mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   Prime95 v30.3 (https://www.mersenneforum.org/showthread.php?t=25823)

kriesel 2020-10-02 19:47

[QUOTE=kladner;558642]Thanks for the note on individual worker memory, James.[/QUOTE]I'll second that!
Most of my systems did not have a worker section in local.txt.
It might look something like

[Worker #1]
Memory=32768

[Worker #2]
Memory=32768 during 7:30-23:30 else 32768

storm5510 2020-10-03 17:58

I have found that running [I]mprime[/I] with [I]Ubuntu on Windows[/I] results in sometimes unexpected behavior, such as results files being deleted after results are submitted or having attributes which makes them invisible. After tracking down the folders' full path, I pinned the location in Windows File Manager. This is most likely the source of the issues. Having its own GUI file manager would help greatly. I do not think such an animal exists. I find [I]mprime[/I] preferable to [I]Prime95[/I]. I have my reasons, long discussed and lambasted over.

Xyzzy 2020-10-03 18:22

You can navigate to the Windows filesystem.

Here is an example from our machine:[CODE]m@i9-10900KF:~$ pwd
/home/m

m@i9-10900KF:~$ cd /mnt/c/Users/m/Desktop/

m@i9-10900KF:/mnt/c/Users/m/Desktop$ ls -l
total 44
-rwxrwxrwx 1 m m 6214 Oct 3 13:16 250w.txt
-rwxrwxrwx 1 m m 6209 Oct 3 13:16 25w.txt
drwxrwxrwx 1 m m 4096 Oct 3 12:41 benchmark
-rwxrwxrwx 1 m m 282 Sep 24 08:32 desktop.ini
-rwxrwxrwx 1 m m 27136 Oct 2 21:51 i9.png
drwxrwxrwx 1 m m 4096 Oct 3 13:20 mfaktc
drwxrwxrwx 1 m m 4096 Oct 3 13:17 prime95[/CODE]You can then use the Windows file manager for your Linux files.

storm5510 2020-10-03 18:58

There is another solution. A dual-boot system using the existing [I]Windows 10[/I] setup and [I]Ubuntu 20.04 LTS[/I]. My SSD is 250GB. [I]Windows 10[/I] is using about 65 GB of that space. I found a walk-thru article on the web. I shrunk the Windows volume by 70 GB. This should be enough. It says to leave the shrink space [U]unallocated[/U]. I have the ISO I download from the Ubuntu web site. All I need to do now is to burn the ISO onto a DVD. The Ubuntu installation prompts are different than they would be on a totally empty drive, according to the article. My HP [I]Windows 7[/I] system is about six feet away. I can load the article there and follow it as I go along.

Aramis Wyler 2020-10-03 19:49

If you were going to dual boot just for the sake of running mprime or the like, you could boot off a usb drive.

storm5510 2020-10-03 22:35

[QUOTE=Aramis Wyler;558786]If you were going to dual boot just for the sake of running mprime or the like, you could boot off a usb drive.[/QUOTE]

This is very true. At most, it would be [I]mprime, mfaktc,[/I] and [I]gpuowl[/I]. I am not sure how I would pick which drive, unless one of the other presented a menu to choose from. Having to reconfigure the BIOS ever time I wanted to switch is not something I would like to to.

To be honest, I am not crazy about messing with the SSD. I have an extra 500 GB spinner in a box. I am leaning toward using it.

Aramis Wyler 2020-10-04 00:39

Last I checked a 64gb USB drive was about $18 at rite-aid, so you could use the usb drive not only to hold the OS and the programs but the result, too. And just plug it into the windows OS to upload it.

kriesel 2020-10-04 01:13

[QUOTE=Aramis Wyler;558813]Last I checked a 64gb USB drive was about $18 at rite-aid, so you could use the usb drive not only to hold the OS and the programs but the result, too. And just plug it into the windows OS to upload it.[/QUOTE]What file system do you use to run linux that is readable by Windows in that scenario?

storm5510 2020-10-04 01:37

[QUOTE=kriesel;558818]What file system do you use to run linux that is readable by Windows in that scenario?[/QUOTE]

I had to shrink my Windows volume on the SSD. I left 80 GB for the Ubuntu Installation. This was the [U]only[/U] way I could get it to install. I had no luck with a USB drive. It presented so many partition options that I could not tell one from another. With the dual-boot from the SSD, it presents a menu which I could choose from.

The Windows partition is not visible in Ubuntu. I have not tried it the other way to see if Windows can read the Ubuntu partition. I sort of doubt that it will.

I did not have any problems getting things set up and running. The GUI portion is somewhat different than what I used before in small areas. The composition is the same.

Just in case anyone here thinks this is off-topic, this is very relevant to [I]mprime[/I].

Xyzzy 2020-10-04 02:32

Your Windows partition is NTFS. There are experimental drivers that will read/write NTFS but we don't trust them.

Create a USB stick with either a FAT32 or ExFAT partition. Both Linux and Windows can read/write that format. Then store your mprime installation on that USB stick. Both Linux and Windows will auto-mount the key if you want them to.

There are many ways to do what you want to do. A shared USB stick might be the easiest.

:mike:

storm5510 2020-10-04 14:37

[QUOTE=Xyzzy;558820]Your Windows partition is NTFS. There are experimental drivers that will read/write NTFS but we don't trust them.

Create a USB stick with either a FAT32 or ExFAT partition. Both Linux and Windows can read/write that format. Then store your mprime installation on that USB stick. Both Linux and Windows will auto-mount the key if you want them to.

There are many ways to do what you want to do. A shared USB stick might be the easiest.

:mike:[/QUOTE]

I have two USB sticks which are FAT32. Both can read them. This is how I shuttle things from one side to the other. As I figured, the Window side cannot see the Linux side. No problem though.

storm5510 2020-10-04 16:56

[QUOTE=Xyzzy;558774]You can navigate to the Windows filesystem.

