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

[James Heinrich]

I don't normally do a lot of PRP, but I'm trying some on my new 7950X/Win11 build, and it's giving me somtimes-problems with proof permissions. Running as administrator-class user, so permissions shouldn't be much of an issue, and yet:

[Sun Dec 25 06:27:56 2022]
Cannot open PRP proof interim residues file: C:\Prime95\proofs\p115978963.residues
Errno: 13, Permission denied
DOSerrno: 32
[Sun Dec 25 08:15:13 2022]
MD5 error reading PRP proof interim residues file.
[Sun Dec 25 08:20:50 2022]
MD5 error reading PRP proof interim residues file.
Proof generation failed.
UID: JamesHeinrich/7950X, M115978963 is not prime.  RES64: 4C549CF1581886__. Wh4: B0659014,55181080,00000000, AID: <masked>

and then today:

[Mon Dec 26 04:43:10 2022]
Cannot open PRP proof interim residues file: C:\Prime95\proofs\p115978957.residues
Errno: 13, Permission denied
DOSerrno: 32
Cannot open PRP proof interim residues file: C:\Prime95\proofs\p115978957.residues
Errno: 13, Permission denied
DOSerrno: 32
Cannot create save files while holding proof interim residues in emergency memory.

The same system/configuration has produced successful proofs recently (e.g. 69842653). 3+TB free on drive. WorkerDiskSpace=50 (is that enough? the temp p115978957.residues I currently see is ~14GB).

[Prime95]

According to
https://learn.microsoft.com/en-us/wi...-codes--0-499-
DOS error number 32 is a sharing violation. Is it possible some other program was accessing the residues file?
Maybe a disk defragmenter or backup utility?

[pepi37]

Originally Posted by Prime95

You still need to do "decent" P-1. I would define decent as "the bounds at which any further increase in the bounds would eliminate candidates slower than a PRP test". Those bounds are not trivial to calculate.

I have 16GB of RAM and 8 core CPU .So how to configure Prime95 to use 4 workers and each worker to use 3 GB of RAM ( something must be left for windows)?

[kriesel]

It would be better to configure it to two workers, so that they can alternate using 12 GB to do adequate P-1 stage 2, or (assuming Mersennes) PRP in short enough time to obtain lower category assignments and avoid assignments expiring before completion.

[James Heinrich]

Originally Posted by Prime95

DOS error number 32 is a sharing violation. Is it possible some other program was accessing the residues file?

I checked and according to the timestamps it's likely that a backup program was accessing the .residues file at around the 06:27:56 timestamp. And that error probably(?) had knock-on effects that caused the MD5 error reading PRP proof interim residues file and proof generation failure about two hours later?

Would it be possible in v30.10 to treat all access errors to .residues files as temporarily-fatal? As in, if Prime95 attempts to access the .residues file and, for whatever reason, encounters an error doing so, it should suspend processing and print (and log) a message to that effect, pause for 5 minutes, and try again later. Maybe a backup utility is running, maybe the HDD is just busy at that moment, maybe it's on a network share an and the network is momentarily down; whatever the reason might be it seems inadvisable to continue processing at that point.

[Gobira]

How do I send data from one to GIMPS without using Prime95?

[pepi37]

Originally Posted by kriesel

It would be better to configure it to two workers, so that they can alternate using 12 GB to do adequate P-1 stage 2, or (assuming Mersennes) PRP in short enough time to obtain lower category assignments and avoid assignments expiring before completion.

If you answer to me: I work on 4*53^n+1 and 4*155^n+1 sequences. Ok, it can be two workers with 4 CPU cores and each take 6 GB of RAM. but how to configure this?

[Uncwilly]

Originally Posted by Gobira

How do I send data from one to GIMPS without using Prime95?

We have been dealing with your question in the thread that you started. https://www.mersenneforum.org/showthread.php?t=28333

[kriesel]

Originally Posted by pepi37

If you answer to me: I work on 4*53^n+1 and 4*155^n+1 sequences. Ok, it can be two workers with 4 CPU cores and each take 6 GB of RAM. but how to configure this?

That is suboptimal.
Let one worker (a) do stage 1 which does not need much ram, while the other (b) does stage 2 with almost all the available ram. Then they automatically switch; when b finishes using lots of ram, a runs stage 2 with lots of ram, b runs stage 1. Setup for that is easy. Just set maximum stage 2 allowable ram to ~12 GB using the prime95 GUI menus or mprime's text menus. Or stop the program, edit the program's local.txt, and restart the program. The program will handle the switching of memory allocation between workers automatically. (Match day and night settings, so it won't restart a stage 2 in progress, from a memory-amount-allowed change with time of day.)
See also the program's readme.txt and undoc.txt.

