Sieving drive for k=2000-3400 n=50K-1M

[mdettweiler]

Originally Posted by gd_barnes

How should I send you the new file? ...or can you instruct me on uploading it to the server? No need to do that in this thread. You can send me a PM or Email. If it's a big hassle, I'm fine with Emailing you the file and you uploading it if you don't mind.

Hmm...well, the process of uploading it is rather geek-intensive, though I suppose I could put together some instructions for you. It's up to you--if you think it would be easier to just have me upload it, you can go ahead and send it to me via email. (Feel free to break it up into multiple chunks if needed.)

[gd_barnes]

Originally Posted by mdettweiler

Hmm...well, the process of uploading it is rather geek-intensive, though I suppose I could put together some instructions for you. It's up to you--if you think it would be easier to just have me upload it, you can go ahead and send it to me via email. (Feel free to break it up into multiple chunks if needed.)

I'm all for avoiding geek hassle. :-) I'll Email you the file with factors removed up to P=5050G a little later tonight. You can upload it to the server when you have time.

[Lennart]

Reserving 18T - 19T Lennart ETA Jun 11

[Svenie25]

Absolutely impressing, what a performance the people here are showing for this sieve, particularly Lennart. Respect to all!

[Flatlander]

Originally Posted by gd_barnes

...
With Lennart's excellent boost, I'll suggest that we now go to P=25T and will change corresponding info. in the 1st post. Once again, if it doesn't look like we'd make that by the 20th, we won't do it.
Gary

It looks like we will hit 25T before 20th June. (Assuming Lennart's continued amazing help.)
I expect you've already done the math and got the t-shirt but, why not 'over' sieve a bit. These will be LLRed twice after all. (Double checking. Albeit with probably faster cores than the average now.)

[Lennart]

Originally Posted by Flatlander

It looks like we will hit 25T before 20th June. (Assuming Lennart's continued amazing help.)
I expect you've already done the math and got the t-shirt but, why not 'over' sieve a bit. These will be LLRed twice after all. (Double checking. Albeit with probably faster cores than the average now.)

I can say that on 16.9T removal rate is 25sec/f on 4 core thats 100sec on 1 core.

That tells me 25T is to low for a singel check.

And if we count in doublecheck we will not oversieve even if we get to 50T

Keep sieving ! Save time.

/Lennart

[gd_barnes]

It's better if we start the drive sooner because of the lack of choices for lower top-5000 primes now. We'll still be sieved further than any other drive that we've had for n<=600K. I'd actually rather start the drive on June 15th. Based on that, here's a note to all: Don't reserve any sieving range that will take past June 15th at this point. If we're not at P=25T by then, then we can continue reserving ranges up through the 20th. Actually another new drive can't start soon enough as far as I'm concerned. k=2000-3000 and k=3000-3400 to a lesser extent has turned into almost as big a mess as k=1000-2000 was 6 months ago.

Later on, we'll need to sieve the P=600K-1M range to P=50T or higher but that will be very low priority. If people want, we could leave the entire n=50K-1M range in that sieve.


Gary

[gd_barnes]

Originally Posted by Lennart

I can say that on 16.9T removal rate is 25sec/f on 4 core thats 100sec on 1 core.

That tells me 25T is to low for a singel check.

And if we count in doublecheck we will not oversieve even if we get to 50T

Keep sieving ! Save time.

/Lennart

I'll do another calculation after we get all factors removed up to about P=15T. We're only planning to LLR with this sieve depth up to n=600K. Because of that, having the removal rate equal the LLR time of a candidate around n=450K would be close to optimal; although I may bump that up to n=500K since our main drive will initially be something like n=420K-600K.

We'll likely sieve to P>50T for LLRing n>600K.

Your machines appear faster than most of ours. Mine are at about P=126000/sec at P=5.5T. What sieve rate are you getting? That infuences the calculation quite a bit.

We won't double check such a large range for many years. We can sieve further when machines are faster in a few years. I'm a big proponent of taking into account future computer speed increases when sieving huge tasks and therefore tend to err on the side of undersieving slightly now. Many times what you are sieving won't be able to be tested for as long as 2-3 years.


G

[Lennart]

Originally Posted by gd_barnes

I'll do another calculation after we get all factors removed up to about P=15T. We're only planning to LLR with this sieve depth up to n=600K. Because of that, having the removal rate equal the LLR time of a candidate around n=450K would be close to optimal; although I may bump that up to n=500K since our main drive will initially be something like n=420K-600K.

We'll likely sieve to P>50T for LLRing n>600K.

Your machines appear faster than most of ours. Mine are at about P=126000/sec at P=5.5T. What sieve rate are you getting? That infuences the calculation quite a bit.

We won't double check such a large range for many years. We can sieve further when machines are faster in a few years. I'm a big proponent of taking into account future computer speed increases when sieving huge tasks and therefore tend to err on the side of undersieving slightly now. Many times what you are sieving won't be able to be tested for as long as 2-3 years.


G

Removal rate @25sec/f at 4 core is on a stock q6600@2,4Ghz.
Removal rate on a Q6850@3,2Ghz is 19sec/f on 4 core
i7@3.5Ghz 16 sec/f on 4 core.

Thats my time :)


EDIT: I started a range on 18T on WinXP64 it is a stock q6600@2,4Ghz i get 168kp/sec on that.
On a Q6850@3,2Ghz i have 237kp/sec
On a i7@3,5Ghz 275kp/sec

But i compare LLR time and sieve time on one computer so i have the same speed on LLR & sieve.

This is on P=16T-18T

Lennart

[gd_barnes]

Thanks for the figures Lennart. Very fast machines!

I'll say this: If people would like to continue sieving through the 20th, if we can get it up to P=30T by then with Lennart's assistance, I'm OK with that assuming that the optimum depth is at least that high. I'll do a new calculation on June 12th. All ranges up to P=18T except Karsten's range will be done by then so removing all of those factors to make the determination will be good enough.

In the mean time, I'm holding off on sending the new file to Max with all factors removed up to P=5050G. There was a small range missing out of one of them. I should get those factors late tonight and will send the new sieve file to him at that time for uploading to the noprimeleftbehind server.


Gary

[gd_barnes]

An updated file with all factors removed up to P=5050G has now been sent to Max for uploading to the server.