Sieving for CRUS

[gd_barnes]

Originally Posted by VBCurtis

Nice work! This is a useful sieve, hardly something to apologise for.

Gary, I'd like Storm5510 to feel like their job is complete- if you decide this needs more sieving, I'll be happy to run it to at least your desired level.

I ran a single instance of a sieve and LLR test on this using CPU software on an 8 core/8 thread I7-9700 machine with nothing else running in the background.

For the sieve, using sr2sieve's expected factors found, which is much more accurate than the actual factors found for determining removal rate, it came up with a removal rate of one every 332 secs.

For the LLR test, I ran it at 60% of the n-range on about a median k: 10068*897^189983-1 to be specific. After running for a couple minutes, the single iteration rate was 0.88 ms. Total iterations was ~1863550. So expected total test time was ~1640 secs.

So a rough estimate of optimum sieve depth for this machine is somewhat less than 1640 / 332 = 4.94 multiplied by the current sieve depth of 15e12. That comes out to ~74e12.

That method is always somewhat high until you get within roughly 20% of the optimum. I'm guessing optimum will be ~65e12 to 70e12 at this point.

I would sieve to 60e12 and then do the calculation once more.

If you have a GPU-based sieve process, you may want to use your removal rate instead.

I'm not asking you to do this. I can just post it as a partially sieved file.

My suggestion if you feel like running a sieve on a file that needs more sieving: Run R60. The file is on the reservations page. That was Storm's first effort and it was only sieved to 1e12. It is for the same n-range of 100k-250k. At 1e12, it needs more small P-range sieving than R897 does at this point but overall will likely be a smaller job. It's a much smaller base with fewer k's remaining so optimum sieve depth will be lower and it will likely get some testing interest in the near future once it is fully sieved.

[storm5510]

Originally Posted by gd_barnes

The file has 284,871 terms in it, which is a lot more than your stated ~249,000 here. I checked above and you had 288,785 terms remaining at 1e13. So 284,871 seems reasonable for a 15e12 sieve and you made a misstatement here.

Let me know if I missed anything.

You used the contents of the sieve so you did not miss anything. Sorry for fudging the numbers. I wasn't feeling so great when I wrote it. Much better today. I rather enjoyed doing this while it lasted. It gave me something to think about and to learn.

Give me a few days and I might pick out something else to take a crack at.

Thank you to everyone for your kind words.

Update: If no one is running R42, I will take it. The reservation page indicates it has been ran to N=200e3. I will run it to 300e3. I will wait a day to start just in case anyone recommends any modifications.

[gd_barnes]

Originally Posted by storm5510

<snip>
Update: If no one is running R42, I will take it. The reservation page indicates it has been ran to N=200e3. I will run it to 300e3. I will wait a day to start just in case anyone recommends any modifications.

The word "run" in this context is ambiguous for our project. It usually means someone is going to test (LLR) a base. I assume you mean you are going to only sieve it. Is that correct?

I have a better idea: Pick up your partial sieve for R60 off of the reservations page and use it as input to your process to sieve it more. It's only sieved to 1e12 and needs plenty more. The optimum sieve depth for it is likely quite a bit less than for R897.

As for R42, if you really want to sieve it instead, please sieve 200e3 to 500e3. It is inefficient to sieve a range where the nmax is less than double the nmin. We usually go for more than double or even triple or more. We have often sieved 100e3 to 500e3 for files that we send to Yoyo since they can sieve so deeply very quickly.

[storm5510]

Originally Posted by gd_barnes

The word "run" in this context is ambiguous for our project. It usually means someone is going to test (LLR) a base. I assume you mean you are going to only sieve it. Is that correct?

I have a better idea: Pick up your partial sieve for R60 off of the reservations page and use it as input to your process to sieve it more. It's only sieved to 1e12 and needs plenty more. The optimum sieve depth for it is likely quite a bit less than for R897.

As for R42, if you really want to sieve it instead, please sieve 200e3 to 500e3. It is inefficient to sieve a range where the nmax is less than double the nmin. We usually go for more than double or even triple or more. We have often sieved 100e3 to 500e3 for files that we send to Yoyo since they can sieve so deeply very quickly.

Actually, I could do both at the same time. R42 and R60. If you remember, I have multiple desktops here. 500e3 it will be for R42.

I will refrain from using "run" in the future.

[gd_barnes]

Originally Posted by storm5510

Actually, I could do both at the same time. R42 and R60. If you remember, I have multiple desktops here. 500e3 it will be for R42.

I will refrain from using "run" in the future.

Sieving both would be great!

[VBCurtis]

I'll give Storm a bit of time to decide what he wants to sieve next before I jump on R897. If R60 or R42 are short-ish, he may wish to resume R897 in a few weeks.

[storm5510]

Originally Posted by VBCurtis

I'll give Storm a bit of time to decide what he wants to sieve next before I jump on R897. If R60 or R42 are short-ish, he may wish to resume R897 in a few weeks.

@VBCurtis: Go ahead and pick up R897 if you wish. My plate will be full for a while yet.

[gd_barnes]

I posted the R897 sieve on the reservations page with a note that it is partially sieved. So it can be picked up again anytime or not at all for a long time. :-)

[storm5510]

Originally Posted by gd_barnes

I posted the R897 sieve on the reservations page with a note that it is partially sieved. So it can be picked up again anytime or not at all for a long time. :-)

About the above. You are changing some of these from ABCD format to plain text. This removes the download capability, at least with Edge. It opens them to read. I have yet to find a "Select All, Copy, and Paste" ability in Edge. If it is there, then it is really buried somewhere. Zip or 7z might be a better way to go.

Just a thought...

[VBCurtis]

Originally Posted by storm5510

I have yet to find a "Select All, Copy, and Paste" ability in Edge.

ctrl-a , ctrl-c , ctrl-v

[kar_bon]

Originally Posted by storm5510

I have yet to find a "Select All, Copy, and Paste" ability in Edge.

RMB -> Link save as : possible for both ZIP and TXT as usual like other browsers, too.