https://mersenneforum.org/showthread.php?t=24280

This replaces 3.8.22 and includes a fix for a bug which could give incorrect results under some situations with AVX-512 on Windows. The headline new feature over 3.8.21 is the addition of AVX-512 support.

When Win64 binaries are formally available I'll recheck the known error case of 3.8.22 and also go through the manual testing part of the LLR release used on previous versions.

Well, 3.8.23 is out.

Win64 executable is faster by a considerable margin on AVX-512.
I happen to have 2 CPUs, 16-cores AVX-512 capable box and obviously keen to get testing.

How about that?
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Well, 3.8.23 is out.

Win64 executable is faster by a considerable margin on AVX-512.
I happen to have 2 CPUs, 16-cores AVX-512 capable box and obviously keen to get testing.

How about that?

Ok, GCW LLR first two task, both waiting for validation.
Single task, 7 threads, each took about 3,5 hours.
Note that this CPU runs only ~2 GHz.

https://www.primegrid.com/workunit.php?wuid=607915830
https://www.primegrid.com/workunit.php?wuid=607854512

Will try other LLR apps.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Forgot to post earlier, I had run both the known bug case for 3.8.22 and also the manual tests as used in previous LLR releases here.

Short version: everything seems to be working.

387*2^3322763+1 is prime! (1000254 decimal digits) Time : 764.756 sec.
129897*68^129897+1 is prime! (238043 decimal digits) Time : 74.772 sec.
59912*5^1500861+1 is prime! (1049062 decimal digits) Time : 1224.062 sec.
1300274732577*2^1290000-1 is prime! (388341 decimal digits) Time : 145.108 sec.
3752948*2^3752948-1 is prime! (1129757 decimal digits) Time : 1437.859 sec.
94112*121^94112-1 is prime! (196021 decimal digits, P = 9) Time : 836.405 sec.
21181*2^1576364+1 is not prime. Proth RES64: 14DF8A91A23FA805 Time : 179.954 sec.
21181*2^1576364+1 is not prime. RES64: 70B8568133B48805. OLD64: 522903839B1D980C Time : 181.302 sec.
152267*2^15220227+1 is not prime. Proth RES64: 897235580F34F0F9 Time : 10864.220 sec.
4344392810277*2^1290000-1 is not prime. LLR Res64: 7D325B0469A1226E Time : 172.028 sec.

Microsoft Windows [Version 10.0.17763.195]
(c) 2018 Microsoft Corporation. All rights reserved.

D:\llr64 3.8.23>test

D:\llr64 3.8.23>llr -d -q"387*2^3322763+1" -t6
Starting Proth prime test of 387*2^3322763+1
Using all-complex AVX-512 FFT length 240K, Pass1=640, Pass2=384, clm=1, 6 threads, a = 5
387*2^3322763+1, bit: 1210000 / 3322771 [36.41%]. Time per bit: 0.209 ms.
Caught signal. Terminating.
Terminate batch job (Y/N)? y

D:\llr64 3.8.23>test

D:\llr64 3.8.23>llr -d -q"387*2^3322763+1" -t2
Resuming Proth prime test of 387*2^3322763+1 at bit 1214538 [36.55%]
Using all-complex AVX-512 FFT length 240K, Pass1=640, Pass2=384, clm=1, 2 threads, a = 5
387*2^3322763+1, bit: 1480000 / 3322771 [44.54%]. Time per bit: 0.325 ms.
Caught signal. Terminating.
Terminate batch job (Y/N)? y

D:\llr64 3.8.23>test

D:\llr64 3.8.23>llr -d -q"387*2^3322763+1" -t6
Resuming Proth prime test of 387*2^3322763+1 at bit 1481864 [44.59%]
Using all-complex AVX-512 FFT length 240K, Pass1=640, Pass2=384, clm=1, 6 threads, a = 5
387*2^3322763+1 is prime! (1000254 decimal digits) Time : 764.756 sec.

