[QUOTE=axn;333150]With the speed at which Titan could complete a test, that might not be necessary.[/QUOTE]

That might not, indeed... but the chance is not low, and... who would double-check the result? :smile:

Luigi

[QUOTE=ET_;333153]who would double-check the result? :smile:[/QUOTE]

Another Titan, naturally :smile: Nothing else is fast enough.

2000h

I may be wrong, but I think at least one "trusted" result is needed for an exponent to be considered "double checked." PrimeNet treats manual results as "untrusted."

[QUOTE=axn;333150]With the speed at which Titan could complete a test, that might not be necessary.[/QUOTE]
Consider [URL="http://mersenneforum.org/showthread.php?p=333018#post333018"]this[/URL]. Makes a lot of sense to me.

The speed is immaterial, but the quality of marksmanship that is deemed to be acceptable in the gamers' GPU field is important. The GPU makers bin chips between computer grade (Tesla) and "good-enough-for-gamers" (Titan).
[URL="http://mersenneforum.org/showthread.php?p=333018#post333018"][/URL]

[QUOTE=Batalov;333199]The speed is immaterial, but the quality of marksmanship that is deemed to be acceptable in the gamers' GPU field is important. The GPU makers bin chips between computer grade (Tesla) and "good-enough-for-gamers" (Titan).
[URL="http://mersenneforum.org/showthread.php?p=333018#post333018"][/URL][/QUOTE]

What is the chance that a GPU with bad memory will have the same residue?

[QUOTE=chalsall;333201]What is the chance that a GPU with bad memory will have the same residue?[/QUOTE]

It should be the equal in respect of two bad LL-tests done with a CPU and having the same final residue. Unless the problem is in the hardware design... or the two tests are submitted by the same user, that's the same chance.

Luigi

[QUOTE=chalsall;333201]What is the chance that a GPU with bad memory will have the same residue?[/QUOTE]

64-bit residue so 1 in 2^64 of false match

[QUOTE=henryzz;333203]64-bit residue so 1 in 2^64 of false match[/QUOTE]
Surely you were joking?

[QUOTE=ixfd64;333198]I may be wrong, but I think at least one "trusted" result is needed for an exponent to be considered "double checked." PrimeNet treats manual results as "untrusted."[/QUOTE]

You are wrong. The issue here is bitshift-edness, which non-testing versions of CuLu don't yet do. (Mostly my fault at the moment.)

[QUOTE=ET_;333202]It should be the equal in respect of two bad LL-tests done with a CPU and having the same final residue. Unless the problem is in the hardware design... or the two tests are submitted by the same user, that's the same chance.[/QUOTE]

Exactly.

I would argue that CUDALucas has proven itself to be as trustworthy as Prime95/mprime.

Primenet should accept as proven a LL and a matching DC done by CUDALucas, but should (of course) reject matching tests done by the same user.

Or, alternatively, George could make a policy decision that only once a candidate has been tested by Prime95/mprime, with a matching residue from CUDALucas (by different users) is the candidate considered proven to be composite.