Prime95 v30.3

Prime95 · 2020-08-10, 02:26

Prime95 version 30.3 build 6 is available.

From whatsnew.txt:

Code:

1) PRP proofs.  This allows GIMPS to double-check a PRP test at less than 1% of the cost of a full PRP test!
   PRP proofs require lots of temporary disk space.  See readme.txt for details.
   PRP proofs require uploading a large proof file.  See readme.txt for details.
   PRP proof verifications require downloading a modest verification file.  See readme.txt for details.
2) Proofs automatically uploaded to server in v30.2.
3) First time LL, World-record LL, 100M-digit LL work preference is deprecated.
4) New resource limits menu choice and dialog box.  Consult readme.txt before making changes to these settings.
   Some options previously in Test/Worker Windows and Options/CPU are moved to the resources dialog box.
5) LL-DC and PRP-DC combined into a single work preference.
6) Warning raised if temporary disk space is less than 1.5GB -- you may not get first time prime tests.
7) Thanks to Mihai Preda, the P-1 probability calculator has been improved.  This change results in a
   lower optimal B1 value and higher optimal B2 value.

Prime95 · 2020-08-10, 02:27

Placeholder for bugs fixes

1) If a proof is not being generated prime95 erroneously tries to which gives a file not found error. Fixed in 30.3 build 2.
2) An error getting the CERT starting value loops forever trying to get the value from the server. Semi-fixed in 30.3 build 2 -- now tries once an hour.
3) PRP test on a big exponent that would benefit from trial factoring was not getting the needed TF. Fixed in 30.3 build 2 -- code copied from the LL code path.
4) Most PRP cofactor proof files will not upload. Fixed in 30.3 build 2
5) Adding PRP=1,2,exponent,-1 defaulted to "TF needed" and "no P-1 needed". This is inconsistent. The new default will be "no TF needed" and "no P-1 needed". Fixed in 30.3 build 3.
6) Debug level 2 in PrimeNet section of prime.txt outputs tons of binary proof data. Fixed in 30.3 build 3 by outputting <binary> instead.
7) Certification error message "MD5 checksum mismatch" can be misleading. Fixed in 30.3 build 4.
8) If SequentialWorkTodo=0 and you are doing PRP work, then your jobs will mistakenly be interrupted for priority work. Fixed in 30.3 build 4.
9) The setting for maximum CERT CPU percentage not working properly. Fixed in 30.3 build 4.
10) An error in pre-allocating the interim residues file disk space improperly increased the proof power rather than decreased the proof power. Fixed in 30.3 build 6.
11) Wrong hwloc DLL included for win32. Fixed in build 6.
12) Crashes during load on Vista 64-bit and earlier with K32GetModuleFileNameExA not found. Possibly fixed in 30.3 build 6.

Prime95 · 2020-08-10, 02:34

The DLL for libgmp is not compatible with the previous one. This was causing issues for some users (finding the wrong version of the DLL). In this release, the DLL has been renamed to libgmp-gw1.dll.

For those with limited disk space, this version can simulate a higher proof power using less disk space but a larger proof file to upload. For example we can simulate a power = 8 proof by doing a power=7 proof on the first half of the PRP test and a second power=7
proof on the second half of a PRP test. This uses half the temporary disk space and twice the upload bandwidth.

LL-DC and PRP-DC are combined into a single work preference.

Error detection/recovery in prime95 is much more robust.

pepi37 · 2020-08-10, 07:23

I try it on sequence 95*10^n-1

Quote:

[Aug 10 09:20:55] Waiting 5 seconds to stagger worker starts.
[Aug 10 09:21:00] Worker starting
[Aug 10 09:21:00] Setting affinity to run worker on CPU core #4
[Aug 10 09:21:00] Setting affinity to run helper thread 1 on CPU core #5
[Aug 10 09:21:00] Setting affinity to run helper thread 2 on CPU core #6
[Aug 10 09:21:08] Resuming PRP test of 95*10^1809228-1 using FMA3 FFT length 448K, Pass1=448, Pass2=1K, clm=4, 3 threads
[Aug 10 09:21:08] Iteration: 4072948 / 6010131 [67.76%].
[Aug 10 09:22:34] Generating proof for 95*10^1809228-1. Proof power = 0, Hash length = 64
[Aug 10 09:22:34] Cannot open PRP proof interim residues file:
[Aug 10 09:22:34] Errno: 2, No such file or directory
[Aug 10 09:22:34] DOSerrno: 3
[Aug 10 09:22:34] Waiting 5 minutes to try proof generation again.
[Aug 10 09:22:34] Waiting five minutes before restarting.

pinhodecarlos · 2020-08-10, 10:01

Do you need a proof for your manual work outside GIMPS boundary?