Here is an example from our machine:[CODE]m@i9-10900KF:~$ pwd
/home/m

m@i9-10900KF:~$ cd /mnt/c/Users/m/Desktop/

m@i9-10900KF:/mnt/c/Users/m/Desktop$ ls -l
total 44
-rwxrwxrwx 1 m m 6214 Oct 3 13:16 250w.txt
-rwxrwxrwx 1 m m 6209 Oct 3 13:16 25w.txt
drwxrwxrwx 1 m m 4096 Oct 3 12:41 benchmark
-rwxrwxrwx 1 m m 282 Sep 24 08:32 desktop.ini
-rwxrwxrwx 1 m m 27136 Oct 2 21:51 i9.png
drwxrwxrwx 1 m m 4096 Oct 3 13:20 mfaktc
drwxrwxrwx 1 m m 4096 Oct 3 13:17 prime95[/CODE]You can then use the Windows file manager for your Linux files.[/QUOTE]

In the [I]Windows 10 File Explorer[/I], there is a search bar above the file and folder names. Enter this: [B]\\wsl$. [/B]Without the period, of course. It will show only the Ubuntu folder in the view area. It is listed as a network connection, for some reason. Double-click the icon and the entire file/folder structure is shown. Nothing appears to be hidden using this.

chalsall 2020-10-04 17:38

[QUOTE=storm5510;558866]It is listed as a network connection, for some reason.[/QUOTE]

This might be using some kind of a loopback device trick to do this.

Tad Mar 2020-10-04 18:48

Upgrading to 30.3
 
I am going to do only LL DCs from now on but should I still upgrade? Also, will LL DCs get replaced by the PRP CERT in the future or is this just for the first time checks?

Prime95 2020-10-04 20:41

[QUOTE=Tad Mar;558876]I am going to do only LL DCs from now on but should I still upgrade? Also, will LL DCs get replaced by the PRP CERT in the future or is this just for the first time checks?[/QUOTE]

For LL DC, there is little reason to upgrade, but no harm.

For DC in general, either an LL DC or a PRP w/CERT will do the job (barring a hardware error in one of the LL tests). Should your computer start exhibiting hardware errors, I'd definitely switch to doing DC using PRP w/CERT.

James Heinrich 2020-10-04 20:51

[QUOTE=Prime95;558884]For DC in general, either an LL DC or a PRP w/CERT will do the job (barring a hardware error in one of the LL tests).[/QUOTE]If the user does a LL-DC and the it doesn't match the first-LL, we don't know which is wrong necessitating an (equally-long) triple-check. If the exponent is cleared by PRP+cert then we can either assume the original LL was correct (which may or may not be the case, but ultimately doesn't matter if we know it's not-prime), or we'll immediately know the PRP was faulty if it fails the cert (has this happened?) What's the current PRP+cert failture rate based on a month or so of data?

Prime95 2020-10-05 00:46

[QUOTE=James Heinrich;558885]we'll immediately know the PRP was faulty if it fails the cert (has this happened?)[/QUOTE]

3 gpuowl proofs failed, the DC showed the PRP run was OK, the problem is in proof generation which Mihai thinks he's got a handle on.

1 prime95 PRP-CF proof failed, which I think was due to errors reading residues during proof generation. This has been hardened with MD5 checks on each residue.

This is out of 30000 or so proofs.

kriesel 2020-10-05 01:39

So, ~0.013% PRP proof error rate (with big error bars on that ratio), versus ~1.5% error rate for LL tests.

Prime95 2020-10-05 04:37

[QUOTE=kriesel;558903]So, ~0.013% PRP proof error rate.[/QUOTE]

75% due to programmer error (probably).

Once we feel we have all the bugs hammered out, you really want to know the error rate where a proof can not be generated (residue file corrupted, FFT error during proof generation).

kriesel 2020-10-05 16:21

[QUOTE=Prime95;558915]75% due to programmer error (probably).

Once we feel we have all the bugs hammered out, you really want to know the error rate where a proof can not be generated (residue file corrupted, FFT error during proof generation).[/QUOTE]Or where a proof can not be confirmed, due to MD5 mismatch or whatever. It's natural to have growing pains, and the early result of better than 100:1 lower error rate is already stellar. There may be ways to allow fallback to lower power proof in some bad residue file cases, rather than pitch the whole set and start over. A gpuowl example: [URL]https://www.mersenneforum.org/showpost.php?p=556391&postcount=27[/URL] (Not sure if prime95 already has this covered)

storm5510 2020-10-05 17:16

[QUOTE=James Heinrich;558885]If the user does a LL-DC and the it doesn't match the first-LL, we don't know which is wrong necessitating an (equally-long) triple-check. If the exponent is cleared by PRP+cert then we can either assume the original LL was correct (which may or may not be the case, but ultimately doesn't matter if we know it's not-prime), or we'll immediately know the PRP was faulty if it fails the cert (has this happened?) What's the current PRP+cert failture rate based on a month or so of data?[/QUOTE]

I put Ubuntu [I]20.04 LTS[/I] on my snail Core2 desktop only to run [I]mprime[/I] doing DC work. About 18 days per test. I am wondering if it might be better to run these as PRP? I have a decent idea as to how to change the worktodo lines. Doing so may void my AID numbers if I leave [C]ManualComm=0[/C] I have it set to [C]DaysOfWork=0[/C] to prevent any auto reservations. Contacting the server would need to remain automatic so the certifications would be uploaded. I am not sure how all this would work.

paulunderwood 2020-10-05 17:30

You could cancel your current Ubuntu LL assignments through mersenne.org and request fresh PRP work.

James Heinrich 2020-10-05 17:44

[QUOTE=paulunderwood;558966]You could cancel your current Ubuntu LL assignments through mersenne.org and request fresh PRP work.[/QUOTE]That would be the best option.

storm5510 2020-10-05 23:15

[B][U]Off-topic[/U][/B]. Consider this machine as going to the dumpster, minus the HD. I found another workstation on eBay similar to the one I have now. 2.8GHz Xeon quad-core and 8 GB of RAM. No HD, no problem. I can put my now boxed GTX 1650 in this and go. Holding the two LL jobs I have for a week should not hurt anything. :smile:

kriesel 2020-10-06 00:34

[QUOTE=storm5510;558988][B][U]Off-topic[/U][/B]. Consider this machine as going to the [STRIKE]dumpster[/STRIKE][/QUOTE]Best Buy recycling, or local hackerspace parts bin.