D:\llr64 3.8.23>llr -d -q"129897*68^129897+1" -t6
Base factorized as : 2^2*17
Base prime factor(s) taken : 17
Starting N-1 prime test of 129897*68^129897+1
Using zero-padded AVX-512 FFT length 120K, Pass1=192, Pass2=640, clm=1, 6 threads, a = 3
129897*68^129897+1 may be prime, trying to compute gcd's
3^((N-1)/17)-1 is coprime to N!
129897*68^129897+1 is prime! (238043 decimal digits) Time : 74.772 sec.

D:\llr64 3.8.23>llr -d -q"59912*5^1500861+1" -t6
Base prime factor(s) taken : 5
Starting N-1 prime test of 59912*5^1500861+1
Using all-complex AVX-512 FFT length 336K, Pass1=896, Pass2=384, clm=1, 6 threads, a = 3
59912*5^1500861+1 may be prime, trying to compute gcd's
3^((N-1)/5)-1 is coprime to N!
59912*5^1500861+1 is prime! (1049062 decimal digits) Time : 1224.062 sec.

D:\llr64 3.8.23>llr -d -q"1300274732577*2^1290000-1" -t6
Starting Lucas Lehmer Riesel prime test of 1300274732577*2^1290000-1
Using zero-padded AVX-512 FFT length 128K, Pass1=128, Pass2=1K, clm=1, 6 threads
V1 = 5 ; Computing U0...done.
1300274732577*2^1290000-1 is prime! (388341 decimal digits) Time : 145.108 sec.

D:\llr64 3.8.23>llr -d -q"3752948*2^3752948-1" -t6
Starting Lucas Lehmer Riesel prime test of 3752948*2^3752948-1
Using zero-padded AVX-512 FFT length 384K, Pass1=1K, Pass2=384, clm=1, 6 threads
V1 = 4 ; Computing U0...done.
3752948*2^3752948-1 is prime! (1129757 decimal digits) Time : 1437.859 sec.

D:\llr64 3.8.23>llr -d -q"94112*121^94112-1" -t6
Base prime factor(s) taken : 11
Starting N+1 prime test of 94112*121^94112-1
Using zero-padded AVX-512 FFT length 84K, Pass1=192, Pass2=448, clm=1, 6 threads, a = 3
94112*121^94112-1 may be prime. Starting Lucas sequence...
Using zero-padded AVX-512 FFT length 84K, Pass1=192, Pass2=448, clm=1, 6 threads, P = 4
94112*121^94112-1 may be prime, trying to compute gcd's
94112*121^94112-1 may be prime, but N divides U((N+1)/11), P = 4
Restarting Lucas sequence with P = 8
Using zero-padded AVX-512 FFT length 84K, Pass1=192, Pass2=448, clm=1, 6 threads, P = 8
94112*121^94112-1 may be prime, trying to compute gcd's
94112*121^94112-1 may be prime, but N divides U((N+1)/11), P = 8
Restarting Lucas sequence with P = 9
Using zero-padded AVX-512 FFT length 84K, Pass1=192, Pass2=448, clm=1, 6 threads, P = 9
94112*121^94112-1 may be prime, trying to compute gcd's
U((N+1)/11) is coprime to N!
94112*121^94112-1 is prime! (196021 decimal digits, P = 9) Time : 836.405 sec.

D:\llr64 3.8.23>llr -d -q"21181*2^1576364+1" -t6
Starting Proth prime test of 21181*2^1576364+1
Using all-complex AVX-512 FFT length 128K, Pass1=128, Pass2=1K, clm=1, 6 threads, a = 3
21181*2^1576364+1 is not prime. Proth RES64: 14DF8A91A23FA805 Time : 179.954 sec.

D:\llr64 3.8.23>llr -d -q"21181*2^1576364+1" -oForcePRP=1 -oFermatPRPtest=1 -t6
Starting probable prime test of 21181*2^1576364+1
Using all-complex AVX-512 FFT length 128K, Pass1=128, Pass2=1K, clm=1, 6 threads, a = 3
21181*2^1576364+1 is not prime. RES64: 70B8568133B48805. OLD64: 522903839B1D980C Time : 181.302 sec.

