Prime95 30.8 (big P-1 changes, see post #551) - Page 10

tha · 2021-12-05, 09:18

Quote:

Originally Posted by Prime95

[*]Save files during P-1 stage 2 cannot be created

I still have thousands of save files done with previous versions. Can I rerun them with the new B2 bound and the old save files? Or does that require a conversion routine in the software? Is it possible at all?

axn · 2021-12-05, 12:47

Quote:

Originally Posted by tha

I still have thousands of save files done with previous versions. Can I rerun them with the new B2 bound and the old save files? Or does that require a conversion routine in the software? Is it possible at all?

Yes. If you keep the same B1 and a larger B2, it will run the prevB2-newB2 range.

petrw1 · 2021-12-05, 15:26

Quote:

Originally Posted by Prime95

This version adds SSE2, FMA, AVX-512 support.

On my i7-7820x:
26.8M exponent
B1=600K/B2=120M

Build 2 (without AVX)
Stage 1: 14 Minutes
Stage 2: 10 Minutes

Build 3 (with AVX)
Stage 1: 9.5 minutes
Stage 2: 6 minutes

axn · 2021-12-05, 16:43

Presumably you're using AVX to mean AVX-512. I wonder how much of the stage 2 improvement is due to AVX-512 specifically, and how much due to the improvements to stage 2 itself. For instance, for my particular case, stage 2 run time (excluding init) went down from 330s to 212s by switching from build 1 to build 3 (build 2 in linux was not timed due to multithreading issues).

petrw1 · 2021-12-05, 19:23

Quote:

Originally Posted by axn

Presumably you're using AVX to mean AVX-512. I wonder how much of the stage 2 improvement is due to AVX-512 specifically, and how much due to the improvements to stage 2 itself. For instance, for my particular case, stage 2 run time (excluding init) went down from 330s to 212s by switching from build 1 to build 3 (build 2 in linux was not timed due to multithreading issues).

Yes AVX is the lazy typers AVX-512.
Considering my Stage 1 time dropped from 14 to 10 I assume a good part of it was AVX ... umm I mean AVX-512.

tha · 2021-12-05, 19:26

My current system is 4 years old. It is based on a Z-170 Asus motherboard. At the time 16 Gb of RAM (4 x 4 Gb) was the sweet spot. So I re-evaluated it and today 64 Gb (4 x 16 Gb) costs the same as 16 Gb then.

So, I've ordered it and it will be delivered later this week.

petrw1 · 2021-12-05, 20:29

And it seems the TBD value of B2 ... at least for this 1 PC ... is brilliant now.
The value chosen is 256xB1; which for this PC gave very similar Stage 1 and Stage 2 times.

chalsall · 2021-12-06, 00:54

George. Seriously. OMGs!!!

I had some time this weekend to experiment with 30.8b3. VERY impressive!

By chance, I was also "pinged" on IBM Cloud offering a special promotion (in addition to the usual $200 USD credit for first-time-users). You have to give your credit card as part of the credentials, but they don't actually charge you (except for a $1 "hold" to prove the CC is valid).

Then, when you're spinning up your first VPC (read: Instance) you're given the opportunity to enter a Promo Code which gives you an immediate $500 USD credit. You're apparently then also able to contact Sales to get another $1,500 credit (I haven't yet done this, as Sales only work weekdays).

This has allowed me to very quickly experiment with just how valuable lots of RAM is. The naming convention with IBM Cloud is very similar to other offerings from the various players.

cx2-8x16 is "compute-optimized" with four (4#) real cores, and 16 GB of RAM. bx2-8x32 is "balanced" with 32 GB of RAM. I haven't yet experimented with memory-optimized. I have found that there is some variation in the CPU provisioned; as usual, if you don't like the one provisioned at launch, restart the VPC.

model name : Intel Xeon Processor (Cascadelake)
stepping : 6
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology eagerfpu pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 arat pku ospke avx512_vnni md_clear spec_ctrl intel_stibp arch_capabilities

model name : Intel Xeon Processor (Skylake, IBRS)
stepping : 4
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology eagerfpu pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 arat pku ospke md_clear spec_ctrl intel_stibp

Some quick empirical showing the difference more RAM results in. Both types using a Cascadelake CPU:

