 GHz that would be ca. 9.52-9.53 ms per iteration. Of course the timings are a bit distorted by a bunch of background tasks running on the box (including a Linux VM).

[monst]

2 for 2:

The residues match for M27816601.

[TheJudger]

Originally Posted by Prime95

...
If there is demand, I could also create a 32-bit mprime (would that run on a 64-bit Linux install)?

If it doesn't take too much of your time why not giving a try? With a x86-64 kernel and a x86/x86-64 userland it should work.

Oliver