Msieve NFS minimum size - Page 3

10metreh · 2009-01-26, 18:38

Quote:

Originally Posted by henryzz

i thought that was the whole point of this thread

I mean, it switched to QS even with the NFS switch.

henryzz · 2009-01-26, 19:01

Quote:

Originally Posted by 10metreh

I mean, it switched to QS even with the NFS switch.

the last binary does that
i will compile again in a few days

Joshua2 · 2009-04-02, 05:00

I found out that for c98 gnfs w/ msieve 1.40 poly select and post proc w/ 32 bit core2 sieve is around 30% faster than c98 w/ yafu 1.08 x64 (and I don't even have 64 bit asm). So the 97 digit thing should be lowered to like 94 so we can test for ourselves. Is that possible to do for the next version?

jasonp · 2009-04-02, 13:52

v1.41 will be released this evening. For v1.42 I want to try adding degree 4 polynomials to the poly selector, and that should hopefully allow NFS to be faster even below 94 digits.

10metreh · 2009-04-02, 14:33

Quote:

Originally Posted by jasonp

v1.41 will be released this evening. For v1.42 I want to try adding degree 4 polynomials to the poly selector, and that should hopefully allow NFS to be faster even below 94 digits.

What about degree 6 for the >C180s?

Andi47 · 2009-04-02, 15:11

Quote:

Originally Posted by 10metreh

What about degree 6 for the >C180s?

If anyone has got such a thing to throw at the c180+'s....

jasonp · 2009-04-02, 15:17

Quote:

Originally Posted by 10metreh

What about degree 6 for the >C180s?

Stage 1 is easy to extend to arbitrary polynomial degree, but stage 2 (the pol51opt equivalent stage) is highly specialized for degree 5. Adding other polynomial degrees to it in a clean way will involve a lot of rearranging. Plus we have to figure out appropriate parameters for everything new.

Actually you're using the SNFS scale to decide when degree 6 is appropriate. I don't know if degree 6 would be useful for GNFS even to the limit of what msieve's postprocessing can do.

Andi: Does your link mean you're volunteering to port the lattice siever to a GPU?

Also, I really hope one GPU in the middle of that cluster doesn't fail...

fivemack · 2009-04-02, 16:18

Has anything yet become of the CADO suggestion to put one undergraduate-project-year into figuring out if you could do any sieving at all on Cell or on a GPU?

http://www.ddj.com/hpc-high-performa...ting/216402188 makes it pretty clear what I'll be buying at Christmas

though from some of the remarks about things which haven't been finalised yet, it's not clear if it will be Christmas 2009 or 2010.

FactorEyes · 2009-04-02, 16:28

Quote:

Originally Posted by fivemack

http://www.ddj.com/hpc-high-performa...ting/216402188 makes it pretty clear what I'll be buying at Christmas

though from some of the remarks about things which haven't been finalised yet, it's not clear if it will be Christmas 2009 or 2010.

When you find that under the tree, will you send me that i7 920 you've been running? I'll give it a good home, with its own water cooling block and a nice 3.8GHz overclock, and plenty to do all day.

Andi47 · 2009-04-02, 16:30

Quote:

Originally Posted by jasonp

Andi: Does your link mean you're volunteering to port the lattice siever to a GPU?

Also, I really hope one GPU in the middle of that cluster doesn't fail...

No, this link was more or less jokingly / saying that you might need a similar sized cluster.

And I am absolutely NOT familiar with GPUs and such things...

fivemack · 2009-04-02, 16:45

I was rather intending to put the Larrabee into one of the slots of the X58 board with the i7 920 in it, though presumably by that point the i7 920 will be a laughably obsolete piece of last-generation technology, there mostly to provide the reasonably quick memory controller through which the Larrabee accesses the large memory with the matrix in it, keeping the vector in its on-board fast memory.

(I will be at least somewhat vexed if Westmere processors don't fit on Nehalem X58 motherboards, given how ridiculously much the motherboards cost, but I suppose I should ready myself for mild vexation)

A graphics card that can run BSD and provide the GPU behaviour as a BSD process is really quite alarmingly capable!

2009-04-02, 05:00	#25
Joshua2 Sep 2004 215₁₆ Posts	I found out that for c98 gnfs w/ msieve 1.40 poly select and post proc w/ 32 bit core2 sieve is around 30% faster than c98 w/ yafu 1.08 x64 (and I don't even have 64 bit asm). So the 97 digit thing should be lowered to like 94 so we can test for ourselves. Is that possible to do for the next version? Last fiddled with by Joshua2 on 2009-04-02 at 05:03

2009-04-02, 16:18	#30
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 6419₁₀ Posts	Has anything yet become of the CADO suggestion to put one undergraduate-project-year into figuring out if you could do any sieving at all on Cell or on a GPU? http://www.ddj.com/hpc-high-performa...ting/216402188 makes it pretty clear what I'll be buying at Christmas though from some of the remarks about things which haven't been finalised yet, it's not clear if it will be Christmas 2009 or 2010. Last fiddled with by fivemack on 2009-04-03 at 09:44

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Estimating minimum relations	bchaffin	Factoring	24	2012-03-24 18:37
What is minimum RAM needed for 6 Core CPU for P-1?	odin	Hardware	15	2010-04-18 14:22
Minimum/desired CPU specs for ECM factoring	Kaboom	PrimeNet	10	2009-04-17 14:58
Minimum primes of various forms database?	jasong	Information & Answers	1	2007-11-01 01:58
Minimum delay between server connections	vaughan	ElevenSmooth	5	2005-09-08 17:17

2009-04-02, 13:52	#26
jasonp Tribal Bullet Oct 2004 3,541 Posts	v1.41 will be released this evening. For v1.42 I want to try adding degree 4 polynomials to the poly selector, and that should hopefully allow NFS to be faster even below 94 digits.

2009-04-02, 16:45	#33
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 14423₈ Posts	I was rather intending to put the Larrabee into one of the slots of the X58 board with the i7 920 in it, though presumably by that point the i7 920 will be a laughably obsolete piece of last-generation technology, there mostly to provide the reasonably quick memory controller through which the Larrabee accesses the large memory with the matrix in it, keeping the vector in its on-board fast memory. (I will be at least somewhat vexed if Westmere processors don't fit on Nehalem X58 motherboards, given how ridiculously much the motherboards cost, but I suppose I should ready myself for mild vexation) A graphics card that can run BSD and provide the GPU behaviour as a BSD process is really quite alarmingly capable!