Tad Mar 2020-10-06 19:09

Will DCs be replaced by the PRP CERT anytime soon or will it just replace the first time LL

James Heinrich 2020-10-06 19:58

[QUOTE=Tad Mar;559067]Will DCs be replaced by the PRP CERT anytime soon or will it just replace the first time LL[/QUOTE]Yes (it has been discussed (again) over the last few pages of this thread if you want to skim over it). Old clients that haven't been upgraded yet will continue to do DC-LL since they don't know anything else, new clients can (slightly) more efficiently clear exponents that need DC with a PRP+cert test.

richs 2020-10-06 20:45

[QUOTE=James Heinrich;559072]Yes (it has been discussed (again) over the last few pages of this thread if you want to skim over it). Old clients that haven't been upgraded yet will continue to do DC-LL since they don't know anything else, new clients can (slightly) more efficiently clear exponents that need DC with a PRP+cert test.[/QUOTE]

Is DC with a PRP+cert one of the options under Worker Windows, or does this have to be handled manually?

James Heinrich 2020-10-06 21:19

[QUOTE=richs;559080]Is DC with a PRP+cert one of the options under Worker Windows, or does this have to be handled manually?[/QUOTE]The worker option (in v30.3) is labelled "Double-check prime tests" without the method. Presumably (but I don't know for certain) the server would hand such work out based on the client requesting (LL to old clients, PRP to new clients).

Prime95 2020-10-07 00:19

[QUOTE=James Heinrich;559083]The worker option (in v30.3) is labelled "Double-check prime tests" without the method. Presumably (but I don't know for certain) the server would hand such work out based on the client requesting (LL to old clients, PRP to new clients).[/QUOTE]

I think the server is handing out LL to both. Correct me if I'm remembering incorrectly.

Uncwilly 2020-10-07 00:24

I have a couple of machines that I converted to v30 that were doing LL-DC before. They are still getting them and certs.

James Heinrich 2020-10-07 00:31

[QUOTE=Prime95;559103]I think the server is handing out LL to both. Correct me if I'm remembering incorrectly.[/QUOTE]Beats me, it's your server... :lol:

However, if it's currently handing out DC-LL to everyone I would strongly suggest that it [i]should[/i] hand out PRP assignments instead of LL to v30+ clients requesting "double-check" work -- the Prime95/mprime menu explicitly does [i]not[/i] specify a worktype for the double-check, just "Double-check prime tests". Obviously pre-v30 clients requesting DC would still get DC-LL assignments.

storm5510 2020-10-07 14:30

[QUOTE=Prime95;559103]I think the server is handing out LL to both. Correct me if I'm remembering incorrectly.[/QUOTE]

The manual reservation page is doing both LL and PRP. First-time and double check. In [I]Prime95[/I], the descriptions are more generic, as you well know. With consideration to the large back-log of LL-DC, it may be better to leave the manual assignment page as it. First time LL could go though.

:two cents:

kriesel 2020-10-09 20:16

Invisible Prime95 V30.3b6
 
On one system, a situation developed where prime95 when restored to screen at less than maximized was not visible. (Windows Vista Business 32 bit OS on a Compaq NC6230)

Normally prime.txt will contain screen locations consisting of modest sized positive integer descriptions such as the following.

Left=685
Top=4
Right=1581
Bottom=790

While this issue was occurring, the prime.txt on disk looked normal.
After closing prime95, prime95.txt contained anomalous values that were all approximately -32000. In screen coordinates, 1,1 being the upper left and maxx,maxy being the lower right, IIRC, the values ~-32000 corresponded to approximately rooftop level in the back yard, for a laptop in the basement.

Editing those lines in prime.txt with a text editor back to positive values no larger than the screen width set things right. I have no idea how prime95 came up with the ~ -32000 values in volatile memory.

jwnutter 2020-10-09 20:21

[QUOTE=kriesel;559356](Windows Vista Business 32 bit OS on a Compaq NC6230[/QUOTE]

I found that I was most amazed by the word "Compaq". :grin:

kriesel 2020-10-09 21:37

[QUOTE=jwnutter;559357]I found that I was most amazed by the word "Compaq". :grin:[/QUOTE]$50 apiece used years ago; 2 of the 3 failed, in such a way that one functional one could be assembled from the wreckage, so 2 survivors still. They make good don't-care-much laptops for travel. And moderately effective space heaters that do a little computing in the process and provide "vintage" test environments. They're actually new compared to some of what I have.

XZT 2020-10-10 02:24

[QUOTE=James Heinrich;559106]Beats me, it's your server... :lol:

However, if it's currently handing out DC-LL to everyone I would strongly suggest that it [I]should[/I] hand out PRP assignments instead of LL to v30+ clients requesting "double-check" work -- the Prime95/mprime menu explicitly does [I]not[/I] specify a worktype for the double-check, just "Double-check prime tests". Obviously pre-v30 clients requesting DC would still get DC-LL assignments.[/QUOTE]

If I've got a machine that I'm very confident in it's reliability (like 100% correct LL-DC so far) and [I]want[/I] to continue LL-DC work, would I still be able to do so easily (i.e. without requiring manual assignments)?

My reasoning for continuing LL-DC is that it seems to be easier to spot flaky machines as they throw a clear LL mismatch and that's useful for verifying the reliability of my machines, and may be of interest data wise for the old first time LLs.

Would there be other ways for a common user to spot errors?
e.g. Are PRP errors recorded? How can they be displayed to the user (without having to manually check logs)?
What happens to the unverified first time LLs?

In another case, if I've got a machine that is known to be unreliable but [I]want[/I] to continue contributing some sort of DC work, the removal of the PRP-DC worktype makes it a bit more difficult to configure (manual assignments).

Understandably, PRP-DC shouldn't be needed going forward with PRP+CERT, so how many non-CERT PRPs remain to DC?
I've also seen some CERT runs where there are already matching PRP(non-CERT) and PRP(CERT) results. That was on PRP-CF work, but I assume this could occur for the main prime search as well. Negligible CPU time, but maybe something to note.

Is the assignment of PRP-CF CERT along with PRP CERT work intentional or is there an option to select which?

