An exchange with DJones

[DJones]

I would like to reserve the following k's:

k=65, k=67, and k=71

I'm intending to take these up to n=1m. This will be my first foray into Riesel searching, so fingers crossed that I get this right...

Edit: Any suggestions as to a ballpark sieving figure, gratefully accepted...

[VBCurtis]

Welcome to the search!

Since the three you've chosen have different starting n values, there's no benefit to sieving them together with sr2sieve. Use sr1sieve (scroll down in the forum for a thread "new sieving software" by geoff to get it).

You reserved k=65, but that one is done to 989k already, so you'd only be testing an 11k range if you intend to go just to 1M.

Optimal sieve depth depends heavily on which architecture you use for sieving. k=67 is rather low-weight, so sieving for just 4 or 5 trillion is probably enough; if you tell me what you're sieving on (and if it is 64bit OS if conroe or Athlon64), I can give you a pretty accurate sieve depth. I sieved 67 a little bit a while back, so I have some data to work with for your estimate.

k=71 is even lower weight, and only needs to be run from 736k to 1M, so 3T or so is very likely enough sieving on a fast-ish machine.

Happy hunting!
-Curtis

[DJones]

Originally Posted by VBCurtis

Welcome to the search!

Thanks.

Originally Posted by VBCurtis

Since the three you've chosen have different starting n values, there's no benefit to sieving them together with sr2sieve. Use sr1sieve (scroll down in the forum for a thread "new sieving software" by geoff to get it).

Ah, ok.

Originally Posted by VBCurtis

You reserved k=65, but that one is done to 989k already, so you'd only be testing an 11k range if you intend to go just to 1M.

My intention was to get these three to n=1M, and then grab 69 and 73 so I have a batch of five to run from 1M onwards. I didn't mention this originally as I didn't want to prevent anyone from reserving those before I'd finished with the current three.

Originally Posted by VBCurtis

Optimal sieve depth depends heavily on which architecture you use for sieving. k=67 is rather low-weight, so sieving for just 4 or 5 trillion is probably enough; if you tell me what you're sieving on (and if it is 64bit OS if conroe or Athlon64), I can give you a pretty accurate sieve depth. I sieved 67 a little bit a while back, so I have some data to work with for your estimate.

A humble little Athlon XP 2100+. I am deliberately holding back from a new computer as I don't run anything which requires anything beyond what I already have.

Originally Posted by VBCurtis

k=71 is even lower weight, and only needs to be run from 736k to 1M, so 3T or so is very likely enough sieving on a fast-ish machine.

Thanks for that. I'd best get cracking

[Kosmaj]

DJones

Welcome to RPS!

I wonder how many machines do you have altogether, besides your Athlon-2100? If that's the only one you have, the ranges you intend to test are going to take quite a while... If you are new to prime search can I suggest you to try for example k=67, say to 1M, sieving initially say to 100bn and then running some LLR tests to have a better feeling what kind of task you are dealing with. If you feel comfortable with exe times you can sieve more later, and test the whole range to 1M.

Happy hunting!

[DJones]

Originally Posted by VBCurtis

k=67 is rather low-weight, so sieving for just 4 or 5 trillion is probably enough

Originally Posted by Kosmaj

...try for example k=67, say to 1M, sieving initially say to 100bn

So, there's one person saying I should sieve to 4 or 5 trillion, and another saying 100 billion will be sufficient. Contradictory information like this is probably what's putting more people off joining this effort.
I'm inclined to go with VBCurtis's '4 or 5 trillion' (mainly because the sieve is already past the 100 billion point) but I'm willing to listen if anyone else has any more numbers they'd like to offer.

Originally Posted by Kosmaj

Welcome to RPS!

Thanks.

Originally Posted by Kosmaj

I wonder how many machines do you have altogether, besides your Athlon-2100? If that's the only one you have, the ranges you intend to test are going to take quite a while...

...and...? Are we in a rush to finish this infinite-size project so that we can get on with something else?

Originally Posted by Kosmaj

If you are new to prime search can I suggest...

I'd be more willing to listen to your suggestions if you had deigned to reply to my PM from some time ago.

Originally Posted by Kosmaj

[so that you] have a better feeling what kind of task you are dealing with. If you feel comfortable with exe times you can sieve more later, and test the whole range to 1M.

