Prime95 v25.3

Prime95 · 2007-06-17, 01:19

Version 25.3
------------

The best candidates for testing this early version are critical thinkers, owners of multi-core machines, those with odd internet settings such as a proxy server. Version 25.3 might be marginally slower than version 24.14.

This version talks to the v4 PrimeNet server by default. I've been using it for 3 weeks and have successfully sent in a matching double-check and a first-time test.

Steps to install version 25.3
----------------------------
DO NOT INSTALL OVER VERSION 24. If you have problems, you want to be able to revert to testing using version 24.
1) Run version 24, uncheck start at bootup, exit version 24.
2) COPY (not move) your version 24 folder to a new place - say c:\v25test
3) Download ftp://mersenne.org/gimps/p95v253.zip. Unzip into c:\v25test.
4) Run version 25.3. Find problems. Report them to me.

Links to other versions
---------------------

Linux: ftp://mersenne.org/gimps/mprime253.tar.gz
Linux64: ftp://mersenne.org/gimps/mprime253-linux64.tar.gz
Mac OSX: ftp://mersenne.org/gimps/mprime253-MacOSX.tar.gz

How to be most useful
---------------------

I'm not only looking for bugs. I'm looking for wording that is confusing, features that are awkward to use, etc. I'll mention some special areas you can test.

1) Can you contact the v4 server? The comm layer is completely different - it uses libcurl.
2) Run Advanced/Time 9900 or Advanced/Time 9901. After hours or days look in results.txt to see if the words "Test failed" appear. If so, you have found a problem in the underlying FFT code. Dual and quad core users are especially encouraged to do some of this!
3) Re-run a past P-1 test that found a factor. There have been some changes to the P-1 code.
4) Re-run a trial factoring run that found a factor. The TF code has changed too.
5) Does Test/Status produce reasonable estimates? Does your RollingAverage get out of whack over time?
6) Is your CPU, L2 cache, hyperthreading, and num cores detected properly.
7) Try running PRP tests against a wide range of known primes from Chris Caldwell's database. There is a new work type in worktodo.txt: "PRP=k,b,n,c"

Register your UI suggestions
----------------------------

1) For example, I'm unhappy with Test / Worker threads dialog box:
a) I'm not sure the average user knows what a "thread" is. Should I instead call them "worker Windows" and if so, what do I call them in the Linux version? And is "worker thread" a misnomer if the worker is actually using multiple threads in the FFT?
b) The CPU affinity setting "Smart assignment" is meaningless. Maybe I should call this "do what makes the most sense"??? It really means if you have as many worker windows as CPUs then assign each worker window to its own CPU - otherwise run on any cpu.
c) The text "Number of CPUs to use" is vague/wrong. It really is how many threads should the FFT code use. On a hyperthreaded machine, a setting of 2 really means use two logical cpus on one physical cpu.
2) Scott, has suggested a one window option. Instead of "Main thread", "Comm thread", Worker 1", they would all appear in one window (kinda like v24) with text next to the timestamp that says what thread generated the message. Do you have better or alternative suggestions?
3) Are the screen outputs well worded? Do the Window and program titles always make sense?

Are there any volunteers out there that want to muck with MFC code? Saving the coordinates and sizes of MDI windows on exit would be nice (and restore on startup). The vertical scroll bars have a devil of a time staying at the bottom.

jasong · 2007-06-18, 00:05

In answer to the concerns Prime95 has with, um, the new Prime95, concerning the wording, I think we should make it "more" confusing. Use mnemonics that only make sense, but make perfect sense, if you read the help file.

Of course, I'm guilty of clicking the "Accept" button without reading online contracts, so it's probably best for me to keep my mouth shut.

Salz · 2007-06-18, 14:59

I don't like the term "threads" either because it is an implementation detail of the running program. And if you use multiple threads for the TTF it is plain wrong.

A much more better term would probably be "Job". So you could e.g. have a "Main Window", "Prime-Net Job 1", "Prime-Net Job 2" and so on. "Job" is a common term on big computing clusters, but also known in common english, and it takes away the link between implementation and graphical interface.

Just my 2¢, and still waiting for the linux/i386 version (and x86_64, too, if there will be one) to further examine the interface.

db597 · 2007-06-19, 08:02

A single "job" can have multiple threads.. no?

Prime95 · 2007-06-19, 17:58

I've added links to the Linux, Linux 64-bit, and Intel Mac OS X versions. These are not well tested. Please report any bugs, though it is unlikely I'll be able to address them until I return around July 4th.