D:\llr64 3.8.23>llr -d -q"152267*2^15220227+1" -t6
Starting Proth prime test of 152267*2^15220227+1
Using all-complex AVX-512 FFT length 1440K, Pass1=768, Pass2=1920, clm=2, 6 threads, a = 3
152267*2^15220227+1 is not prime. Proth RES64: 897235580F34F0F9 Time : 10864.220 sec.

D:\llr64 3.8.23>llr -d -q"4344392810277*2^1290000-1" -t6
Starting Lucas Lehmer Riesel prime test of 4344392810277*2^1290000-1
Using zero-padded AVX-512 FFT length 128K, Pass1=128, Pass2=1K, clm=1, 6 threads
V1 = 5 ; Computing U0...done.
4344392810277*2^1290000-1 is not prime. LLR Res64: 7D325B0469A1226E Time : 172.028 sec.

D:\llr64 3.8.23>

So far, I've several live LLR tasks validated agains live LLR version.

Validated
GCW
PPS
PPSE
MEGA
SGS
SR5

Waiting: 321
In progress: SOB.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Doing things in a rush at night means mistakes. I forgot to point app_info to 3.8.23 and it was still using 3.8.22. Doh! Only noticed as I went in to add more projects. About to try TRP on it, but will run some PPSE now corrected to 3.8.23 first.

This only affects boinc results overnight, the manual testing posted before was on 3.8.23.

Will it eventually become the standard for everyone if we just wait a bit?

I played with an app_info for another project the other day and lost some cpu units here, obviously I got it wrong, so am hoping this exe will just be the default at some point.

Will it eventually become the standard for everyone if we just wait a bit?

Yes, if there are no errors found.

About how long do you think that will take? I have a Skylake-X that I would like to try out. If it is going to be just a day or two, I'll wait. Otherwise, I will go down the app_info.xml path.

That said, can someone please post an app_info.xml sample? I would be running it with multiple cores, if that makes a difference.

This should be (about) complete list for PG's LLR apps.
I've created the list sorted by app names and made it somehow grouped.
Number of threads can be easily changed in app_config.xml. Note that shorter tasks have 3 threads, longer have 7.
Also note that you need those 3 files mentioned downloaded manually and placed in \projects\www.primegrid.com\


app_info.xml

You take the file from here (extract from the archive, obviously) and rename to cllr64.3.8.23.exe
The llr.ini.6.07 and primegrid_llr_wrapper_8.01_windows_x86_64.exe you can get from running normal PG LLR tasks. Most likely you already have them in your projects\www.primegrid.com\ folder.

thx Honza for app_info/config.xml

run very quick test on i9-7920x, happy with the result 3.8.21 (FMA3) vs 3.8.23 (AVX512)

llr -d -q"152267*2^15220227+1" -t12
Resuming Proth prime test of 152267*2^15220227+1 at bit 711237 [4.67%]
Using all-complex FMA3 FFT length 1440K, Pass1=384, Pass2=3840, 12 threads, a = 3
152267*2^15220227+1, bit: 729200 / 15220244 [4.79%]. Time per bit: 0.704 ms.

llr512 -d -q"152267*2^15220227+1" -t12
Resuming Proth prime test of 152267*2^15220227+1 at bit 332924 [2.18%]
Using all-complex AVX-512 FFT length 1440K, Pass1=768, Pass2=1920, clm=2, 12 threads, a = 3
152267*2^15220227+1, bit: 711200 / 15220244 [4.67%]. Time per bit: 0.548 ms.


AVX2/FMA3 3500MHz: ~0.700ms/iter, CPU Power 189W, hottest CPU core 69C, llr 3.8.21
AVX512 3000MHz: ~0.550ms/iter, CPU Power 160W, hottest CPU core 61C, llr 3.8.23

quite unexpectedly (well not so much for me, ive been running mersenne with AV512 for some time) AVX512 is faster even with better CPU power and better temps. what a progress ;)

How did you get the time per bit info? The stderr for my completed tasks doesn't that level of detail.