Prime95 2020-10-10 03:06

[QUOTE=XZT;559385]If I've got a machine....[/QUOTE]

A lot of questions there. For now, v30 clients that request double-check work will get LL-DC.

When an LL-DC mismatches, you do not know which machine is bad -- yours or the first testers. In that respect, LL-DC is not perfect for your needs. However, there are easy to spot indicators at mersenne.org's web pages that make it a good choice for monitoring a PC's reliability.

Doing PRP work (first time or DC) will be 100% (we think) reliable even on unreliable hardware. Thus, the only way to monitor machine health is to look at the logs. The data is available to the server, so a web page could be created that reports on machine health.


Doing a PRP with proof for a DC of an LL test leaves the original LL test in "limbo". We don't know if it was good or bad. Somewhat untidy and unsatisfying.

As to CERT work, there are some options you can set described in undoc.txt to control which CERTs you might get.

Your unreliable machine should do PRP work of some kind. You might try setting WorkPreference to 151 (no idea if that will work), or manually get PRP-DC assignments,
or get LL-DC and turn them into PRP with proof by editing worktodo.txt. Not very satisfying solutions. Sorry.

ixfd64 2020-10-10 07:38

On some Windows 10 machines, Prime95 now shows a different font in the windows after I upgraded to the latest version. Is this to be expected?

ATH 2020-10-10 11:17

If you get a mismatch in LL DC and you want to know if yours is good, you can always ask if someone wants to do a triple check right away in this thread:
[url]https://mersenneforum.org/showthread.php?t=24148[/url]

endless mike 2020-10-15 20:16

I upgraded all my machines to 30.3b6. On a few of them, they did not remember my userid or their computerid. I caught and corrected the problem before results were submitted except for one. How can I get the exponent that I completed added to my user statistics instead of it saying it was completed by -Anonymous- ?

Uncwilly 2020-10-15 20:30

[QUOTE=endless mike;559986]I upgraded all my machines to 30.3b6. On a few of them, they did not remember my userid or their computerid. I caught and corrected the problem before results were submitted except for one. How can I get the exponent that I completed added to my user statistics instead of it saying it was completed by -Anonymous- ?[/QUOTE]Send George a PM or email. His email address is one of the graphics that are available. [C]:woltman:[/C]

James Heinrich 2020-10-15 20:37

[QUOTE=Uncwilly;559992]Send George a PM or email. His email address is one of the graphics that are available. [C]:woltman:[/C][/QUOTE]It's also in the top of [c]readme.txt[/c] :smile:
Or you can PM him on the forum, username [c]Prime95[/c].

JuanTutors 2020-10-19 15:57

1 Attachment(s)
Not sure if this has been reported yet but this version of Prime95 seems to have the same problem that some apps/programs have on 4k monitors. The text is exceedingly small. I am restarting the computer to see if things get fixed but I have attached what the program looks like right now.

JuanTutors 2020-10-19 16:02

[QUOTE=JuanTutors;560325]Not sure if this has been reported yet but this version of Prime95 seems to have the same problem that some apps/programs have on 4k monitors. The text is exceedingly small. I am restarting the computer to see if things get fixed but I have attached what the program looks like right now.[/QUOTE]

Restarting my computer didn't fix the text size for me.

JuanTutors 2020-10-19 16:21

I found another very minor bug. When I installed the new version of Prime95 into the same old folder that had the previous version of Prime95 (28.6), it got a new PRP assignment despite the fact that I am currently in the last few days of a current 100M digit PRP assignment and already have my next 100M digit PRP assignment on deck. That means I got a PRP assignment despite having a half year's worth of work.

I can only speculate on the exact reason why this could possibly happen. (I pasted the new .exe file but didn't paste the new .dll files? Maybe the sequence in which certain checks can be done caused this to happen?)

kruoli 2020-10-19 16:25

[QUOTE=JuanTutors;560325]Not sure if this has been reported yet but this version of Prime95 seems to have the same problem that some apps/programs have on 4k monitors.[/QUOTE]

It's not the 4K that causes the problem, it's the Windows scaling that does not get recognized by Prime95. Yes, if you have set the magnification exceedingly high, it looks like the text is tiny.

Uncwilly 2020-10-19 16:28

[QUOTE=JuanTutors;560328]I found another very minor bug. When I installed the new version of Prime95 into the same old folder that had the previous version of Prime95 (28.6), it got a new PRP assignment despite the fact that I am[/QUOTE]Are you sure that you did not just get a Cert assignment? That should not last more than an hour or 3, depending on your machine.

JuanTutors 2020-10-19 16:30

[QUOTE=Uncwilly;560332]Are you sure that you did not just get a Cert assignment? That should not last more than an hour or 3, depending on your machine.[/QUOTE]
I don't think so. Here is the assignment:
[CODE]PRP=AID EXPUNGED,1,2,332316059,-1,79,0,3,[/CODE]

kriesel 2020-10-19 16:48

[QUOTE=JuanTutors;560333]I don't think so. Here is the assignment:
[CODE]PRP=xxx,1,2,332316059,-1,79,0,3,[/CODE][/QUOTE]
DON'T post AIDs for current assignments!
Don't make pointless work for moderators to remove what should not be posted.

DrobinsonPE 2020-10-20 04:33

I have a computer running mprime 30.3 b6 that today finished PRP on exponent 100362923. It experienced 5 Gerbicz/double-check errors during the run but always said confidence in final result is excellent so I did not start over. All of the errors happened at about 50% and all in the period of a few hour time. I just checked [url]https://www.mersenne.org/report_exponent/?exp_lo=100362923&full=1[/url] and discovered that the result is suspect, two people tried to do a cert and now it has been assigned to someone else to do a PRP.

I am curious, was the PRP bad, did the cert fail? The computer is currently working on 100495651. Should I stop it if it starts with Gerbicz/double-check errors again or just let it keep going?