Code:

[Work thread Dec 5 23:34] P-1 on M15899909 with B1=960000, B2=TBD
[Work thread Dec 5 23:34] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Dec 5 23:34] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Dec 5 23:34] Using AVX-512 FFT length 840K, Pass1=192, Pass2=4480, clm=2, 4 threads
[Work thread Dec 5 23:34] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Dec 5 23:35] M15899909 stage 1 is 7.21% complete. Time: 71.994 sec.
[Work thread Dec 5 23:36] M15899909 stage 1 is 14.43% complete. Time: 72.088 sec.
[Work thread Dec 5 23:37] M15899909 stage 1 is 21.66% complete. Time: 72.187 sec.
[Work thread Dec 5 23:38] M15899909 stage 1 is 28.88% complete. Time: 72.557 sec.
[Work thread Dec 5 23:40] M15899909 stage 1 is 36.10% complete. Time: 72.066 sec.
[Work thread Dec 5 23:41] M15899909 stage 1 is 43.32% complete. Time: 72.679 sec.
[Work thread Dec 5 23:42] M15899909 stage 1 is 50.54% complete. Time: 72.038 sec.
[Work thread Dec 5 23:43] M15899909 stage 1 is 57.76% complete. Time: 72.377 sec.
[Work thread Dec 5 23:45] M15899909 stage 1 is 64.98% complete. Time: 72.572 sec.
[Work thread Dec 5 23:46] M15899909 stage 1 is 72.21% complete. Time: 71.292 sec.
[Work thread Dec 5 23:47] M15899909 stage 1 is 79.43% complete. Time: 72.210 sec.
[Work thread Dec 5 23:48] M15899909 stage 1 is 86.65% complete. Time: 71.730 sec.
[Work thread Dec 5 23:49] M15899909 stage 1 is 93.87% complete. Time: 71.888 sec.
[Work thread Dec 5 23:50] M15899909 stage 1 complete. 2769682 transforms. Total time: 998.978 sec.
[Work thread Dec 5 23:50] Conversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 4.975 sec.
[Work thread Dec 5 23:50] With trial factoring done to 2^73, optimal B2 is 327*B1 = 313920000.
[Work thread Dec 5 23:50] If no prior P-1, chance of a new factor is 6.5%
[Work thread Dec 5 23:50] Switching to AVX-512 FFT length 960K, Pass1=1536, Pass2=640, clm=1, 4 threads
[Work thread Dec 5 23:50] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Dec 5 23:50] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Dec 5 23:50] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Dec 5 23:50] With trial factoring done to 2^73, optimal B2 is 272*B1 = 261120000.
[Work thread Dec 5 23:50] If no prior P-1, chance of a new factor is 6.34%
[Work thread Dec 5 23:50] Using 13853MB of memory.  D: 3570, 384x1447 polynomial multiplication.
[Work thread Dec 5 23:50] Setting affinity to run polymult helper thread on CPU core #2
[Work thread Dec 5 23:50] Setting affinity to run polymult helper thread on CPU core #3
[Work thread Dec 5 23:50] Setting affinity to run polymult helper thread on CPU core #4
[Work thread Dec 5 23:51] Round off: 0, poly_size: 2, EB: 1.67556, SM: 2.39624
[Work thread Dec 5 23:51] Round off: 0, poly_size: 4
[Work thread Dec 5 23:51] Round off: 0, poly_size: 8
[Work thread Dec 5 23:51] Round off: 0, poly_size: 16
[Work thread Dec 5 23:51] Round off: 0, poly_size: 32
[Work thread Dec 5 23:51] Round off: 0, poly_size: 64
[Work thread Dec 5 23:51] Round off: 0, poly_size: 128
[Work thread Dec 5 23:51] Round off: 0, poly_size: 256
[Work thread Dec 5 23:51] Round off: 0, poly_size: 512
[Work thread Dec 5 23:51] Stage 2 init complete. 10134 transforms. Time: 24.187 sec.
[Work thread Dec 5 23:51] Round off: 0
[Work thread Dec 5 23:56] M15899909 stage 2 is 0.00% complete. Time: 288.071 sec.
[Work thread Dec 6 00:00] M15899909 stage 2 is 0.00% complete. Time: 286.923 sec.
[Work thread Dec 6 00:05] M15899909 stage 2 is 0.00% complete. Time: 286.369 sec.
[Work thread Dec 6 00:10] M15899909 stage 2 is 0.00% complete. Time: 287.448 sec.
[Work thread Dec 6 00:10] M15899909 stage 2 complete. 833973 transforms. Total time: 1172.719 sec.
[Work thread Dec 6 00:10] Stage 2 GCD complete. Time: 3.159 sec.
[Work thread Dec 6 00:10] M15899909 completed P-1, B1=960000, B2=261641730, Wi8: 94D99AFF
[Comm thread Dec 6 00:10] Sending result to server: UID: ***/ibm1, M15899909 completed P-1, B1=960000, B2=261641730, Wi8: 94D99AFF

