Oops - New Prime! (M49 related)

[Dubslow]

Originally Posted by bgbeuning

So GIMPS really stands for "George's Internet Mersenne Prime Search"?

So George is user id 4 on the forum. There is one before him.

http://www.mersenneforum.org/member.php?u=3

That would be the founder and administrator

[ixfd64]

http://techreport.com/news/29629/gim...mersenne-prime

'Dat first comment...

[Prime95]

Originally Posted by airsquirrels

Is there any reason Primenet should still even hand out TF assignments, ever?

Assignment rules have been tweaked for "what makes most sense". You now need a 400MHz P4 equivalent (was 800MHz) to get a double-check. If your computer is not that fast (Atom processors, Semprons with 128KB L2 caches, computers marked as running just a few hours a day) you will now get ECM work instead of TF work.

BTW, there are some recent TF assignments that I cannot figure out. I think the computers should be getting LL work based on the computer info in the database. Has anyone here seen a case where "what makes most sense" returned TF work when you think LL work would have been more appropriate?

[Madpoo]

Originally Posted by Madpoo

Well... whatever.

I set the mockup version to have a baseline of zero for now... it just compresses the vertical data even more. I'm still not convinced it's any better. It's only use is for a quick eyeball to see changes over the previous week and I'm not sure a baseline of zero contributes to that at all.

Other opinions out there?

I added a graph to show LL results for the past year (and modified the # of new users to be logarithmic since it has really spiked recently).

Interesting thing about the LL results over the past year... I had to exclude all of my own results that were for exponents < 7M since I did those thousands of triple-checks and other random things. It was seriously skewing the totals.

I'm still not liking the baseline=0 for the tflop graph. LOL

[kladner]

Originally Posted by ixfd64

http://techreport.com/news/29629/gim...mersenne-prime

'Dat first comment...

I sure hope that was a joke.

[Mark Rose]

Originally Posted by Madpoo

I added a graph to show LL results for the past year (and modified the # of new users to be logarithmic since it has really spiked recently).

Interesting thing about the LL results over the past year... I had to exclude all of my own results that were for exponents < 7M since I did those thousands of triple-checks and other random things. It was seriously skewing the totals.

Would it be possible to separate out the DC? That help visualize the growing/shrinking gap between the two. Perhaps two lines on one graph?

I'm still not liking the baseline=0 for the tflop graph. LOL

I do appreciate you were willing to try it out. I prefer it, but I'm also not the one who would get the most out of the graph.

[Madpoo]

Originally Posted by Mark Rose

Would it be possible to separate out the DC? That help visualize the growing/shrinking gap between the two. Perhaps two lines on one graph?

Hmm... I was just putting some thought into that.

While it's technically possible to look at test results and figure out which ones were first time and which were double-checks, it's actually a little time consuming to do so on the fly, especially quantizing per day.

The only way that might work would be to cache the result set rather than have it generated each time the page is viewed (which, of course, makes sense to do even if the query is relatively "cheap").

One of those things that I might have to put some think time in to. The existing graph on the page (the tflops/second) actually does come from pre-calculated data so that wasn't an issue... the new ones I added are spontaneously created but I'm sure I could put them into some table that's updated hourly if they prove useful.

[ATH]

Originally Posted by Prime95

BTW, there are some recent TF assignments that I cannot figure out. I think the computers should be getting LL work based on the computer info in the database. Has anyone here seen a case where "what makes most sense" returned TF work when you think LL work would have been more appropriate?

Maybe this bug was not completely crushed?
http://www.mersenneforum.org/showthread.php?t=20549

[ATH]

Originally Posted by bgbeuning

So George is user id 4 on the forum. There is one before him.

George at least has that the first post in the first thread:

http://www.mersenneforum.org/showpos...=1&postcount=1
http://www.mersenneforum.org/showthread.php?t=1

The forum started in Aug 2002, only 6 months before I joined GIMPS in Feb 2003. Although it did not sign up on the forum before Dec 2003 apparently.

[Madpoo]

Originally Posted by ATH

Maybe this bug was not completely crushed?
http://www.mersenneforum.org/showthread.php?t=20549

That particular one should be resolved.

[airsquirrels]

Originally Posted by Prime95

Assignment rules have been tweaked for "what makes most sense". You now need a 400MHz P4 equivalent (was 800MHz) to get a double-check. If your computer is not that fast (Atom processors, Semprons with 128KB L2 caches, computers marked as running just a few hours a day) you will now get ECM work instead of TF work.

BTW, there are some recent TF assignments that I cannot figure out. I think the computers should be getting LL work based on the computer info in the database. Has anyone here seen a case where "what makes most sense" returned TF work when you think LL work would have been more appropriate?

In the past few months there have been a handful of times where all the workers on a new machine that was set to LL/WR or What Makes Sense suddenly started getting 120M or 200M TF work. Upon going to the settings page on the website it was all set to TF and I had to switch it back. Once I switched it back it never reoccured for that machine.