I think the problem is the memory because the two sticks were in a different computer that had a bad LL-D result about this time last year. Over the last year, whatever computer this memory is in has errors about once every 3-4 months.

moebius 2020-10-20 05:59

[QUOTE=DrobinsonPE;560381]I am curious, was the PRP bad, did the cert fail? The computer is currently working on 100495651. Should I stop it if it starts with Gerbicz/double-check errors again or just let it keep going? [/QUOTE]
I bet that the PRP test is good, and only the .proof-file you uploaded are bad.

moebius 2020-10-20 15:04

[QUOTE=Prime95;557098]I've made 30.3 build 6 the official download version at mersenne.org![/QUOTE] Does anyone have statistics available, how many users already use a Prime95 / mprime / gpuowl version with proof generation. It seems to have improved a lot in the last month, but there are still many first-time LL tests and PRP tests without proofs, mostly from CPUs which could use a PRP-proof version without any problems.It is best, to be mean like Microsoft and force an update under duress (or even better under torture:deadhorse:) since the software does not have a notification or update option.

Uncwilly 2020-10-20 15:11

Your answer is here (somewhat):

[url]https://www.mersenneforum.org/showthread.php?p=559472[/url]

rainchill 2020-10-21 14:20

Yep - I've brought up auto updates for years for it to only be treated like the worst idea anyone has ever heard. At some point all the old clients out there genuinely are counter productive.

At a minimum there should be a rule that the server does not hand out first time checks to clients older than ~2-3 years.

JuanTutors 2020-10-21 17:50

[QUOTE=rainchill;560538]Yep - I've brought up auto updates for years for it to only be treated like the worst idea anyone has ever heard. At some point all the old clients out there genuinely are counter productive.

At a minimum there should be a rule that the server does not hand out first time checks to clients older than ~2-3 years.[/QUOTE]
I don't know about it being counter-productive, but there are a whole lot of options that seem to not be explored. For example, Primer connects to the internet and downloads information, so why not also allow it to get messages such as "a new update of Prime95 is available", etc?

kriesel 2020-10-21 18:33

An occasional check of current released version versus running version, such as at startup, would probably be ok, as long as it doesn't behave too obnoxiously or interfere with prime95 starting as a service, require user acknowledgement before running all workers, etc.
Something like, if the running version is not the latest on [URL]https://www.mersenne.org/download/[/URL], and the running version has internet access,
[CODE]Current released version is Vxx.y, running version is Vww.v. Please upgrade if/when practical.[/CODE]Of course this would only be effective for versions sometime after V30.3b6, as the older version, since the feature has not been included yet, so would not address the great majority still running V29.x or earlier, until those users can be persuaded to update, by other means.

rainchill 2020-10-21 18:43

You are correct. Moving forward, the sooner some type of upgrade alert feature can be included the better.

storm5510 2020-10-22 01:06

[QUOTE=moebius;560421][COLOR=Gray]Does anyone have statistics available, how many users already use a Prime95 / mprime / gpuowl version with proof generation. It seems to have improved a lot in the last month, but there are still many first-time LL tests and PRP tests without proofs, mostly from CPUs which could use a PRP-proof version without any problems.[/COLOR]It is best, to be mean like Microsoft and force an update under duress (or even better under torture:deadhorse:) since the software does not have a notification or update option.[/QUOTE]

The current flavor of Windows 10 [U]does[/U] have a notification option. It simply has to be configured in the Automatic Update dialogs. I have it set to show a notice when a restart is needed. This way, I can properly shut-down any running processes, then allow it to proceed.

James Heinrich 2020-10-22 01:09

I think the reference was more to the fact that unless you try really hard, Windows no longer lets you [i]not[/i] install updates (maybe not this second, or even today, but sooner or later "you will comply").

kladner 2020-10-23 05:00

[QUOTE=James Heinrich;560631]I think the reference was more to the fact that unless you try really hard, Windows no longer lets you [I]not[/I] install updates (maybe not this second, or even today, but sooner or later "you will comply").[/QUOTE]
I just recently resolve an issue with Updates which failed installation repeatedly, causing one reboot on install, and two reboots on the automatic rollback. I kept Update paused for weeks at a time. In the last day or two I searched more carefully and found a script online which flushed caches and reset things. This resolved the update loop at last.
However, I came to this thread to post a link to a report I first dropped in the Server Problems thread. On reflection, it is probably better off here.
[URL]https://mersenneforum.org/showpost.php?p=560486&postcount=2014[/URL]
WTF. Here's the post:
On one of my Win 10 64 bit machines I get this error if Cert is checked:
Quote:
[Sun Oct 18 15:25:53 2020]
Error getting CERT starting value.
[Sun Oct 18 16:25:53 2020]
Error getting CERT starting value.
[Sun Oct 18 17:25:53 2020]
Error getting CERT starting value.
[Sun Oct 18 18:05:58 2020]
Error getting CERT starting value.
[Sun Oct 18 18:06:42 2020]
Error getting CERT starting value.
I removed Cert from worktodo and unchecked it in Worker Windows to stop this. This is an i7 6700k with 32 GiB of Ram. Asus 170 board. A newer machine, i7 9700k, 32 GiB of RAM has no problems.
On both machines I upgraded to V30.3 b6 by copying the unzipped files to the existing directory.

JuanTutors 2020-10-23 19:43

[QUOTE=JuanTutors;560325]Not sure if this has been reported yet but this version of Prime95 seems to have the same problem that some apps/programs have on 4k monitors. The text is exceedingly small. I am restarting the computer to see if things get fixed but I have attached what the program looks like right now.[/QUOTE]

I noticed this isn't listed in the bugs list yet. It's not obvious from the picture, but the letters in 30.3 are actually smaller than is legible. Is there any plans to fix this?

Uncwilly 2020-10-23 20:07

[QUOTE=JuanTutors;560891]I noticed this isn't listed in the bugs list yet. It's not obvious from the picture, but the letters in 30.3 are actually smaller than is legible. Is there any plans to fix this?[/QUOTE]They are fine on my different displays. I have a desktop of 3000x3000 on one machine.

Prime95 2020-10-23 23:36

[QUOTE=JuanTutors;560891]I noticed this isn't listed in the bugs list yet. It's not obvious from the picture, but the letters in 30.3 are actually smaller than is legible. Is there any plans to fix this?[/QUOTE]

I wouldn't know how.

ixfd64 2020-10-23 23:56

I'm guessing this isn't intentional?

