Sierp base 16 - team drive #1

[tnerual]

Originally Posted by gd_barnes

Fast machine Karsten! I got 38-45 secs. testing time (vs. your 25 secs.) in that range when I LLR'd k=64518. The high k-value shouldn't have made any difference because LLR treats it like k=32259 since it's a multiple of 2, which is around the avg. sized k for this effort.

i don't know wich kind of processor you use but on my side, just before 30000 i got those times:

athlon X2-4200 :190 secs
centrino 1600 :280 secs
athlon 2500+ : 260 secs

karsten machine is 10 times faster than the centrino 1600 it's incredible ...
i want to know what to buy ...

[em99010pepe]

Originally Posted by tnerual

karsten machine is 10 times faster than the centrino 1600 it's incredible ...
i want to know what to buy ...

Intel Core 2 at 2.4 GHz but overclocked to 3.0GHz...got that timings, less to be accurate (23 sec), on the 35K-40K range.

[VBCurtis]

Nothing can touch an overclocked Core2 for LLR or sieving. I have a Core2-1866 running at 2850. It runs LLR at about P4-4400 speed. An Athlon-64 runs LLR at roughly a P4 of equal clockspeed (not rating), if not slower, so an X2-4200 would LLR at less than half the rate of an overclocked Core2.
If you read the hardware discussions elsewhere on this forum, they discuss memory-bandwidth issues for Core2Quads and the difficulty in using all 4 cores fully. For LLR with FFTs smaller than 256k or so, Carlos will confirm that all 4 cores run at full speed- our small searches are perfect for a Quad.

Quads *do* overclock, too. :)
-Curtis

[kar_bon]

ok guys,
i own a Core2Quad Q6600 at 2.4GHz, no overclocking.
it was a cheap offer i can't resist. but i read about overclocking up to 3.2GHz without problems!
the timings are quite good, because of LLR: reducing it to base 2 automatically and when k is divisible by 8 another fftlen smaller needed so the runtime was about 16s per candidate (only few ones).
for Riesel-base 6 k=1597 at n=153k it needs about 900s to complete one test.

[tnerual]

Originally Posted by VBCurtis

Nothing can touch an overclocked Core2 for LLR or sieving. I have a Core2-1866 running at 2850. It runs LLR at about P4-4400 speed.
-Curtis

yes but my P4-1600 takes about 600 secs at n=35000

so a P4-4800 would take 200 secs ... or would do your core2duo overclockerd ...

or it's 8 time faster (about 25secs) ... maybe it's because i use LLRNET in place of LLR ... will have a try to solve the mistery

edit:
LLRNET: 31347*16^36249+1 is not prime. RES64: C9DDFFA46F2BB6FB Time: 490.180 sec.
LLR 3.7.1C : 31347*2^144996+1 is not prime. Proth RES64: F25ECE59387BB41F Time: 138.144 sec.

timings are incredible ...

[em99010pepe]

LLRNET uses LLR 3.5.0.

tnerual,

Use the manual client and control all your machines with LogMeIn Free Edition.

[tnerual]

Originally Posted by em99010pepe

LLRNET uses LLR 3.5.0.

tnerual,

Use the manual client and control all your machines with LogMeIn Free Edition.

i will do that but i like the LLRnet way : just manage on one computer ...(this way i can put LLRNET on friends computer ... )

i will keep llrnet for non power of 2 ...

[kar_bon]

44k to 47k complete, no primes

reserving 50k to 53k

[mdettweiler]

Originally Posted by gd_barnes

Anon and Karsten, if you think it's worth your trouble, you can use srfile to remove k=19687 from your files before testing. I'll remove it from the remaining file links here shortly.

No problem, actually I'd already done that before I started.

[mdettweiler]

Originally Posted by tnerual

i will do that but i like the LLRnet way : just manage on one computer ...(this way i can put LLRNET on friends computer ... )

i will keep llrnet for non power of 2 ...

Yeah, I don't think there were any speed boosts for PRP tests (non power of 2) since LLR 3.5, so that shouldn't be a problem there.

Maybe we should bug whoever wrote LLRNet to make a new version based on LLR 3.7.1c? (Hopefully it wouldn't be too hard of a job, so it could be all ready by the time we've got our upcoming Conjectures 'R Us LLRNet server running!)

Edit: I posted a message about this over at the Riesel Sieve forums, since I think most of the development of LLRNet has been done there in the past. I asked how easy it would be to compile LLRNet with the LLR 3.7.1c source code--hopefully, if it's a relatively easy task, then either someone at RS will work on it, or, if they don't get around to doing it soon, maybe someone here could do it.

[axn]

Originally Posted by Anonymous

Yeah, I don't think there were any speed boosts for PRP tests (non power of 2) since LLR 3.5, so that shouldn't be a problem there.

Well, there has been a slight speed improvement after 3.5; nothing earth-shattering, about 5-7% (for base-2 as well as non-base-2).