Prime95 · 2020-08-10, 15:14

Proofs only work for base 2 numbers. I'll work on a fix for the bug you found.

kriesel · 2020-08-10, 16:47

re prime95/mprime v30.3:

Quote:

Originally Posted by Prime95

Error detection/recovery in prime95 is much more robust.

Always good. In what ways was it improved?
(Did GEC get implemented for the CERT runs?)

Prime95 · 2020-08-10, 17:16

Quote:

Originally Posted by kriesel

re prime95/mprime v30.3:Always good. In what ways was it improved?
(Did GEC get implemented for the CERT runs?)

GEC for CERT -- no.

Interim residues are now written with an MD5 checksum.

Entire proof file is read back after being written to compare MD5 checksums.

Numerous cases of sloppy error handling fixed. I code proof-of-concept with sloppy error handling first. Then I go back and clean up.

ATH · 2020-08-10, 18:54

Error on trying to upload a proof file for a very small PRP CF in mprime on EC2:

"Error getting number from proof header"

Edit: The file is still not uploaded. Trying to attach it here.

Code:

[Comm thread Aug 10 14:32:19] Got assignment 9B32734729207C3C235DD65DBD241B7A: PRP M2344571
[Work thread Aug 10 14:32:19] Resuming.
[Comm thread Aug 10 14:32:19] Sending expected completion date for M2344571: Aug 10 2020
[Work thread Aug 10 14:32:19] Starting Gerbicz error-checking PRP test of M2344571/5123586391565541063572897 using AVX-512 FFT length 120K, Pass1=192, Pass2=640, clm=1
[Work thread Aug 10 14:32:19] Preallocating disk space for the proof interim residues file p2344571.residues
[Work thread Aug 10 14:32:19] PRP proof using power=7 and 64-bit hash size.
[Work thread Aug 10 14:32:19] Proof requires 0.0GB of temporary disk space and uploading a 2MB proof file.
[Comm thread Aug 10 14:32:20] Done communicating with server.
[Work thread Aug 10 14:36:30] Iteration: 1000000 / 2344576 [42.651635%], roundoff: 0.101, ms/iter:  0.249, ETA: 00:05:34
[Work thread Aug 10 14:36:30] Gerbicz error check passed at iteration 1000000.
[Work thread Aug 10 14:36:30] M2344571/5123586391565541063572897 interim PRP residue A3E2916C933308D5 at iteration 1000000
[Work thread Aug 10 14:40:41] Iteration: 2000000 / 2344576 [85.303270%], roundoff: 0.110, ms/iter:  0.249, ETA: 00:01:25
[Work thread Aug 10 14:40:41] Gerbicz error check passed at iteration 2000000.
[Work thread Aug 10 14:40:41] M2344571/5123586391565541063572897 interim PRP residue 5A3CD751E40E3CA4 at iteration 2000000
[Work thread Aug 10 14:42:08] Gerbicz error check passed at iteration 2344569.
[Work thread Aug 10 14:42:13] Generating proof for M2344571.  Proof power = 7, Hash length = 64
[Work thread Aug 10 14:42:13] Root hash = 616605A749A0C958D8DF3E72FC32C804BBD2B1288AA709C47FE65853A297850D
[Work thread Aug 10 14:42:13] hash0 = 1ACBF839AA5DF27F
[Work thread Aug 10 14:42:13] hash1 = 782DBDD6964A10A1
[Work thread Aug 10 14:42:13] hash2 = D1EC5C0BEAC72ADA
[Work thread Aug 10 14:42:14] hash3 = 49AB07B48B882843
[Work thread Aug 10 14:42:14] hash4 = 0055006F7821EA7E
[Work thread Aug 10 14:42:15] hash5 = E7ADE8B93091DF2B
[Work thread Aug 10 14:42:17] Proof construction cost 9348 squarings
[Work thread Aug 10 14:42:17] Proof verification will cost 18317 squarings
[Work thread Aug 10 14:42:17] M2344571/5123586391565541063572897 is not prime.  Type-5 RES64: CD8AD0D724D8D9__. Wh8: 396B1DA3,1632460,00000000
[Comm thread Aug 10 14:42:17] Sending result to server: UID: athath/ec2-PRPCF, M2344571/5123586391565541063572897 is not prime.  Type-5 RES64: CD8AD0D724D8D9__. Wh8: 396B1DA3,1632460,00000000, AID: 9B32734729207C3C235DD65DBD241B7A
[Comm thread Aug 10 14:42:17] 