Code:

[Work thread Dec 5 23:46] P-1 on M15886219 with B1=960000, B2=TBD
[Work thread Dec 5 23:46] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Dec 5 23:46] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Dec 5 23:46] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Dec 5 23:46] Using AVX-512 FFT length 840K, Pass1=1344, Pass2=640, clm=1, 4 threads
[Work thread Dec 5 23:48] M15886219 stage 1 is 7.21% complete. Time: 88.113 sec.
[Work thread Dec 5 23:49] M15886219 stage 1 is 14.43% complete. Time: 87.418 sec.
[Work thread Dec 5 23:50] M15886219 stage 1 is 21.66% complete. Time: 87.636 sec.
[Work thread Dec 5 23:52] M15886219 stage 1 is 28.88% complete. Time: 87.543 sec.
[Work thread Dec 5 23:53] M15886219 stage 1 is 36.10% complete. Time: 86.928 sec.
[Work thread Dec 5 23:55] M15886219 stage 1 is 43.32% complete. Time: 86.476 sec.
[Work thread Dec 5 23:56] M15886219 stage 1 is 50.54% complete. Time: 87.378 sec.
[Work thread Dec 5 23:58] M15886219 stage 1 is 57.76% complete. Time: 87.532 sec.
[Work thread Dec 5 23:59] M15886219 stage 1 is 64.98% complete. Time: 87.120 sec.
[Work thread Dec 6 00:01] M15886219 stage 1 is 72.21% complete. Time: 86.941 sec.
[Work thread Dec 6 00:02] M15886219 stage 1 is 79.43% complete. Time: 87.057 sec.
[Work thread Dec 6 00:03] M15886219 stage 1 is 86.65% complete. Time: 86.790 sec.
[Work thread Dec 6 00:05] M15886219 stage 1 is 93.87% complete. Time: 86.376 sec.
[Work thread Dec 6 00:06] M15886219 stage 1 complete. 2769682 transforms. Total time: 1207.141 sec.
[Work thread Dec 6 00:06] Conversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 4.986 sec.
[Work thread Dec 6 00:06] With trial factoring done to 2^73, optimal B2 is 655*B1 = 628800000.
[Work thread Dec 6 00:06] If no prior P-1, chance of a new factor is 7.12%
[Work thread Dec 6 00:06] Switching to AVX-512 FFT length 960K, Pass1=1536, Pass2=640, clm=1, 4 threads
[Work thread Dec 6 00:06] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Dec 6 00:06] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Dec 6 00:06] Setting affinity to run helper thread 2 on CPU core #3
[Work thread Dec 6 00:06] With trial factoring done to 2^73, optimal B2 is 540*B1 = 518400000.
[Work thread Dec 6 00:06] If no prior P-1, chance of a new factor is 6.94%
[Work thread Dec 6 00:06] Using 29696MB of memory.  D: 6930, 720x3210 polynomial multiplication.
[Work thread Dec 6 00:06] Setting affinity to run polymult helper thread on CPU core #2
[Work thread Dec 6 00:06] Setting affinity to run polymult helper thread on CPU core #3
[Work thread Dec 6 00:06] Setting affinity to run polymult helper thread on CPU core #4
[Work thread Dec 6 00:06] Round off: 0, poly_size: 2, EB: 1.11845, SM: 2.68872
[Work thread Dec 6 00:06] Round off: 0, poly_size: 4
[Work thread Dec 6 00:07] Round off: 0, poly_size: 8
[Work thread Dec 6 00:07] Round off: 0, poly_size: 16
[Work thread Dec 6 00:07] Round off: 0, poly_size: 32
[Work thread Dec 6 00:07] Round off: 0, poly_size: 64
[Work thread Dec 6 00:07] Round off: 0, poly_size: 128
[Work thread Dec 6 00:07] Round off: 0, poly_size: 256
[Work thread Dec 6 00:07] Round off: 0, poly_size: 512
[Work thread Dec 6 00:07] Round off: 0, poly_size: 1024
[Work thread Dec 6 00:07] Stage 2 init complete. 20334 transforms. Time: 51.794 sec.
[Work thread Dec 6 00:08] Round off: 0
[Work thread Dec 6 00:12] M15886219 stage 2 is 0.00% complete. Time: 315.540 sec.
[Work thread Dec 6 00:18] M15886219 stage 2 is 0.00% complete. Time: 315.115 sec.
[Work thread Dec 6 00:23] M15886219 stage 2 is 0.00% complete. Time: 315.800 sec.
[Work thread Dec 6 00:26] M15886219 stage 2 complete. 777679 transforms. Total time: 1119.132 sec.
[Work thread Dec 6 00:26] Stage 2 GCD complete. Time: 3.176 sec.
[Work thread Dec 6 00:26] M15886219 completed P-1, B1=960000, B2=518523390, Wi8: 85E1C98F
[Comm thread Dec 6 00:26] Sending result to server: UID: ***/ibm4, M15886219 completed P-1, B1=960000, B2=518523390, Wi8: 85E1C98F

