Prime95 30.8 (beta, big P-1 changes, see post #551)

[Zhangrc]

Stage 2 switches between threads constantly during stage 2, unlike PRP or LL where no switching happens.
That might be why the "system" and "program manager" steal my CPU cycles.

[Prime95]

Originally Posted by Zhangrc

Stage 2 switches between threads constantly during stage 2, unlike PRP or LL where no switching happens.
That might be why the "system" and "program manager" steal my CPU cycles.

This bug report basically says "something odd is happening".

More useful would be the output from prime95 showing what the assigned affinity was for each worker/helper. Is the machine hyperthreaded? Prime95 is configured for how many workers, how many threads/worker, how many additional polymult threads, etc.?

[Zhangrc]

See some of my previous posts for more information.
This CPU is AMD R7 4800H, which has 8 cores and 16 threads. Prime95 is set to run 1 worker at core 2,3,4,5 (not hyperthreaded). However, it seems that Prime95 is switching between threads of each core.
For comparison, this is my CPU usage when running LLDC (M63759613, 1 worker using core 1,2,3,4, not hyperthreaded).

[jocelynl1204]

George, currently on build 8, when there's a "out of memory" it just simply quits. On your next build could it just output a message in the result file and then continue on to the next one in the worktodo list? This would really be appreciated. Thanks.

[Xyzzy]

[Work thread Jan 16 09:17] With trial factoring done to 2^72, optimal B2 is 6670*B1 = 20010000000.

6670? (!)

[petrw1]

Originally Posted by Xyzzy

[Work thread Jan 16 09:17] With trial factoring done to 2^72, optimal B2 is 6670*B1 = 20010000000.

6670? (!)

Not surprising with 30.8 and lots of RAM.
It will fly through Stage 2.
How much RAM?

[Xyzzy]

Originally Posted by petrw1

How much RAM?

[Work thread Jan 16 10:18] Using 229377MB of memory. D: 90090, 8640x34699 polynomial multiplication.

[petrw1]

@George: Do you still recommend these 2 settings?

ErrorCheck=1
SumInputsErrorCheck=1

[petrw1]

[Main thread Jan 17 12:32] Mersenne number primality test program version 30.8
[Main thread Jan 17 12:32] Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 8x1 MB, L3 cache size: 11 MB
[Main thread Jan 17 12:32] Starting worker.
[Work thread Jan 17 12:32] Worker starting
[Work thread Jan 17 12:32] 
[Work thread Jan 17 12:32] P-1 on M20847979 with B1=800000, B2=TBD
[Work thread Jan 17 12:32] Using AVX-512 FFT length 1152K, Pass1=192, Pass2=6K, clm=2, 8 threads
[Work thread Jan 17 12:35] M20847979 stage 1 complete. 736816 transforms. Total time: 176.568 sec.
[Work thread Jan 17 12:35] Round off: 0.0483828336
[Work thread Jan 17 12:35] Inversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 6.813 sec.
[Work thread Jan 17 12:35] With trial factoring done to 2^75, optimal B2 is 359*B1 = 287200000.
[Work thread Jan 17 12:35] If no prior P-1, chance of a new factor is 5.42%
[Work thread Jan 17 12:35] Switching to AVX-512 FFT length 1200K, Pass1=640, Pass2=1920, clm=2, 8 threads
[Work thread Jan 17 12:35] With trial factoring done to 2^75, optimal B2 is 345*B1 = 276000000.
[Work thread Jan 17 12:35] If no prior P-1, chance of a new factor is 5.39%
[Work thread Jan 17 12:35] Switching to AVX-512 FFT length 1280K, Pass1=128, Pass2=10K, clm=2, 8 threads
[Work thread Jan 17 12:35] With trial factoring done to 2^75, optimal B2 is 324*B1 = 259200000.
[Work thread Jan 17 12:35] If no prior P-1, chance of a new factor is 5.34%
[Work thread Jan 17 12:35] Using 24579MB of memory.  D: 4290, 480x1936 polynomial multiplication.

[Prime95]

Originally Posted by petrw1

@George: Do you still recommend these 2 settings?
ErrorCheck=1
SumInputsErrorCheck=1

ErrorCheck=1 won't hurt. I think the polymult library is good at keeping roundoff error well below 0.5. However, I've not thoroughly studied the matter.

SumInputs is only useful for SSE2 machines -- in other words its pretty much obsolete.

[petrw1]

Originally Posted by Prime95

ErrorCheck=1 won't hurt. I think the polymult library is good at keeping roundoff error well below 0.5. However, I've not thoroughly studied the matter.

SumInputs is only useful for SSE2 machines -- in other words its pretty much obsolete.

ok thanks.

BTW: I added this line to eliminate all the "Setting Affinity..." lines; a bigger deal for 8+core machines.
AffinityVerbosity=0