Having an idea of what kind of task I'm dealing with isn't going to affect the run times. In fact, I'd be quite happy with the run times regardless of what they are, as I'd rather my comp was doing something instead of nothing. I take the practical view that as there are fundamentally an infinite number of k's to be gone through, some contribution is better than none at all regardless of how long it takes. As you seem so concerned about my speed though, I can quite happily take my CPU cycles and go and do SETI or something if that's what you want.
It basically boils down to - would you rather I was doing these k's slowly or not at all?

[VBCurtis]

Hooo, boy. If you have just the one machine, getting those three to 1M will be more than a year-long project. I sieved 67 to 1B, and 13,220 candidates remain from 500k to 1M. Say there are 10,000 left after you run the sieve all the way. The average test will take ~20 min on a P4-2400, so we're talking 3,333 hours, or 20 weeks on a P4. I believe athlonXP's are about half the speed of P4 at same Mhz for LLR, so you're looking at 40 to 45 weeks of LLR just on 67 if you have just the one machine. On a non-ancient P4, you'll get all 3 k's to 1M in less than a CPU-year, give or take.

A standard estimate for sieving is 5% of the total project time, so sieving 24/7 for 14-16 days on your XP should be good for 67.
-Curtis

[DJones]

Originally Posted by VBCurtis

I believe athlonXP's are about half the speed of P4 at same Mhz for LLR

I take it from this that LLR, like Prime95, is geared towards working better on Pentiums then? Even so, I think half the speed is rather an underestimate.

[VBCurtis]

Combative a bit, are we?

Kosmaj & co. are very very used to people jumping into a project without a clue how big it is, and losing interest long before useful work is done. Most of these people do not bother to unreserve the work they set out to do, and we are left with orphan k's, which does nobody any good. For these people, SETI is indeed a better use of their CPU cycles, as workunits are quite small, and the progress easily measured.

His comments are intended solely to prevent such events from occuring. The vast majority who start a new project are not in the business of making multiyear commitments to said project. I'm not sure why you are so defensive, nor why these reasons for us warning you come as an attack rather than trying to inform you.

Further, if you re-read Kosmaj's comment, he suggested to sieve to 100B or so, then *play with* LLR to get a sense for testing times and project length. Nowhere did he say that depth was sufficient. His suggestion is what he and I (and I'm sure quite a few others) did to learn what we know about sieve depths, project lengths, etc. Again, I do not understand your snottiness about his wisdom.
-Curtis

[VBCurtis]

Originally Posted by DJones

I take it from this that LLR, like Prime95, is geared towards working better on Pentiums then? Even so, I think half the speed is rather an underestimate.

LLR uses predominantly the same routines as Prime95. I looked up the benchmark page on mersenne.org, picked a FFT size with round timing numbers, and compared a P4-2400 to AthlonXP-2400. The time was nearly exactly double. If you know otherwise, I suggest you post a new benchmark to George.

The software has significant SSE2 optimizations, which are not available to XP owners.
-Curtis

[DJones]

Originally Posted by VBCurtis

His comments are intended solely to prevent such events from occuring.

What his comments are intended to do and what they actually achieve are two different things. "I wonder how many computers you have" is not the same as "how many computers do you have?". The latter is a question whilst the former carries the implication "if it's only the one, why are you bothering?"
Telling me that I need to "understand what kind of task [I'm] dealing with" before I can make a useful contribution is nonsense. I don't need to understand everything about the maths to donate CPU cycles and type in a few commands at a DOS prompt.
Talking down to people like that is not going to encourage them to take part.

Originally Posted by VBCurtis

Kosmaj & co. are very very used to people jumping into a project without a clue how big it is

Oh, I see, so because a few people have chosen to help in the past without realising the scale of the project, that entitles Kosmaj to assume that no-one who chooses to take part understands the scale of the project.

Originally Posted by VBCurtis

I do not understand your snottiness about his wisdom.

His wisdom isn't the issue. It's his presentation. Like Dr. Silverman, Kosmaj may know what he's talking about but when he starts off on the offensive he just comes across as an arrogant berk. Yes, I was defensive. I usually am when I'm being attacked.
I appreciate Kosmaj probably did not intend to offend me but, again, intentions and results are not the same thing; all I can suggest is that he is more careful in future not just about what he says but also how he says it.

[Kosmaj]

DJones

Please don't waste our time.

SETI sounds great.