Linux specific tests and questions include:
1) load average option - I probably broke it.
2) Is the -B option needed anymore.
3) Should the check for another running version of mprime be changed or eliminated?

S00113 · 2007-06-19, 22:38

Quote:

Originally Posted by Prime95

I've added links to the Linux, Linux 64-bit, and Intel Mac OS X versions.

Finally a 64-bit version Linux!

But not without problems.. Could you link the 64-bit Linux version with libcurl version 3 instead of 4? Both RedHat RHEL 4 and 5 and the latest stable version of Ubuntu include libcurl version 3, not 4. The 32-bit version of mprime links with libcurl 3, and works fine for me so far on the systems I've tested.

S00113 · 2007-06-19, 23:58

Clarification of above: By version 3 i refer to the soname, which was bumped from 3 to four in curl version 7.16.0.

A bug: If the ComputerID in local.txt (converted from local.ini) is more than 13 characters, communication with the server will fail.

Prime95 · 2007-06-20, 00:12

Quote:

Originally Posted by S00113

Finally a 64-bit version Linux!

You can thank Agner Fog.

Quote:

But not without problems.. Could you link the 64-bit Linux version with libcurl version 3 instead of 4? Both RedHat RHEL 4 and 5 and the latest stable version of Ubuntu include libcurl version 3, not 4.

I guess Debian's default is 4. I'll work on problems with the static linking makefile when I return. Get me a nice long bug list to work on

tha · 2007-06-20, 14:15

In the past two months the number of exponents assigned to people who abandon assignments early on has dropped by about 15-20%. Most of these people will be overclockers who use the server anyway. The sudden drop is either caused by an individual who installs prime95 on an enormous amount of machines, or a website that is frequented by a lot of overclockers has changed it's advisories.

It would would be nice if the overclockers somehow would not make the switch to the new server. Any creative ideas besides the new policies that will be implemented and require the client to contact the server more often?

antiroach · 2007-06-26, 12:33

This might be a bit 'off topic' but I recently tried to do LMH work using mprime version 25.3.2. I used the FactorOverride=xx setting in the prime.txt file, however that setting never took effect and the numbers to be factored end up being factored to like 70 bits (when FactorOverride was set to 64). Is this a known issue? Should I just go back to using v24 of mprime?

sonjohan · 2007-07-03, 11:21

Quote:

Originally Posted by antiroach

This might be a bit 'off topic' but I recently tried to do LMH work using mprime version 25.3.2. I used the FactorOverride=xx setting in the prime.txt file, however that setting never took effect and the numbers to be factored end up being factored to like 70 bits (when FactorOverride was set to 64). Is this a known issue? Should I just go back to using v24 of mprime?

Did you uncheck 'Use primenet to get work & exponents'?

