Communications

[ET_]

I will be away from the computer for one or two weeks starting from 29th of November. If you have reservations to do, please communicate them immediately. Thank you!

[ET_]

I will be away from the computer for three weeks starting from 15th of October. Please be patient, as I will not take a computer with me.
Feel free to find as many Fermat factors as you like.

Thank you!

[rogue]

The page http://www.fermatsearch.org/stat/stats.php, relies on Adobe Flash, which has been delegated to the dust bin of history. I don't know if other pages rely on Adobe Flash, but they should be updated to remove it.

[ET_]

Originally Posted by rogue

The page http://www.fermatsearch.org/stat/stats.php, relies on Adobe Flash, which has been delegated to the dust bin of history. I don't know if other pages rely on Adobe Flash, but they should be updated to remove it.

I forgot it... shame on me.

I will do the updates as soon as I can, thank you for the head-up!

Luigi
---

[rogue]

On this page http://www.fermatsearch.org/productivity.html, it doesn't state how the rate is calculated for gfndsieve+pfgw or ppsieve+pfgw. The calculation is fairly straightforward for the other programs.

Also, should there be a column for gfndsieve with -x? This would be close to GMP-Fermat in speed. The column for gfndsieve+pfgw assumes that -x is not used as pfgw does the GFN testing.

Should the page state that pfgw supports xGFN testing while the others do not (to my knowledge).

Should the page also have a column for size of the range of k tested? I ask because as n increases, the size of the range for k will decrease.

It might also make sense when using gfndsieve+pfgw or ppsieve+pfgw to specify an approximate sieve depth for the n.

[ET_]

Originally Posted by rogue

On this page http://www.fermatsearch.org/productivity.html, it doesn't state how the rate is calculated for gfndsieve+pfgw or ppsieve+pfgw. The calculation is fairly straightforward for the other programs.

Also, should there be a column for gfndsieve with -x? This would be close to GMP-Fermat in speed. The column for gfndsieve+pfgw assumes that -x is not used as pfgw does the GFN testing.

Should the page state that pfgw supports xGFN testing while the others do not (to my knowledge).

Should the page also have a column for size of the range of k tested? I ask because as n increases, the size of the range for k will decrease.

It might also make sense when using gfndsieve+pfgw or ppsieve+pfgw to specify an approximate sieve depth for the n.

Hi Mark, thank you for jumping in

The page tends to give an approximate idea of the "productivity" of different programs on a defined N and k range. There are just too many variables to take into account to offer n analytical comparison: CPU frequency, overclocking, available RAM RAM speed, number of cores, inter-process communication speed, size and speed of caches and so on.

What I did (and Gary and Roman helped me with it) was to calculate well-defined ranges using different programs, and show the relative speed in keys per second on a standard PC core. This way to measure performance obviously created some issues with programs that work in pairs like gfndsieve and ppsieve and pfgw, because the mentioned variables now apply to 2 different programs.

I didn't play much with the -x switch of gfndsieve, to be honest. I used it mostly as a siever, so I initially chose to not add such option to the list, but you are right in saying that people could benefit from it.

I guess I could split the section into 2 different columns: plain gfndsieve and gfndsieve + pfgw. - Added to the to-do list ;-)

For the sieve depth, I will add a note saying that the sieve should continue until the key elimination time will be greater than the primality testing of each key. - Added to the to-do list.

[rogue]

Originally Posted by ET_

Hi Mark, thank you for jumping in

The page tends to give an approximate idea of the "productivity" of different programs on a defined N and k range. There are just too many variables to take into account to offer n analytical comparison: CPU frequency, overclocking, available RAM RAM speed, number of cores, inter-process communication speed, size and speed of caches and so on.

What I did (and Gary and Roman helped me with it) was to calculate well-defined ranges using different programs, and show the relative speed in keys per second on a standard PC core. This way to measure performance obviously created some issues with programs that work in pairs like gfndsieve and ppsieve and pfgw, because the mentioned variables now apply to 2 different programs.

I didn't play much with the -x switch of gfndsieve, to be honest. I used it mostly as a siever, so I initially chose to not add such option to the list, but you are right in saying that people could benefit from it.

I guess I could split the section into 2 different columns: plain gfndsieve and gfndsieve + pfgw. - Added to the to-do list ;-)

For the sieve depth, I will add a note saying that the sieve should continue until the key elimination time will be greater than the primality testing of each key. - Added to the to-do list.

The top of the pages states the CPU used to generate the data. It doesn't show the GPU that was used and that will likely have a much bigger impact than the CPU. Obviously different hardware configurations will yield different results, but I would expect that for the CPU that the numbers will scale similarly from CPU to CPU.

gfndsieve with -x is only meaningful up to about n=5000. Above that I expect it to be slower than other options. It will likely be slower than GMP-Fermat, but only because it validates factors. I do not know if any of the other programs validate their factors before testing.