Prime95 version 29.2

[Prime95]

Prime95 version 29.2 build 4 is available.

This is for early testers to find bugs. I'm sure I've added some new ones.

From whatsnew.txt:

1)  Throughput benchmarks of all FFT implementations are written to gwnum.txt. 
    These benchmarks will then be used to pick the best FFT implementation to use.
    This can lead to minor performance improvements.
2)  Benchmarks are run at 5AM (and every 21 hours after that) for any FFT sizes that
    will be needed within the next 7 days.  Once enough benchmark data is accumulated,
    these automatic benchmarks no longer take place.
3)  Default FFT implementations for Ryzen added.
4)  A few new FFT sizes (up to 50M) are enabled for FMA3-capable CPUs.
5)  ECM and P-1 on Fermat numbers will automatically add known Fermat factors to
    the worktodo.txt entries.

This version is not heavily tested - consider it a beta version. Bug reports welcome!

Download links:
Windows 64-bit: ftp://mersenne.org/gimps/p95v292.win64.zip
Linux 64-bit: ftp://mersenne.org/gimps/p95v292.linux64.tar.gz
Mac OS X: ftp://mersenne.org/gimps/p95v292.MacOSX.zip
Windows 32-bit: ftp://mersenne.org/gimps/p95v292.win32.zip
Linux 32-bit: ftp://mersenne.org/gimps/p95v292.linux32.tar.gz
FreeBSD11 64-bit: ftp://mersenne.org/gimps/p95v292.FreeBSD11-64.tar.gz
Source: ftp://mersenne.org/gimps/p95v292.source.zip

[Prime95]

1) The Windows version does not use new benchmark data in FFT selection until prime95 is restarted. Fixed in 29.2 build 3.
2) Throughput benchmarks using less than all cores improperly used all cores. Fixed in 29.2 build 4.
3) Using benchmark data on AMD Bulldozer with SUM(INPUTS) checking enabled can lead to errors initializing FFTs. Fixed in 29.2 build 5.
4) Auto benchmark runs even if user is not running any workers. Auto benchmark runs on battery power even if "run on battery power" option is not selected. Fixed in 29.2 build 5.

[Prime95]

I found the first bug. Prime95 is not using the results of the automatic benchmarks to select the best FFT implementation until prime95 is restarted.

[GP2]

Originally Posted by Prime95

I found the first bug. Prime95 is not using the results of the automatic benchmarks to select the best FFT implementation until prime95 is restarted.

I was about to ask when gwnum.txt was supposed to show up in the directory. I think I stopped and restarted it a few times (for unrelated reasons).

[Prime95]

Originally Posted by GP2

I was about to ask when gwnum.txt was supposed to show up in the directory. I think I stopped and restarted it a few times (for unrelated reasons).

You can generate benchmark data for FFT selections manually using the benchmark dialog box. Do a throughput benchmark using all CPU cores and the number of workers you have currently configured.

Or you can let prime95 generate the benchmark data automatically. The first auto-benchmark will take place around 5AM (and occur about every 21 hours thereafter).

[Prime95]

Originally Posted by Prime95

I found the first bug. Prime95 is not using the results of the automatic benchmarks to select the best FFT implementation until prime95 is restarted.

Found the problem. It is a compiler optimization problem. I can't complain about Microsoft too much since it is a 12 year old compiler.

[S485122]

The "File Version" of Prime95 29.2 is still 29.1.1.0, the "Product Version" is 29.1.0.0.

Something that would be useful is the possibility to differentiate between the different versions of the program, there have been several versions of Prime95 29.1. What you could do is name them xx.x.B for beta until the version is OK Then you would remove the B.Going by the date is not always reliable.

Jacob

[GP2]

Originally Posted by S485122

Something that would be useful is the possibility to differentiate between the different versions of the program, there have been several versions of Prime95 29.1.

In practice you can do that by going to https://mersenne.org/cpus/ while logged in, and seeing what build number is reported there under "CPU / Software".

[S485122]

Originally Posted by GP2

In practice you can do that by going to https://mersenne.org/cpus/ while logged in, and seeing what build number is reported there under "CPU / Software".

Not completely relevant : first of all one has to run an instance with the mprime or Prime95 to have the version of the software there. It is quicker to run the software and use the menu to get the version and the build.

But my point is about the software as found on ftp://ftp.mersenne.org and the fact that it would be useful to have the version and subversion explicitly in the name.

Jacob

[henryzz]

Originally Posted by Prime95

You can generate benchmark data for FFT selections manually using the benchmark dialog box. Do a throughput benchmark using all CPU cores and the number of workers you have currently configured.

Or you can let prime95 generate the benchmark data automatically. The first auto-benchmark will take place around 5AM (and occur about every 17 hours thereafter).

5 am won't work for anyone who only runs during the daytime.

[tului]

You need any specific benchmarks you want me to run on my Ryzen 1800X now that I have DDR4-3200 available?