2007-06-17, 01:19	#1
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·4,079 Posts	Prime95 v25.3 Version 25.3 ------------ The best candidates for testing this early version are critical thinkers, owners of multi-core machines, those with odd internet settings such as a proxy server. Version 25.3 might be marginally slower than version 24.14. This version talks to the v4 PrimeNet server by default. I've been using it for 3 weeks and have successfully sent in a matching double-check and a first-time test. Steps to install version 25.3 ---------------------------- DO NOT INSTALL OVER VERSION 24. If you have problems, you want to be able to revert to testing using version 24. 1) Run version 24, uncheck start at bootup, exit version 24. 2) COPY (not move) your version 24 folder to a new place - say c:\v25test 3) Download ftp://mersenne.org/gimps/p95v253.zip. Unzip into c:\v25test. 4) Run version 25.3. Find problems. Report them to me. Links to other versions --------------------- Linux: ftp://mersenne.org/gimps/mprime253.tar.gz Linux64: ftp://mersenne.org/gimps/mprime253-linux64.tar.gz Mac OSX: ftp://mersenne.org/gimps/mprime253-MacOSX.tar.gz How to be most useful --------------------- I'm not only looking for bugs. I'm looking for wording that is confusing, features that are awkward to use, etc. I'll mention some special areas you can test. 1) Can you contact the v4 server? The comm layer is completely different - it uses libcurl. 2) Run Advanced/Time 9900 or Advanced/Time 9901. After hours or days look in results.txt to see if the words "Test failed" appear. If so, you have found a problem in the underlying FFT code. Dual and quad core users are especially encouraged to do some of this! 3) Re-run a past P-1 test that found a factor. There have been some changes to the P-1 code. 4) Re-run a trial factoring run that found a factor. The TF code has changed too. 5) Does Test/Status produce reasonable estimates? Does your RollingAverage get out of whack over time? 6) Is your CPU, L2 cache, hyperthreading, and num cores detected properly. 7) Try running PRP tests against a wide range of known primes from Chris Caldwell's database. There is a new work type in worktodo.txt: "PRP=k,b,n,c" Register your UI suggestions ---------------------------- 1) For example, I'm unhappy with Test / Worker threads dialog box: a) I'm not sure the average user knows what a "thread" is. Should I instead call them "worker Windows" and if so, what do I call them in the Linux version? And is "worker thread" a misnomer if the worker is actually using multiple threads in the FFT? b) The CPU affinity setting "Smart assignment" is meaningless. Maybe I should call this "do what makes the most sense"??? It really means if you have as many worker windows as CPUs then assign each worker window to its own CPU - otherwise run on any cpu. c) The text "Number of CPUs to use" is vague/wrong. It really is how many threads should the FFT code use. On a hyperthreaded machine, a setting of 2 really means use two logical cpus on one physical cpu. 2) Scott, has suggested a one window option. Instead of "Main thread", "Comm thread", Worker 1", they would all appear in one window (kinda like v24) with text next to the timestamp that says what thread generated the message. Do you have better or alternative suggestions? 3) Are the screen outputs well worded? Do the Window and program titles always make sense? Are there any volunteers out there that want to muck with MFC code? Saving the coordinates and sizes of MDI windows on exit would be nice (and restore on startup). The vertical scroll bars have a devil of a time staying at the bottom. Last fiddled with by Prime95 on 2007-06-19 at 16:36

2007-06-18, 00:05	#2
jasong "Jason Goatcher" Mar 2005 110110110011₂ Posts	In answer to the concerns Prime95 has with, um, the new Prime95, concerning the wording, I think we should make it "more" confusing. Use mnemonics that only make sense, but make perfect sense, if you read the help file. Of course, I'm guilty of clicking the "Accept" button without reading online contracts, so it's probably best for me to keep my mouth shut.

2007-06-18, 14:59	#3
Salz Jun 2007 3 Posts	I don't like the term "threads" either because it is an implementation detail of the running program. And if you use multiple threads for the TTF it is plain wrong. A much more better term would probably be "Job". So you could e.g. have a "Main Window", "Prime-Net Job 1", "Prime-Net Job 2" and so on. "Job" is a common term on big computing clusters, but also known in common english, and it takes away the link between implementation and graphical interface. Just my 2¢, and still waiting for the linux/i386 version (and x86_64, too, if there will be one) to further examine the interface.

2007-06-19, 08:02	#4
db597 Jan 2003 2×103 Posts	A single "job" can have multiple threads.. no?

2007-06-19, 17:58	#5
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·4,079 Posts	I've added links to the Linux, Linux 64-bit, and Intel Mac OS X versions. These are not well tested. Please report any bugs, though it is unlikely I'll be able to address them until I return around July 4th. Linux specific tests and questions include: 1) load average option - I probably broke it. 2) Is the -B option needed anymore. 3) Should the check for another running version of mprime be changed or eliminated?

2007-06-19, 23:58	#7
S00113 Dec 2003 2³·3³ Posts	Clarification of above: By version 3 i refer to the soname, which was bumped from 3 to four in curl version 7.16.0. A bug: If the ComputerID in local.txt (converted from local.ini) is more than 13 characters, communication with the server will fail.

2007-06-20, 14:15	#9
tha Dec 2002 360₁₆ Posts	In the past two months the number of exponents assigned to people who abandon assignments early on has dropped by about 15-20%. Most of these people will be overclockers who use the server anyway. The sudden drop is either caused by an individual who installs prime95 on an enormous amount of machines, or a website that is frequented by a lot of overclockers has changed it's advisories. It would would be nice if the overclockers somehow would not make the switch to the new server. Any creative ideas besides the new policies that will be implemented and require the client to contact the server more often?

2007-06-26, 12:33	#10
antiroach Jun 2003 2²·61 Posts	This might be a bit 'off topic' but I recently tried to do LMH work using mprime version 25.3.2. I used the FactorOverride=xx setting in the prime.txt file, however that setting never took effect and the numbers to be factored end up being factored to like 70 bits (when FactorOverride was set to 64). Is this a known issue? Should I just go back to using v24 of mprime?