How did you get the time per bit info? The stderr for my completed tasks doesn't that level of detail.

That's what the -d command line option does. You can see it in the command line that was posted.

Yes, I see that. But how is it used? Does it go somewhere in the app_info.xml or app_config.xml? Or some other method?

This should be (about) complete list for PG's LLR apps.
I've created the list sorted by app names and made it somehow grouped.
Number of threads can be easily changed in app_config.xml. Note that shorter tasks have 3 threads, longer have 7.
Also note that you need those 3 files mentioned downloaded manually and placed in \projects\www.primegrid.com\


app_info.xml

<app_info> <app><name>llr321</name><user_friendly_name>321 (LLR)</user_friendly_name></app> <app><name>llrCUL</name><user_friendly_name>Cullen (LLR)</user_friendly_name></app> <app><name>llrESP</name><user_friendly_name>Extended Sierpinski Problem</user_friendly_name></app> <app><name>llrGCW</name><user_friendly_name>Generalized Cullen/Woodall (LLR)</user_friendly_name></app> <app><name>llrMEGA</name><user_friendly_name>PPS-Mega (LLR)</user_friendly_name></app> <app><name>llrPPS</name><user_friendly_name>PPS (LLR)</user_friendly_name></app> <app><name>llrPPSE</name><user_friendly_name>PPSE (LLR)</user_friendly_name></app> <app><name>llrSOB</name><user_friendly_name>Seventeen or Bust</user_friendly_name></app> <app><name>llrSR5</name><user_friendly_name>Sierpinski/Riesel Base 5 Problem (LLR)</user_friendly_name></app> <app><name>llrTPS</name><user_friendly_name>Sophie Germain (LLR)</user_friendly_name></app> <app><name>llrTRP</name><user_friendly_name>The Riesel Problem (LLR)</user_friendly_name></app> <app><name>llrWOO</name><user_friendly_name>Woodall (LLR)</user_friendly_name></app> <file><name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</name><executable/></file> <file><name>llr.ini.6.07</name><executable/></file> <file><name>cllr64.3.8.23.exe</name></file> <app_version> <app_name>llr321</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrCUL</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrESP</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrGCW</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrMEGA</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrPPS</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrPPSE</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrSOB</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrSR5</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrTPS</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrTRP</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> <app_version> <app_name>llrWOO</app_name><version_num>801</version_num><api_version>6.12.43</api_version> <file_ref><file_name>primegrid_llr_wrapper_8.01_windows_x86_64.exe</file_name><main_program/></file_ref> <file_ref><file_name>cllr64.3.8.23.exe</file_name><open_name>primegrid_cllr.exe</open_name><copy_file/></file_ref> <file_ref><file_name>llr.ini.6.07</file_name><open_name>llr.ini</open_name><copy_file/></file_ref> </app_version> </app_info>

app_config.xml

<app_config> <app><name>llr321</name></app> <app_version><app_name>llr321</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrCUL</name></app> <app_version><app_name>llrCUL</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrESP</name></app> <app_version><app_name>llrESP</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrGCW</name></app> <app_version><app_name>llrGCW</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrMEGA</name></app> <app_version><app_name>llrMEGA</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrPPS</name></app> <app_version><app_name>llrPPS</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrPPSE</name></app> <app_version><app_name>llrPPSE</app_name><cmdline>-t 3</cmdline><avg_ncpus>3</avg_ncpus></app_version> <app><name>llrSOB</name></app> <app_version><app_name>llrSOB</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrSR5</name></app> <app_version><app_name>llrSR5</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrTPS</name></app> <app_version><app_name>llrTPS</app_name><cmdline>-t 3</cmdline><avg_ncpus>3</avg_ncpus></app_version> <app><name>llrTRP</name></app> <app_version><app_name>llrTRP</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> <app><name>llrWOO</name></app> <app_version><app_name>llrWOO</app_name><cmdline>-t 7</cmdline><avg_ncpus>7</avg_ncpus></app_version> </app_config>