[kriesel]

[Dec 27 11:03:08] Worker starting
[Dec 27 11:03:08] Setting affinity to run worker on CPU core #1
[Dec 27 11:03:11] Setting affinity to run helper thread 1 on CPU core #2
[Dec 27 11:03:11] Setting affinity to run helper thread 2 on CPU core #3
[Dec 27 11:03:11] Setting affinity to run helper thread 9 on CPU core #10
[Dec 27 11:03:11] Setting affinity to run helper thread 4 on CPU core #5
[Dec 27 11:03:11] Setting affinity to run helper thread 6 on CPU core #7
[Dec 27 11:03:11] Setting affinity to run helper thread 5 on CPU core #6
[Dec 27 11:03:11] Setting affinity to run helper thread 7 on CPU core #8
[Dec 27 11:03:11] Setting affinity to run helper thread 8 on CPU core #9
[Dec 27 11:03:11] Setting affinity to run helper thread 3 on CPU core #4
[Dec 27 11:03:11] Setting affinity to run helper thread 10 on CPU core #11
[Dec 27 11:03:11] Setting affinity to run helper thread 11 on CPU core #12
[Dec 27 11:03:20] Resuming Gerbicz error-checking PRP test of M543656371 using AVX FFT length 30M, Pass1=2K, Pass2=15K, clm=4, 12 threads
[Dec 27 11:03:20] PRP proof using power=7x2 and 64-bit hash size.
[Dec 27 11:03:20] Proof requires 8.7GB of temporary disk space and uploading a 1087MB proof file.
[Dec 27 11:03:21] Iteration: 543644754 / 543656371 [99.997%].

Some time after that, it crashes and disappears. The worker window capture above is its third try in progress. It seems to reach the proof building stage before it crashes.

The other worker is a few hours behind but may have the same issue. The proof temporary residues files are currently hosted by a separate system and may have some missing residues due to the other system having gone offline at times.

[Dec 27 11:03:08] Waiting 5 seconds to stagger worker starts.
[Dec 27 11:03:13] Worker starting
[Dec 27 11:03:13] Setting affinity to run worker on CPU core #13
[Dec 27 11:03:16] Setting affinity to run helper thread 1 on CPU core #14
[Dec 27 11:03:16] Setting affinity to run helper thread 6 on CPU core #19
[Dec 27 11:03:16] Setting affinity to run helper thread 3 on CPU core #16
[Dec 27 11:03:16] Setting affinity to run helper thread 4 on CPU core #17
[Dec 27 11:03:16] Setting affinity to run helper thread 5 on CPU core #18
[Dec 27 11:03:16] Setting affinity to run helper thread 2 on CPU core #15
[Dec 27 11:03:16] Setting affinity to run helper thread 7 on CPU core #20
[Dec 27 11:03:16] Setting affinity to run helper thread 8 on CPU core #21
[Dec 27 11:03:16] Setting affinity to run helper thread 9 on CPU core #22
[Dec 27 11:03:16] Setting affinity to run helper thread 10 on CPU core #23
[Dec 27 11:03:16] Setting affinity to run helper thread 11 on CPU core #24
[Dec 27 11:03:25] Resuming Gerbicz error-checking PRP test of M503021377 using AVX FFT length 28M, Pass1=1792, Pass2=16K, clm=4, 12 threads
[Dec 27 11:03:25] PRP proof using power=7x2 and 64-bit hash size.
[Dec 27 11:03:25] Proof requires 8.0GB of temporary disk space and uploading a 1006MB proof file.
[Dec 27 11:03:25] Iteration: 502746596 / 503021377 [99.945%].
[Dec 27 11:06:12] Iteration: 502750000 / 503021377 [99.946%], ms/iter: 48.833, ETA: 03:40:52

results.txt contains

[Tue Dec 27 07:58:41 2022]
MD5 error reading PRP proof interim residues file.
[Tue Dec 27 08:58:27 2022]
MD5 error reading PRP proof interim residues file.

I would really like to salvage these long runs and proofs if possible. Residue files may have been accidentally split, with early residues landing locally, later remotely, and some later ones lost due to emergency memory filling up. Not a memory issue; ECC ram, 20GB free.

Highest exponent PRP I've completed so far in prime95 was https://www.mersenne.org/report_expo...2643801&full=1 (Xeon Phi 7210 AVX512, local temporaries file). This is not quite as high, but I think the highest I've attempted on AVX (dual xeon e5-2697v2, one xeon per worker; supposedly could go to 595.xM given sufficient time).

[kriesel]

The 543M finally completed oddly, after several app restarts; its results.json.txt record includes a null md5:

"errors":{"gerbicz":0}, "proof":{"version":2, "power":7, "power-multiplier":2, "hashsize":64, "md5":""}

There are no .bu* files to attempt it again from. But there are save files from nearly every 10M iterations, including 540M ~99.33%.