GPU to 72 status...

[kladner]

Originally Posted by James Heinrich

On the P-1 Top 500, Axon is blazing up the rankings and has found 3193 P-1 factors in the past year (probably less). :surprised

Sheesh! I see what you are saying about Axon, but The Judger has already overrun most of the GPU272 heavy hitters in PrimeNet ranking. I'm watching to see how that steep trajectory evens out. It seems that he might blow just about everyone down a notch on GPU272.

[LaurV]

Originally Posted by Dubslow

But those aren't actually P-1; there's a thread somewhere here about the low TF work he's doing. This will disappear when PrimeNet's credit assignment system is fixed.

Yes, see here.

[flashjh]

Originally Posted by chalsall

@petrw1: "This is curious?

Not terribly.

I haven't yet finished the code required to predict P-1 factors found.

It's much more complicated than predicting simple TFing.

But I am amused that someone finally noticed (or, perhaps, noticed and brought it forward).

To get the P-1 prediction going, I came up with the following using the data from G272's P-1 stats page at ~0500 UTC today.

The data shows an average across the board of 1 factor per every 21.28 P-1 exponents. Predictions can be made per the following:

 
ROUNDDOWN( ( P-1 done + factors found ) / 21.28)

This does not take into account everyone's actual B1/B2 or E value, only the raw data and a median calcualted off of the results from G272. I know that actual P-1 factor probablity is based on a lot more, but I thought it a good start to see where everyone stands.

[Bdot]

Originally Posted by chalsall

Not NFS version 2 nor 3.

I now tested all (accessible to me) variations of NFSv3, Samba, cifs, and of course, local file systems. All behave as expected with the O_EXCL | O_CREAT options. (I did not test hard for race conditions, though.)

The mkdir approach may work as well, and it may also be atomically on some systems, but this is not documented (opposed to "open"), so I will not rely on it.

On NFSv2, or when running a Linux kernel before 2.6, the open call may contain a race condition - not sure if anyone has a 64-bit Linux still running a 2.4 kernel ... probably mfakto would not run on that anyway.

So I think I will keep the current implementation, and release 0.11 soon.

[chalsall]

Originally Posted by Bdot

So I think I will keep the current implementation, and release 0.11 soon.

Cool!!!

I guess this means I'll have to reciprocate, and deliver my promised work fetching spider....

[petrw1]

Originally Posted by kladner

Sheesh! I see what you are saying about Axon, but The Judger has already overrun most of the GPU272 heavy hitters in PrimeNet ranking. I'm watching to see how that steep trajectory evens out. It seems that he might blow just about everyone down a notch on GPU272.

Torchwood is moving up awful fast too.

Here is the progress of the top 10 so far in 2012

                                  Total                     |      Rank Change       | Overtake date based on
 Rank  Member Name              GHz-Days Attempts Successes |90day 30day 7 day 1 day | 90 days 30 days  7 days
------ -------------------- ------------ -------- --------- |----- ----- ----- ----- | ------- ------- -------
     1 ANONYMOUS                8383.446     3184        96 |   1                   |          366dy   285dy 
     2 Torchwood Institute      6308.100     1657        84 | ***    1             |                        
     3 Jerry Hallett            5814.907     1963       233 |  13    5    1       |                        
     4 James Heinrich           5558.519     4976       212 |   3          1       |                        
     5 TheJudger                5410.186     1431        72 |   1    2    1       |                        
     6 Never Odd Or Even        4991.143      130         1 |   5    4    3       |   95dy    76dy    46dy 
     7 kdgehman                 4843.124      978        45 |   2    1             |                        
     8 curtisc                  4076.383     1768        60 |   5    3             |                        
     9 GrunwalderGIMP           3588.970     1099        50 |  20    2    1       |                        
    10 Axon                     3283.901    16716         0 |  78          1       |

And since April 1-2012

                                  Total                     |      Rank Change       | Overtake date based on
 Rank  Member Name              GHz-Days Attempts Successes |90day 30day 7 day 1 day | 90 days 30 days  7 days
------ -------------------- ------------ -------- --------- |----- ----- ----- ----- | ------- ------- -------
     1 Jerry Hallett            2317.640      731        94 |   7    2    1       |                        
     2 James Heinrich           2019.427     2226        83 |   3    2    1       |                 >999dy 
     3 TheJudger                1640.387      452        19 |   1    4    2       |                        
     4 GrunwalderGIMP           1634.590      506        20 |  30    2          1 |                        
     5 Torchwood Institute      1512.187      410        20 | ***    4    4    1 |           14dy    13dy 
     6 kdgehman                 1433.976      299        13 |         1             |          365dy         
     7 ANONYMOUS                1331.446      485         9 |   6    5             |   48dy    53dy   112dy 
     8 David Campeau            1140.745      403        14 |   9                   |                        
     9 monst                     751.597      240        18 |   2    1             |  146dy   207dy    27dy 
    10 curtisc                   743.543      335        16 |   7    1    1    1 |   62dy    93dy    69dy

Looks like a pair of "JH" has it recently....could that be a "Jack-of-Hearts" in disguise?

[Dubslow]

Hey chalsall,
I've noticed that login sessions are not shared between the separate language subdomains. While it doesn't really have a practical impact, since in theory a user will choose one language and stick with it, it would be nice for translators or the curious who like to switch back and forth if you could log in the English site and have that transfer to the French site (et cetera).
Dubslow

(It seems like such a... 'sticky' (gooey? floppy-with-formalness?) post. )

[chalsall]

Originally Posted by Dubslow

I've noticed that login sessions are not shared between the separate language subdomains.

No kidding. Do you understand why?

Originally Posted by Dubslow

While it doesn't really have a practical impact, since in theory a user will choose one language and stick with it, it would be nice for translators or the curious who like to switch back and forth if you could log in the English site and have that transfer to the French site (et cetera).

The curious will probably not be accessing password protected areas.

Translators will hopefully understand the reason for the need to log in to password protected areas for each language subdomain.

[bcp19]

No new DC assignments?

[chalsall]

Originally Posted by bcp19

No new DC assignments?

Xyzzy came by, and reserved all the DC TF candidates. I've topped up the supply again.

[chalsall]

Hey all.

Just to let you know, the "GPU to 72" team on PrimeNet is now #1 for P-1 work, #1 for TF work, and #3 overall.

Thanks for all the cycles everyone!!!