[Comm thread Aug 10 14:54:13] MD5 of p2344571.proof is 015d98c962f822406773acc18ae0933d
[Comm thread Aug 10 14:54:13] Error getting number from proof header
[Comm thread Aug 10 15:59:13] MD5 of p2344571.proof is 015d98c962f822406773acc18ae0933d
[Comm thread Aug 10 15:59:13] Error getting number from proof header



[Comm thread Aug 10 17:04:13] MD5 of p2344571.proof is 015d98c962f822406773acc18ae0933d
[Comm thread Aug 10 17:04:13] Error getting number from proof header
[Comm thread Aug 10 18:09:13] MD5 of p2344571.proof is 015d98c962f822406773acc18ae0933d
[Comm thread Aug 10 18:09:13] Error getting number from proof header

Prime95 · 2020-08-10, 23:30

Quote:

Originally Posted by ATH

Error on trying to upload a proof file for a very small PRP CF

Will have a fix tonight.

Prime95 · 2020-08-11, 00:59

Please upgrade to 30.3 build 2.

Two of the reported bugs were serious:
1) PRPs without proofs (say, if you upgraded from v29) will fail at the the end of the PRP test by trying to create a proof file.
2) Inability to upload proofs from cofactor PRP tests.

2020-08-10, 02:27	#2
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 73×113 Posts	Placeholder for bugs fixes 1) If a proof is not being generated prime95 erroneously tries to which gives a file not found error. Fixed in 30.3 build 2. 2) An error getting the CERT starting value loops forever trying to get the value from the server. Semi-fixed in 30.3 build 2 -- now tries once an hour. 3) PRP test on a big exponent that would benefit from trial factoring was not getting the needed TF. Fixed in 30.3 build 2 -- code copied from the LL code path. 4) Most PRP cofactor proof files will not upload. Fixed in 30.3 build 2 5) Adding PRP=1,2,exponent,-1 defaulted to "TF needed" and "no P-1 needed". This is inconsistent. The new default will be "no TF needed" and "no P-1 needed". Fixed in 30.3 build 3. 6) Debug level 2 in PrimeNet section of prime.txt outputs tons of binary proof data. Fixed in 30.3 build 3 by outputting <binary> instead. 7) Certification error message "MD5 checksum mismatch" can be misleading. Fixed in 30.3 build 4. 8) If SequentialWorkTodo=0 and you are doing PRP work, then your jobs will mistakenly be interrupted for priority work. Fixed in 30.3 build 4. 9) The setting for maximum CERT CPU percentage not working properly. Fixed in 30.3 build 4. 10) An error in pre-allocating the interim residues file disk space improperly increased the proof power rather than decreased the proof power. Fixed in 30.3 build 6. 11) Wrong hwloc DLL included for win32. Fixed in build 6. 12) Crashes during load on Vista 64-bit and earlier with K32GetModuleFileNameExA not found. Possibly fixed in 30.3 build 6. Last fiddled with by Prime95 on 2020-09-10 at 21:51

2020-08-10, 02:34	#3
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 73·113 Posts	New in v30.3 compared to 30.2 The DLL for libgmp is not compatible with the previous one. This was causing issues for some users (finding the wrong version of the DLL). In this release, the DLL has been renamed to libgmp-gw1.dll. For those with limited disk space, this version can simulate a higher proof power using less disk space but a larger proof file to upload. For example we can simulate a power = 8 proof by doing a power=7 proof on the first half of the PRP test and a second power=7 proof on the second half of a PRP test. This uses half the temporary disk space and twice the upload bandwidth. LL-DC and PRP-DC are combined into a single work preference. Error detection/recovery in prime95 is much more robust. Last fiddled with by Prime95 on 2020-08-10 at 02:35

2020-08-10, 10:01	#5
pinhodecarlos "Carlos Pinho" Oct 2011 Milton Keynes, UK 3×1,721 Posts	Do you need a proof for your manual work outside GIMPS boundary?

2020-08-10, 15:14	#6
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 73×113 Posts	Proofs only work for base 2 numbers. I'll work on a fix for the bug you found.

2020-08-11, 00:59	#11
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 73·113 Posts	Please upgrade to 30.3 build 2. Two of the reported bugs were serious: 1) PRPs without proofs (say, if you upgraded from v29) will fail at the the end of the PRP test by trying to create a proof file. 2) Inability to upload proofs from cofactor PRP tests.