[QUOTE=ixfd64;559406]On some Windows 10 machines, Prime95 now shows a different font in the windows after I upgraded to the latest version. Is this to be expected?[/QUOTE]

storm5510 2020-10-24 00:35

[QUOTE=JuanTutors;560891]I noticed this isn't listed in the bugs list yet. It's not obvious from the picture, but the letters in 30.3 are actually smaller than is legible. Is there any plans to fix this?[/QUOTE]

i noticed this when I was updated to Windows 10 v2004. I set the default screen magnification to 125%. Doing so made the text larger. I have two older Windows 7 machines with lower resolution monitors. 30.3 looks alright on them. Just because you can run a really high resolution does not mean you must. Lower it until things are easy to read.

JuanTutors 2020-10-24 01:32

[QUOTE=Prime95;560908]I wouldn't know how.[/QUOTE]

I wouldn't either, but I was running 29.8v6 until a few days ago and the text size on that one was fine, with no changes to the computer in between the upgrade. Was there something you changed between that session and this one? Some size measurement difference? em/rem/px?

LaurV 2020-10-24 01:57

[QUOTE=JuanTutors;560891]I noticed this isn't listed in the bugs list yet. It's not obvious from the picture, but the letters in 30.3 are actually smaller than is legible. Is there any plans to fix this?[/QUOTE]
Kinda offtopic, some years ago we developed a PC interface for a product we designed for a Belgian customer and somewhere in the middle of the developing cycle (which took over one year, and had a lot of loops and ping-pongs) turned out he had a 4k laptop, and the app screen, which was about 1000x600 in size (to be displayable on 768 lines of most laptop's screens which were 1366x768 at the time, without overlapping the windoze taskbar, and was still looking more than reasonable on 1920x1080 monitors we use at work), [U]was extremely small and unreadable on his monitor[/U]. The biggest letters looked like small fleas, and the smallest looked like pixels :lol:

The application was in Java (not our choice!), it wasn't designed to be scaled, the window was fixed-size, with lots of icons (bitmap style), and it turned out that scalable fonts are out of question in Java, being "close to impossible" to implement. Java doesn't offer a full TTF handling, as true-type fonts are somehow "unsecure" due to pointers and executable code involved to handle the drawing process. Which we didn't know at the time. (you can google "scalable fonts in Java").

We ended up adding a new config parameter that offered 3 possible resolutions, "small", "medium" and "large" to accommodate all monitors, and of course, we had to re-map all fonts and re-design (re-paint) all the icons and bitmaps not once, but twice, for the two new resolutions. Which was a pain in the butt, and it took another half year.

So, sometimes simple tasks are not so simple..

But maybe that would be a solution for P95 interface too, to avoid a scalable interface.

OTOH, in msvc, implementing scalable fonts is not so difficult... Charles Petzold has a couple of examples in his bible book.

Edit: I may put my nose into it in the future, I intended to do it in the past, as the issue with windows arrangement still bothers me... but the RL[SUP](TM)[/SUP] stuff didn't allow me, yet...

JuanTutors 2020-10-24 03:35

Not quoting LaurV for size but based on his post, if this is not a problem for all upgrades to 30.3 on 4k computers, a solution as easy as a menu option like "double the font size" can do it. I would bet there is a minor change in there somewhere that is causing the weird font difference. I have a lot of programs that have random things the wrong size, but not everything. Sometimes it's menu options, sometimes it's icons, sometimes text. Makes me suspect that sizes are defined differently in different situations within the program.

Prime95 2020-10-24 07:44

[QUOTE=JuanTutors;560924]I wouldn't either, but I was running 29.8v6 until a few days ago and the text size on that one was fine, with no changes to the computer in between the upgrade. Was there something you changed between that session and this one? Some size measurement difference? em/rem/px?[/QUOTE]

The source code did not change. 29.6 was built with MSVC 2005. 30.3 was built with Visual Studio 2019.

storm5510 2020-10-24 17:34

When I got up this morning, I found that my Xeon system had rebooted itself. I was running [I]Prime95 v30.3[/I] using four workers on large stage-one ECM's. Having some I/T experience, I went to the event logs. One item in the system log was listed as "Critical Error." The associated message said "the previous restart was unexpected." There was nothing in the application log. Also, nothing in [I]Prime95's[/I] log with debug set to one.

So, I am trying to recreate this with 30.3, same configuration, and four workers, but [U]not[/U] on the same machine. This one is an i5. Both machines run [I]Windows 7 Pro x64, SP1[/I] and have the same amount of RAM. If the i5 reboots itself, then I would lean towards it being a software problem. If it does [U]not[/U], then I would suspect a hardware problem.

In the interim, I restarted [I]Prime95[/I] on the Xeon with only two workers. I am trying to keep GMP-ECM fed which is also running on the Xeon, but was not running when the restart happened. So far, so good. All I can do now is wait and see what happens, or does not happen.

ckdo 2020-10-26 02:06

I upgraded Prime95 from 29.3 to 30.3b6 on a machine I do not regularly monitor on October 7th.

Ever since, whenever I took a look at the machine I found Prime95 was no longer running, leaving behind a bunch of ".write" files, indicating a crash when writing its state to disk. 29.3 always ran fine on that machine.

By pure chance I witnessed one of those "crashes" today.

It turns out that Bitdefender Total Security (which also received an update recently) is actually killing Prime95 because the latter is "encrypting my files", namely all the ".txt" files it happens to use.

A message along these lines is displayed on screen, but will close automatically after a while, leaving you puzzled unless you look into Bitdefender's event log.

I added Prime95 to Bitdefender's ransomware detection exception list. Prepare for the same to happen to more users...



While I'm here, I found the following in "prime.log":

[code]
[Wed Oct 7 18:38:50 2020 - ver 30.3]
CURL library error: Failed to connect to v5.mersenne.org port 80: Bad access
...
[Wed Oct 14 15:10:25 2020 - ver 30.3]
CURL library error: Failed to connect to www.mersenne.org port 80: Bad access
[/code]That's also new with 30.3. There's only these two instances so far.

JuanTutors 2020-10-26 03:06

[QUOTE=Prime95;560947]The source code did not change. 29.6 was built with MSVC 2005. 30.3 was built with Visual Studio 2019.[/QUOTE]

Is the following link (about DPIUNAWARE) the solution? If you would like to try it on my computer I don't mind testing it out.

