me, myself and I

ramgeis · 2013-06-07, 18:08

Hi,

I started a LL Test and simultaneously already the DoubleCheck on different hardware (CPU vs GPU) on another system.

Now my question is, will the PrimeNet server accept my 2nd result of the same exponent when I'm submitting it?

lycorn · 2013-06-07, 18:10

Yes, it will.
If both tests were on a GPU, it wouldn´t.

LaurV · 2013-06-08, 04:09

To be sure, report the manual one (GPU) first. It will, anyhow, finish faster. The other way around you risk to get "this computer already reported this result" when you do the manual report after P95 reported the CPU result.

ramgeis · 2013-06-11, 01:30

Thanks for the info, I'll give it a try.
I wouldn't bet on the GPU though. ;)
At the moment the estimations are that the GPU finishes 2nd.

kladner · 2013-06-11, 05:21

Quote:

Originally Posted by ramgeis

Thanks for the info, I'll give it a try.
I wouldn't bet on the GPU though. ;)
At the moment the estimations are that the GPU finishes 2nd.

So- really fast CPU, or relatively slow GPU? Some combination? Inquiring minds, and all that.....

Brian-E · 2013-06-11, 08:55

And another one for the inquiring minds:

Has GIMPS finally ditched the principle that double checks ought to be carried out by a different user than the original test was? Or, more accurately, that an LL test result is considered definitive when two independent users have produced the same residue?

LaurV · 2013-06-12, 13:06

Different LL runs with P95 are certainly accepted and relatively "safe" even if they are ran by the same user, as they contain different shifts and checksums. It will take an user with some knowledge of the P95's internals to falsify such reports, and those users won't waste the time and won't risk their reputation just to get few more GHzDays. When a "manual" (GPU) run comes into equation, the things are tricky, as such report can be done by anybody. One childish user could run LL test and report (eventually incorrect) results as both first-time-LL and DC-LL, therefore causing GIMPS to miss primes for a while (until triple tests). Two GPU results are not accepted anyhow, no matter if they come from the same user or not. There should be at least a P95 run with a different shift (all GPU runs have the shift zero).

ramgeis · 2013-06-12, 18:23

Quote:

Originally Posted by kladner

So- really fast CPU, or relatively slow GPU? Some combination? Inquiring minds, and all that.....

Actually both quite fast in their area. :) In the blue corner a dual e5-2670 xeon system (which means 32 cores with hyperthreading) and in the red corner a quadro k5000... for the given exponent the time per iteration is approx. 7ms vs approx. 8ms

kladner · 2013-06-12, 21:30

Quote:

Originally Posted by ramgeis

Actually both quite fast in their area. :) In the blue corner a dual e5-2670 xeon system (which means 32 cores with hyperthreading) and in the red corner a quadro k5000... for the given exponent the time per iteration is approx. 7ms vs approx. 8ms

Yikes!

Nice box, if I may say so!

Batalov · 2013-06-13, 00:39

~~Producing~~Faking the second manual result "on the GPU" after having the complete residue known from the CPU can be easily done even by a 9-year old. Changing the order and submitting the "GPU" result first - by a clever 9-year old ...or by any 10-year old.

It has been demonstrated that two matching results (with different shifts, and produced with unfettered Prime95 binary) are easily produced for any input. Maybe not by a 10-year old though...

ramgeis · 2013-06-13, 14:01

Quote:

Originally Posted by Batalov

~~Producing~~Faking the second manual result "on the GPU" after having the complete residue known from the CPU can be easily done even by a 9-year old. Changing the order and submitting the "GPU" result first - by a clever 9-year old ...or by any 10-year old.

It has been demonstrated that two matching results (with different shifts, and produced with unfettered Prime95 binary) are easily produced for any input. Maybe not by a 10-year old though...

In other words, the whole LL-checking process only reduces the chance to prove that M( p ) is prime and can't prove that M( p ) isn't prime, right?

If false reports are possible, wouldn't that mean that the announcement of M( p ) as the n-th Mersenne prime must include at least one factor for each non-prime M( p' ) with p' < p ?

2013-06-07, 18:08	#1
ramgeis Apr 2013 165₈ Posts	me, myself and I Hi, I started a LL Test and simultaneously already the DoubleCheck on different hardware (CPU vs GPU) on another system. Now my question is, will the PrimeNet server accept my 2nd result of the same exponent when I'm submitting it?

2013-06-12, 13:06	#7
LaurV Romulan Interpreter Jun 2011 Thailand 3³×347 Posts	Different LL runs with P95 are certainly accepted and relatively "safe" even if they are ran by the same user, as they contain different shifts and checksums. It will take an user with some knowledge of the P95's internals to falsify such reports, and those users won't waste the time and won't risk their reputation just to get few more GHzDays. When a "manual" (GPU) run comes into equation, the things are tricky, as such report can be done by anybody. One childish user could run LL test and report (eventually incorrect) results as both first-time-LL and DC-LL, therefore causing GIMPS to miss primes for a while (until triple tests). Two GPU results are not accepted anyhow, no matter if they come from the same user or not. There should be at least a P95 run with a different shift (all GPU runs have the shift zero). Last fiddled with by LaurV on 2013-06-12 at 13:19

2013-06-07, 18:10	#2
lycorn Sep 2002 Oeiras, Portugal 1,451 Posts	Yes, it will. If both tests were on a GPU, it wouldn´t.

2013-06-08, 04:09	#3
LaurV Romulan Interpreter Jun 2011 Thailand 3³·347 Posts	To be sure, report the manual one (GPU) first. It will, anyhow, finish faster. The other way around you risk to get "this computer already reported this result" when you do the manual report after P95 reported the CPU result.

2013-06-11, 01:30	#4
ramgeis Apr 2013 3²·13 Posts	Thanks for the info, I'll give it a try. I wouldn't bet on the GPU though. ;) At the moment the estimations are that the GPU finishes 2nd.

2013-06-11, 08:55	#6
Brian-E "Brian" Jul 2007 The Netherlands 7·467 Posts	And another one for the inquiring minds: Has GIMPS finally ditched the principle that double checks ought to be carried out by a different user than the original test was? Or, more accurately, that an LL test result is considered definitive when two independent users have produced the same residue?

2013-06-13, 00:39	#10
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 41×229 Posts	~~Producing~~Faking the second manual result "on the GPU" after having the complete residue known from the CPU can be easily done even by a 9-year old. Changing the order and submitting the "GPU" result first - by a clever 9-year old ...or by any 10-year old. It has been demonstrated that two matching results (with different shifts, and produced with unfettered Prime95 binary) are easily produced for any input. Maybe not by a 10-year old though...