Holy new Mersenne prime, Batman! (M47 related)

[davieddy]

Originally Posted by Mini-Geek

Ah, but at the same time, it's likely that he will use Enigma again in his next war

Hitler has just left the building

[Andi47]

Originally Posted by davieddy

Hitler has just left the building

There are buildings down there?

[ADBjester]

Originally Posted by davieddy

Hitler has just left the building

I don't think Godwin's Law applies to web-based forums. Only Usenet.

Nice try, though.

Jester

[jinydu]

Originally Posted by davieddy

This thread is nearing unfathomable proportions, but I recall Jeff
suggesting that Ernst might beat him to it.
So on the strength(?) of that I would guess the 12th

Well I can try to use that information to estimate an upper bound...

Gilchrist started his verification run for M44 on Sept. 5 and finished on Sept. 15. Since I'm looking for an upper bound on the size of M45, I'll round down to 9 days. This time, Gilchrist appears to have started his verification on August 27. If he finishes on Sept. 12, that will be 14 days. Assuming he is using less CPU power this time, this means that the new exponent requires at most 56% more CPU time to test.

According to the CPU years calculator on the TPR website (which admittedly is out of date), that means M45 takes up at most 9.84 P90 CPU years. So the exponent must be at most 40.25M.

I think the last paragraph is far off the mark though because the FFT boundaries have since been adjusted...

[ewmayer]

Originally Posted by Batalov

Someone simply had to do this edit - http://en.wikipedia.org/wiki/M45

I believe the first link there provides the vital clue which is needed to answer the riddle of why the verification is taking so much longer, compared to what the estimated GIMPS "wavefront" of the past few months and the speed of the last verify would imply:


wait for it...



Answer: The verify is taking longer because it's "messier" than the last one.

[Mini-Geek]

Originally Posted by jinydu

Well I can try to use that information to estimate an upper bound...

Gilchrist started his verification run for M44 on Sept. 5 and finished on Sept. 15. Since I'm looking for an upper bound on the size of M45, I'll round down to 9 days. This time, Gilchrist appears to have started his verification on August 27. If he finishes on Sept. 12, that will be 14 days. Assuming he is using less CPU power this time, this means that the new exponent requires at most 56% more CPU time to test.

According to the CPU years calculator on the TPR website (which admittedly is out of date), that means M45 takes up at most 9.84 P90 CPU years. So the exponent must be at most 40.25M.

I think the last paragraph is far off the mark though because the FFT boundaries have since been adjusted...

So, lesse here...

14=new time for M45=full speed M45 time*1.56, 14/1.56=new time=8.97...
M44/time for M44*full speed M45 time=estimate before FFT change accounted for
32582657/9*8.97=32474048.14333...

That upper bound on the slower CPU time would make it smaller than M44 (unless I made some mistake). Let's try a smaller difference, like 25%

14=new time for M45=full speed M45 time*1.25, 14/1.25=new time=11.2
M44/time for M44*full speed M45 time=estimate before FFT change accounted for
32582657/9*11.2=40547306.4888...

That's an FFT change from 1792K to 2560K. The average factor for that size change on SSE2 is 1.49.
40547306.49*1.49=60415486.6701

With that size, it'd need an even larger FFT, but I'm not even going to look at that, since it's way out of the range of anything completed during the time. I think these calculations are being way too pessimistic to get anything like a real number outta this. Let's try it with pessimistic dates but 100% CPU speed of last time.

M44/time for M44*estimated M45 time=estimate before FFT change accounted for
32582657/9*14=50684133.11111111111111111111

Hm, even that, before accounting for the FFT difference, is way too high. I think the days picked are far too pessimistic. Let's try 9 days for M44 and 12 days for M45 with a 15% slowdown...

12=new time for M45=full speed M45 time*1.15, 12/1.15=new time=10.43...
M44/time for M44*full speed M45 time=estimate before FFT change accounted for
32582657/9*10.43=37759679.16777...

Which would put it into 2048K range, so I need to adjust for the 2048K change...
37759679.16778*1.12=42290840.6679136
Which would put it into 2560K range, so I need to adjust for the 2560K change...
37759679.16778*1.49=56261921.9599922
Which would put it into 3072K range, so I need to adjust for the 3072K change...
37759679.16778*1.83=69100212.8770374
and so on...this'll go on for just about ever...what am I doing wrong?
Edit: Is it supposed to stop after the first time, and suggest 42290841? If so, my guess is AD9500DD (big surprise, right?)

[FactorEyes]

No big deal. Even if it is prime, there's another prime, Mersenne or not, that's bigger.

[davieddy]

Originally Posted by ewmayer

Answer: The verify is taking longer because it's "messier" than the last one.

Very drole.
But presumably the same could be applied to M44....

[Jud McCranie]

How about this for faster confirmations - have Prime95 save its data after 10%, 20%, 30%, etc of the iterations. Then 10 instances of a confirmation program could be run in parallel, making the wait for a confirmation much shorter.

[jrk]

You can already do that with InterimFiles=xxxxxxx.

[cheesehead]

Originally Posted by Mini-Geek

So, lesse here...

14=new time for M45=full speed M45 time*1.56, 14/1.56=new time=8.97...

< snip >

I think these calculations are being way too pessimistic to get anything like a real number outta this. Let's try it

< snip >
...what am I doing wrong?

The icosahedral dice rolls. http://mersenneforum.org/showpost.ph...&postcount=189

... and we're still waiting for those raccoon pelts http://mersenneforum.org/showpost.ph...3&postcount=24