GIMPS Credit for Exponents Between 79.3 Million and 596 Million

[cheesehead]

Originally Posted by jinydu

So in other words, I would need to have a P90 in order to find the iteration times,

No, there is enough information on the mersenne.org and mersenneforum.org, and perhaps on mersenndewiki, to enable calculating theoretical P90 iteration times (close enough for credit assignments, at least) from the iteration times on later models. (And no, I'm not implying that it would be a linear extrapolation.) Search around!

Perhaps someone can put together a table on mersennewiki (I presume there's none there now).

not to mention the fact that I don't know what the FFT length is for exponents larger than 79.3 million.

George, do you have these?

[jinydu]

For some reason, it looks like Mr. Woltman doesn't want to answer these questions; I don't know why.

[Prime95]

Originally Posted by jinydu

Given that the LL tests for M100,000,007 are scheduled to finish next month, I am curious as to how GIMPS credit will be assigned for exponents from 79.3 million to 596 million.

I'll have to assign credit manually, but calculating the amount will be easy.

[jinydu]

Thank you for the reply Mr. Woltman. So how will the amount of credit be calculated?

[cheesehead]

Originally Posted by jinydu

I don't know what the FFT length is for exponents larger than 79.3 million...

Look in Prime95 source module mult.asm at tables xjmptable and xjmptablep.

Some raw excerpts:

xjmptable DD 743, 32, 0.00000111, 896
...
allfft DD 68130000, 3670016, 0.323, 9456000
...
allfft DD 77910000, 4194304, 0.382, 5724160
...
allfft DD 96830000, 5242880, 0.485, 7045376
...
allfft DD 115300000, 6291456, 0.668, 8364416
...
allfft DD 134200000, 7340032, 0.886, 9683456
...


xjmptablep DD 739, 32, 0.00000111, 1152
...
allfft DD 58410000, 3145728, 0.289, 4143104
...
allfft DD 77700000, 4194304, 0.425, 5462016
...
DD 115000000, 6291456, 0.668, 9650176
...
DD 153100000, 8388608, 1.042, 12828672
...

I'll leave those to your interpretation. :-)

[StarQwest]

I have a fair way of calculating how much credit I should recieve for M100,000,007. Since I am testing two exponents (on a two processor computer so they recieve equal resources), M36,234,713 (known credit after completion) and M100,000,007 (unknown credit), we can simply compare the two:
Fastest time per iteration possible:
M36,234,713: 0.074 seconds
M100,000,007: 0.254 seconds
Fastest possible time to test:
M36,234,713: (0.074 sec / iter)(36,234,713 iter) = 2,681,368 sec = 31.034 days
M100,000,007: (0.254 sec / iter)(100,000,007 iter) = 25,400,002 sec = 293.982 days
Obviously, these tests are taking slightly longer than this as the computer is slowed by other applications. However, this is the ideal time. Now we take the ratio of the two:
(293.982 days / 31.034 days) = 9.4729
Note that (100,000,007 / 36,234,713)^2 = 7.6164, so the time actually increases slightly more than quadratic. However, once M36,234,713 finishes, I can check and see how much credit I recieved for that test and multiply it by 9.4729 to find out how much credit I will recieve for M100,000,007. It should end up being ~65 - 70 P90 CPU years.

[StarQwest]

To George Woltman:

When are you going to update your site to include work on exponents larger than 79.3 million, not just M100,000,007, but also the 100M digits prefactor project, MasterPrime2006, and Operation Billion Digits? There is significant factoring work going on in this range, and LL tests are coming soon. Also, there are some banners on http://www.mersenne.org/primenet/status.shtml that need to be updated due to the incredible improvement and growth of GIMPS over the last few years:

Some banners (that need to be updated) say:

GIMPS / Primenet = 38 Cray T916 Supercomputers. Whoa.
GIMPS does 79 years of computing- every day. Help out.
GIMPS / Primenet. Join 15,000 fellow computer users.
38 of them are known in the universe. Find one more.
27,000 computers. Working as one. GIMPS / Primenet.
It exists. Hiding. Cloaked. Find Mersenne Prime #39.
GIMPS. Because 2,098,960 digits just aren't enough.
GIMPS. Because 2^6,972,593-1 just isn't big enough.
950,000,000,000 floating point operations / sec. Nifty.


These banners should say instead:

GIMPS / Primenet = 400 Cray T932 Supercomputers. Whoa.
GIMPS does 1900 years of computing- every day. Help out.
GIMPS / Primenet. Join 50,000 fellow computer users.
44 of them are known in the universe. Find one more.
76,000 computers. Working as one. GIMPS / Primenet.
It exists. Hiding. Cloaked. Find Mersenne Prime #45.
GIMPS. Because 9,808,358 digits just aren't enough.
GIMPS. Because 2^32,582,657-1 just isn't big enough.
22,800,000,000,000 floating point operations / sec. Nifty.

[S485122]

Originally Posted by StarQwest

Note that (100,000,007 / 36,234,713)^2 = 7.6164, so the time actually increases slightly more than quadratic. However, once M36,234,713 finishes, I can check and see how much credit I recieved for that test and multiply it by 9.4729 to find out how much credit I will recieve for M100,000,007. It should end up being ~65 - 70 P90 CPU years.

No need to wait until M36 234 713 finishes to know the credit you will get, somewhere in th eforum is the formula (the algorythm to be exact), it will be : 7.7477 P90 CPU years.
According to YOUR formula 100 000 007 should get you 73.3932 P90 CPU years.

[jinydu]

You can use this: http://www.teamprimerib.com/rr1/bin/...onent=36234713

But is it really correct to give the smallest iteration times? Wouldn't it be more accurate to give the average iteration time?

Originally Posted by 'www.mersenne.org/bench.htm'[/QUOTE

Where more than one user has reported timings for a given CPU, anomalous timings are discarded and the best times are reported.

And I can't seem to reconcile cheesehead's data with that given the table given in http://www.mersenneforum.org/showthr...&highlight=P90 .

[StarQwest]

Originally Posted by Jacob Visser

No need to wait until M36 234 713 finishes to know the credit you will get, somewhere in th eforum is the formula (the algorythm to be exact), it will be : 7.7477 P90 CPU years.
According to YOUR formula 100 000 007 should get you 73.3932 P90 CPU years.

Sorry. I thought M36,234,713 would be about 7.2 CPU years. But if M100,000,007 is 73.3932 CPU years, then that is even better. I'll know for sure in about a week (when M36,234,713 finishes). If this is so, then it will mean that M100,000,007 requires 7.34 x 10^16 calculations:surprised to prove whether or not it is prime!

[cheesehead]

Originally Posted by jinydu

I can't seem to reconcile cheesehead's data

My data? Are you referring to the entries in the "[x]jmptable[p]" tables in Prime95 source module mult.asm?

the table given in http://www.mersenneforum.org/showthr...&highlight=P90 .

To which of the tables in http://www.mersenneforum.org/showthread.php?t=2731 are you referring?