Xyzzy · 2021-12-06, 01:45

axn · 2021-12-06, 02:02

Quote:

Originally Posted by Xyzzy

Thanks for the benchmark, Mike!

ATH · 2021-12-06, 04:55

Switched from build2 to build3 during Stage1. When it reaches P-1 Stage2 it freezes during initialization and Prime95 is using 0% CPU and 2.9 GB RAM for hours (instead of the allotted 18GB).
Prime95 cannot be closed normally but has to be killed. I tried 3 times with the same result (starting from the same build 2 stage1 savefile). I sent the files in a private message.

2021-12-06, 04:55	#110
ATH Einyen Dec 2003 Denmark D7C₁₆ Posts	Switched from build2 to build3 during Stage1. When it reaches P-1 Stage2 it freezes during initialization and Prime95 is using 0% CPU and 2.9 GB RAM for hours (instead of the allotted 18GB). Prime95 cannot be closed normally but has to be killed. I tried 3 times with the same result (starting from the same build 2 stage1 savefile). I sent the files in a private message. Last fiddled with by ATH on 2021-12-06 at 05:02

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Do not post your results here!	kar_bon	Prime Wiki	40	2022-04-03 19:05
what should I post ?	science_man_88	science_man_88	24	2018-10-19 23:00
Where to post job ad?	xilman	Linux	2	2010-12-15 16:39
Moderated Post	kar_bon	Forum Feedback	3	2010-09-28 08:01
Something that I just had to post/buy	dave_0273	Lounge	1	2005-02-27 18:36

2021-12-05, 16:43	#103
axn Jun 2003 2³×683 Posts	Presumably you're using AVX to mean AVX-512. I wonder how much of the stage 2 improvement is due to AVX-512 specifically, and how much due to the improvements to stage 2 itself. For instance, for my particular case, stage 2 run time (excluding init) went down from 330s to 212s by switching from build 1 to build 3 (build 2 in linux was not timed due to multithreading issues).

2021-12-05, 19:26	#105
tha Dec 2002 881 Posts	My current system is 4 years old. It is based on a Z-170 Asus motherboard. At the time 16 Gb of RAM (4 x 4 Gb) was the sweet spot. So I re-evaluated it and today 64 Gb (4 x 16 Gb) costs the same as 16 Gb then. So, I've ordered it and it will be delivered later this week.

2021-12-05, 20:29	#106
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 3·5²·71 Posts	And it seems the TBD value of B2 ... at least for this 1 PC ... is brilliant now. The value chosen is 256xB1; which for this PC gave very similar Stage 1 and Stage 2 times.

2021-12-06, 01:45	#108
Xyzzy Aug 2002 2·3²·13·37 Posts