[URL="https://docs.microsoft.com/en-us/visualstudio/designers/disable-dpi-awareness?view=vs-2019"]https://docs.microsoft.com/en-us/visualstudio/designers/disable-dpi-awareness?view=vs-2019[/URL]

kruoli 2020-10-26 11:45

[QUOTE=Prime95;560908]I wouldn't know how.[/QUOTE]

[URL="https://docs.microsoft.com/en-us/windows/win32/api/shellscalingapi/nf-shellscalingapi-getscalefactorformonitor"]GetScaleFactorForMonitor[/URL] and increase the font size accordingly?

JuanTutors 2020-10-28 05:19

[QUOTE=Uncwilly;560332]Are you sure that you did not just get a Cert assignment? That should not last more than an hour or 3, depending on your machine.[/QUOTE]

This bug repeated itself on a different computer. I downloaded v30.3b6 on a separate computer, directly from Mersenne.org (not from a flash drive). when I restarted P95, it gave me a new PRP assignment.

[code]PRP=XXX,1,2,332316437,-1,79,2[/code]

bayanne 2020-10-30 13:37

Any progress in making version 30.* build * available for Mac OSX yet?

Prime95 2020-10-30 19:53

[QUOTE=bayanne;561546]Any progress in making version 30.* build * available for Mac OSX yet?[/QUOTE]

There was a link to a command line version that some have reported worked OK. Try [url]https://www.dropbox.com/s/xo7psmmjfkb52pr/p95v303b6.MacOSX.tar.gz?dl=0[/url]

simon389 2020-10-31 19:14

Since v30.3 my computer bank is completing only 75% of the PRPs that it did in v29.6. They spend a large amount of time doing other peoples CERT work. Is there a setting to ask the server for just PRP and skip all these CERTs for other peoples PRPs?

James Heinrich 2020-10-31 19:19

[QUOTE=simon389;561706]Is there a setting to ask the server for just PRP and skip all these CERTs for other peoples PRPs?[/QUOTE][c]Test | Worker Windows...[/c] uncheck: [c]Get occasional proof certification work[/c]

Alternately, set [c]CertWork=0[/c] in [i]local.txt[/i]

You can also see under [c]Options | Resource Limits... | Advanced...[/c] the setting [c]Certification work limit (% of CPU time)[/c] if you want to do some Cert work but less than you're currently getting. With default settings your CPU shouldn't spend more than 10% of it's time doing Cert.

storm5510 2020-11-01 18:10

Advanced Resource Limits & Settings Dialog
 
1 Attachment(s)
Please consider the attached image. This seems to be a Windows 7 Pro issue [U]only[/U]. Clicking the "OK" button in the upper right corner produces the inset message. Clicking "OK" on this message highlights the value in the dialog beneath in blue. I have this shaded in bright yellow. The only way out is to click "Cancel" on the larger dialog. I have learned where most of the setting involved are in [I]prime.txt[/I] and [I]local.txt[/I], so I adjust them manually.

Viliam Furik 2020-11-01 18:20

[QUOTE=storm5510;561843]Please consider the attached image. This seems to be a Windows 7 Pro issue [U]only[/U]. Clicking the "OK" button in the upper right corner produces the inset message. Clicking "OK" on this message highlights the value in the dialog beneath in blue. I have this shaded in bright yellow. The only way out is to click "Cancel" on the larger dialog. I have learned where most of the setting involved are in [I]prime.txt[/I] and [I]local.txt[/I], so I adjust them manually.[/QUOTE]

I have the issue on Windows 10, too. It seems to be caused by the program limiting the size of the "Emergency memory" to at most quarter of the total RAM available but has some rounding issues (or issues when getting the size of installed RAM) so it won't allow exactly 2 GB when having 8GB total, but only 1.99951 GB. Or in my case, 7.97949 GB out of 32 GB, instead of 8GB.

PhilF 2020-11-01 18:23

I'll bet $1.99 that if you typed 1.99 over that 2 that the program would take it.

Viliam Furik 2020-11-01 23:28

[QUOTE=PhilF;561846]I'll bet $1.99 that if you typed 1.99 over that 2 that the program would take it.[/QUOTE]

I have tested it, you're absolutely right. If you put in a number which is in the range required by the error message, it will accept it, but it will round it to one decimal place.

In my case, 8 is bad, as well as 7.9999, 7.999, 7.99. 7.98 is still bad, but 7.97 is good, and it gets rounded to 8 when I apply the setting and open it again. 7.95 gets rounded to 7.9.

Prime95 2020-11-01 23:48

I'll also round the maximums to the nearest tenth.

storm5510 2020-11-02 18:30

Windows 10 did not give me any problems. I just checked. It is set to 1. I have never tried any fractional values, like 1.5. In this case, I would have gone into [I]local.txt[/I] and set the value to 1536. For some reason, I always use powers of 2 for things of this type, although I do not know why. Regardless, I am pleased a few here paid some attention to this. :smile:

Viliam Furik 2020-11-03 02:06

[QUOTE=storm5510;561973]Windows 10 did not give me any problems. I just checked. It is set to 1. I have never tried any fractional values, like 1.5. In this case, I would have gone into [I]local.txt[/I] and set the value to 1536. For some reason, I always use powers of 2 for things of this type, although I do not know why. Regardless, I am pleased a few here paid some attention to this. :smile:[/QUOTE]

If you try to set the value to [I]half[/I] of the total installed RAM, I think you will get the error again.

Runtime Error 2020-11-03 13:20

I'm curious what happened with [M]100574899[/M]. It seems that Mihai Preda did PRP, and then curtisc & George Woltman both did CERTs, but then I still got it as a first-time check. Was there something suspicious with Preda's run? Thanks.

Prime95 2020-11-03 20:01

[QUOTE=Runtime Error;562062]I'm curious what happened with [M]100574899[/M]. It seems that Mihai Preda did PRP, and then curtisc & George Woltman both did CERTs, but then I still got it as a first-time check. Was there something suspicious with Preda's run? Thanks.[/QUOTE]

Gpuowl 7.1 had a bug in proof generation. The two Certs failed, the result was marked suspicious and the exponent was reassigned as a first time check.

