[QUOTE=rogue;158857]It might have something to do with Geoff's build. If you can give me a couple of examples of where LLR is faster, I can investigate.[/QUOTE]

@Rogue:

LLR test: 3677878*3^342573-1 (2580.474 s)
Phrot test: 3677878*3^342656-1 (2751.30 s)

As you can see Geoff, though there is only an increase in n by 83 the time increases using Phrot with almost 171 seconds.

The next test and a bit higher, using LLR is still faster than the single mentioned Phrot test.

@Gary:
I know that there is a difference between Phrot and Proth. Unfortunantly I've had a hard time learning to spell to Phrot since I find it much easier for some reasons to spell Proth everytime I'm actually talking about Phrot :smile: But thanks for pointing out the differences, since I've actually done some testings with Proth, however I didn't seem to be able to make it work so I abandoned it again, however if Phrot appears to be faster at base 63 on sierp side, then I'll use it for n>1000 :smile:

Regards

Kenneth

[QUOTE=KEP;159008]@Rogue:

LLR test: 3677878*3^342573-1 (2580.474 s)
Phrot test: 3677878*3^342656-1 (2751.30 s)
[/QUOTE]

Can you tell me what phrot says WRT the line "Actually Testing"?

[QUOTE=rogue;159033]Can you tell me what phrot says WRT the line "Actually Testing"?[/QUOTE]

Input 3677878*3^342656-1 : Actually testing 24130557558*531441^28554-1 (witness
=3 28556/57344 limbs)

My own suggestion as to why it takes so considerablely longer to do a Phrot test compared to an LLR test may be that it relys to or has something to do with the k value. A k value of 24G must also have a high FFT boundry, so therefor it will be slower to test.

The problem of being slower using phrot appears to happen for base 63 also

k tested: 888
n tested: 25000

LLR time: 888*63^25000+1 is not prime. RES64: 9FE7B73F28056F20. OLD64: CD7C598FA045F9E4 Time: 127.730 sec.

Phrot time: Input 888*63^25000+1 is composite LLR64=9fe7b73f28056f20. (t=133.40s)

Input 888*63^25000+1 : Actually testing 55944*250047^8333+1 (witness=3 8334/184
32 limbs)

Hope it helps.

Kenneth

What CPU are you running this on? Is this Windows or Linux?

On a Core 2 Duo, your examples took about 1500 seconds with phrot and about 1600 seconds with LLR.

[QUOTE=rogue;159052]What CPU are you running this on? Is this Windows or Linux?

On a Core 2 Duo, your examples took about 1500 seconds with phrot and about 1600 seconds with LLR.[/QUOTE]

Running on Windows.

Base 3 tests were carried out on a Core 2 Duo while the base 63 tests were done on a Q6600 quad core 2.4 GHz.

Hope it helps!

Sierp. base 33 k=766 is now complete to n=100K. No primes. :sad: Results are attached.

Now proceeding with k=1678 and k=1818...

I wonder if it has anything to with running a 64-bit machine and O.S.

I run 64-bit Linux Ubuntu version 8.04 on all of my quads and only have a couple of slower 32-bit Window's machines. On base 12, Phrot, it is a full 40% faster than LLR on 64-bit Linux for exponents n>200K. I was exstatic when I first tested it.

But when I've tried either base 3 or any base on a 32-bit Windows machine, LLR is usually faster my a marginal amount like what Kenneth has experienced.


Gary

[quote=KEP;159007]@Gary:

Wow that was a lot of very insightfull answers. I'll try in a couple of months, to find a way to put the amount of candidates removed and remaining in to a spreadsheet for following n's:

n=1
n=2
n=3 to 4
n=5 to 8
n=9 to 16
n=17 to 32
n=33 to 64
n=65 to 128
n=129 to 256
n=257 to 512

By finding out how many candidates is removed and remaining for each doubeling in n, eventually a more accurate prediction will be able to be projectured for Sierp. base 63. However as you state, none of us is likely to ever know if our predictions for a highest completion n, is ever correct estimated.

But it will be a couple of months before I can make this kind of prediction.

Now I'll go dig up the n value for base 3 to Rogue, so he can start his investigation :smile:

Kenneth![/quote]


You don't need to take a couple of months to do it. Do it like I did. Just take a small subset like k<10000 and run it up to n=1000. (I did k<2000.) You don't have to run 30M+ k's to get a good estimate. :smile:

The 37% that I got is a little lower than what I was seeing at n=1000 (~39-40%) for k<2000 but the higher k's and n's have a lower % so it's best to estimate a little lower than what you see at the low k and n-ranges.

BTW, I prefer to use:
n=100-200
n=200-400
n=400-800
n=800-1600
etc.

It's too big of a pain to deal with so many k's and primes starting from n=1 and you don't need to do that to come up with a reasonable estimate.


Gary

[QUOTE=gd_barnes;159100]You don't need to take a couple of months to do it. Do it like I did. Just take a small subset like k<10000 and run it up to n=1000. (I did k<2000.) You don't have to run 30M+ k's to get a good estimate. :smile:

The 37% that I got is a little lower than what I was seeing at n=1000 (~39-40%) for k<2000 but the higher k's and n's have a lower % so it's best to estimate a little lower than what you see at the low k and n-ranges.

BTW, I prefer to use:
n=100-200
n=200-400
n=400-800
n=800-1600
etc.

It's too big of a pain to deal with so many k's and primes starting from n=1 and you don't need to do that to come up with a reasonable estimate.


Gary[/QUOTE]

Here is the removal stats for the k's <=1M (483871 candidates):

n k's rem. primes k's rem. % removed
1 483871 103654 380217 21,4%
2 380217 67082 313135 17,6%
4 313135 81042 232093 25,9%
8 232093 75792 156301 32,7%
16 156301 57897 98404 37,0%
32 98404 38437 59967 39,1%
64 59967 23771 36196 39,6%
128 36196 14346 21850 39,6%
256 21850 8553 13297 39,1%
512 13297 5039 8258 37,9%

Hope this can help someone come up with a good prediction :smile:

Kenneth

[quote=gd_barnes;159099]But when I've tried either base 3 or any base on a 32-bit Windows machine, LLR is usually faster my a marginal amount like what Kenneth has experienced.[/quote]
Hmm...that's odd. What version of Phrot are you using on 32-bit Windows?