Runtime Error 2020-11-04 02:59

[QUOTE=Prime95;562120]Gpuowl 7.1 had a bug in proof generation. The two Certs failed, the result was marked suspicious and the exponent was reassigned as a first time check.[/QUOTE]

Interesting, & thanks for the reply - I appreciate it!

James Heinrich 2020-11-27 15:01

Just a minor oddity I noticed:[quote]Mersenne number primality test program version 30.3
Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 6x1 MB, L3 cache size: [b][COLOR="Red"]6x[/COLOR][/b]19712 KB[/quote]The 6x1MB L2 is correct, but there's only a single 19.25MB L3 cache. It is a VM though so perhaps that's confusing it (I don't have any other system running mprime to compare output).

kriesel 2020-11-27 20:24

V30.3b6 crash instead of completing exponent
 
[CODE][Nov 27 14:15] Worker starting
[Nov 27 14:15] Setting affinity to run worker on CPU core #1
[Nov 27 14:15] Setting affinity to run helper thread 1 on CPU core #2
[Nov 27 14:15] Resuming Gerbicz error-checking PRP test of M182000029 using FMA3 FFT length 10M, Pass1=1280, Pass2=8K, clm=2, 2 threads
[Nov 27 14:15] PRP proof using power=7x2 and 64-bit hash size.
[Nov 27 14:15] Proof requires 2.9GB of temporary disk space and uploading a 364MB proof file.
[Nov 27 14:15] Iteration: 181998738 / 182000128 [99.99%].
[Nov 27 14:16] Iteration: 182000000 / 182000128 [99.99%], ms/iter: 29.771, ETA: 00:00:03[/CODE]It will crash during producing the proof. I can't capture the worker window at the crash because the app disappears from the screen.
This was repeatable, as results.txt excerpt shows.
[CODE][Sun Nov 22 03:24:00 2020]
Error reading PRP proof interim residues file.
Errno: 13, Permission denied
DOSerrno: 23
[Fri Nov 27 14:06:22 2020]
Error reading PRP proof interim residues file.
Errno: 13, Permission denied
DOSerrno: 23
[Fri Nov 27 14:14:56 2020]
Error reading PRP proof interim residues file.
Errno: 13, Permission denied
DOSerrno: 23[/CODE]So wouldn't you know it, I go to document it, and the fourth time it succeeds and reports a result. Yay. I didn't change anything, except to stop prime95 and continue, to avoid running the last 0:03:35 over and over. The drive had over 300GB of free space at the time. System sat idle for over 5 days as a result of the first crash in the series.

PhilF 2020-11-27 22:03

DOSerrno 23 is "Too many file open in system". I guess when you restarted Prime95 it closed some files.

It's hard to speculate at this point whether your system ran out of file handles overall, or if Prime95 is possibly leaking open files over time.

kriesel 2020-11-27 22:18

[QUOTE=PhilF;564617]DOSerrno 23 is "Too many file open in system". I guess when you restarted Prime95 it closed some files.

It's hard to speculate at this point whether your system ran out of file handles overall, or if Prime95 is possibly leaking open files over time.[/QUOTE]I restarted it multiple times after multiple crashes, and only the fourth try worked.
System has not yet been up for 16 days.
[URL="https://www.stanislavs.org/helppc/dos_error_codes.html"]This[/URL] says 23 is "FCB unavailable".
[URL="https://docs.microsoft.com/en-us/windows/win32/debug/system-error-codes--0-499-"]Others [/URL]say CRC error.
04 is too many files open.

PhilF 2020-11-27 22:23

In Visual Studio, which I think George builds with, it shows as [URL="https://docs.microsoft.com/en-us/cpp/c-runtime-library/errno-doserrno-sys-errlist-and-sys-nerr?view=msvc-160"]Too many open files in system[/URL].

Not to be confused with doserrno 24, which is "Too many open files", lol.

kriesel 2020-11-28 01:17

[QUOTE=PhilF;564622]In Visual Studio, which I think George builds with, it shows as [URL="https://docs.microsoft.com/en-us/cpp/c-runtime-library/errno-doserrno-sys-errlist-and-sys-nerr?view=msvc-160"]Too many open files in system[/URL].

Not to be confused with doserrno 24, which is "Too many open files", lol.[/QUOTE]
Thanks.
Event viewer is showing a lot of these errors in system log.
The device, \Device\Harddisk0\DR0, has a bad block.
Kinda poor since it's only ~3-4 years old and has led a sheltered life.

Application log had this from a p95 crash
Faulting application name: prime95.exe, version: 30.3.1.0, time stamp: 0x5f5ae7c7
Faulting module name: prime95.exe, version: 30.3.1.0, time stamp: 0x5f5ae7c7
Exception code: 0xc0000409
Fault offset: 0x000000000245e79c
Faulting process id: 0x5d6c
Faulting application start time: 0x01d6c4f8cce1c42d
Faulting application path: C:\Users\ken\Documents\p95\prime95.exe
Faulting module path: C:\Users\ken\Documents\p95\prime95.exe
Report Id: e4c7e7ce-48d7-49e7-b623-3addef909bbc
Faulting package full name:
Faulting package-relative application ID:

PhilF 2020-11-28 01:33

[QUOTE=kriesel;564637]Thanks.
Event viewer is showing a lot of these errors in system log.
The device, \Device\Harddisk0\DR0, has a bad block.[/QUOTE]

Does the timestamp on the bad block event match one of the Prime95 crash times?

EDIT: Never mind, I just noticed you said your log is showing a lot of bad block entries. The good news is that I think you found your problem. The bad news is that I think you found your problem. :max:

Ensigm 2020-11-29 16:50

Benchmarking for P-1
 
I wonder if anyone else is also interested in the P-1 benchmarks I posted [URL="https://www.mersenneforum.org/showpost.php?p=564685&postcount=831"]here[/URL]. I believe this could mean we can potentially extract more P-1 power on some machine configurations (by using different FFT implementations for Stage 1 and Stage 2).


All times are UTC. The time now is 02:49.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.