Hey all. In preparation for the Winter Solstice Challenge (GFN21) in December, I petition everyone to sieve GFN21 before it! :)

Comparing current sieverate and testing speed on my GTX1070, I calculated that sieving GFN21 is currently 239 times more faster at excluding tasks from the project (counting in the DC) as running the testing. That's a years worth of GFN21 tests in one and a half day.

That said, it would be very desirable to sieve GFN21 as much as possible before the challenge, to increase our chances of finding a record prime worthy of the TOP 6th largest prime ever found! A world-first GFN21 find would also be the largest prime Primegrid has ever found!

I already see heavy hitters like tng* working on it, but we will need the help of everyone to sieve as deep as possible before the challenge :)

(I decided to post this in this subforum, since this is where challenge discussion is mostly had)

And also it's a very good credit generator! I manually sieved for the first time yesterday and got 12,900 credits for about an hour of crunching.

I also want to help. How can I do this sieving?

Sieving is 1.99 times as efficient as genefer at the moment. While there are a lot more candidates removed than that, they're not candidates we're going to test during the challenge. In fact the 1.99 number is based on the candidates we'll test over the next five years.

The numbers below were as of my last spreadsheet run on June 20, 2018. Next run is in late December.

GFN21 sieving is for b values from 1 to 2G. But our sieve file is for 1 to 100M, since that's the initial sieve we started with. The sieving results graph reflects this - it's as though we're sieving to only 100M. Since we've tested with genefer up to b=247160 (highest b then in progress), any candidate sieved under that value is not useful. The calculated leading edge for five years from now is b=1,005,325. So the current useful range is 1005325-247160=758165. That's 0.00758165 of the 100M sieving range.

We're generating 470/380=1.23684211 sieve removals per 1P sieved. And my benchmarks say that in the time genefer can test one candidate, we can sieve 106.122P. But of course genefer is run twice on every candidate. So we can remove 2*106.122*1.23684211*0.00758165=1.99 relevant candidates in the time it takes to run genefer

As far as affecting the number of candidates we'll actually test during the challenge, sieving is less effective as there's an even smaller range of candidates affected. So while sieving is a good thing, it's certainly not 200x as effective in the short term.

I also want to help. How can I do this sieving?

I also want to help. How can I do this sieving?

Myrskylyhty: very well explained.

[As you can see, there are a lot of steps and nyances related to manual sieving, so you need to be aware of what you're doing.

Yes, Radeon RX560 4GB

For how long PSA points will be pending?

Less than 240 000 P left now to sieve. I'll try to keep posting statistics every now and then so we can keep up with the speed :)

Were closing in on 230 000 P left. This implies a rough speed estimate of 5000P per day. If we keep up this speed, we can finish the sieve before the challenge! :)

Were at about 223 000 P left currently, so it seems to have slowed down a little. The 400P sized ranges can take about 3-5 days to finish so the number goes down in steps.

Avg. speed is at 3400P per day now. Luckily the sieve should speed up as we near the end.

Were at about 223 000 P left currently, so it seems to have slowed down a little. The 400P sized ranges can take about 3-5 days to finish so the number goes down in steps.

Avg. speed is at 3400P per day now. Luckily the sieve should speed up as we near the end.

203830P left! Rate is about 4250P/day.

We can do it! :)

I'm also getting full and optimal GPU usage on GFN21 sieve by just running one task (GTX1060 and GTX1070). On GFN17 sieve two tasks was optimal.

I think the sieve depth and maybe also the n affects the behaviour.

Back from my vacation, so I'll be back at it now.

We are at 182700P left now. The total speed has gone down a bit (3000P) per day, but were still advancing nicely. My total speed was something like 370P, so hopefully I can boost it back up a little.

Looks like tng* stopped sieving and is concentrating on Genefer tests now? I don't think we can finish the current range without him before the challenge starts.

167050P left, total speed since my 240000P left post is 3315P/day. With that speed we'd do 132600P before the challenge. So were gonna get close at least :)

Does anyone know if the sieve will speed up anymore when going deeper?

155161P left! Still at a steady 3500P/day speed.

Seems like we are now at under 19M candidates in the 100M range (in terms of just from sieving)! I guess that means were at over a 81 % sieve "depth".

155161P left! Still at a steady 3500P/day speed.

Seems like we are now at under 19M candidates in the 100M range (in terms of just from sieving)! I guess that means were at over a 81 % sieve "depth".

155161P left! Still at a steady 3500P/day speed.

Seems like we are now at under 19M candidates in the 100M range (in terms of just from sieving)! I guess that means were at over a 81 % sieve "depth".

No. In the 100M range, there are only 50M candidates (odd numbers only!). 19M remaining out of 50M means 31M factors were sieved out, or 62%. Not 81%.

We are now at around 100,000P left to reserve. That's 140,000P sieved in 5 weeks, which translates to 28,000P per week or 4000P per day. I think the rate has nicely gone up a little in the past few weeks.

With current rate it could be possible to sieve the whole range before the challenge!

Cheers and thanks everyone! Lets keep at it :)

As expected, the AP27 challenge put a dent in the sieving speed.

But now, we can still continue sieving for 20 days before the start of the challenge. Lets do our best!

Remember the server migration on Friday, and maybe reserve an extra range or two for manual sieving to not be left without work :) (Remember this also for your CPU projects!)

We're at 56400P and 13 days left. With the overall rate of 3612P/day, we are currently running a bit short, but lets do our best! Lets do a final push!

Anyhow it goes, we are currently sieved very deep for the challenge! I estimate that during the time of this thread until the challenge, we will have factored about 1% of the current remaining candidates out of the GFN21 project. It is not very large, but it all adds up!

If we were to test all those ~200,000 candidates we will have factored out, it would take 310 years on a GTX1070 and use up about 9 GWh of energy! That's roughly the amount 2,000 large households use up per year! By my estimations, sieving those candidates took about 1/100 of the work and energy than testing them would have! This efficiency however goes down the deeper we sieve. (Disclaimer: these are shaky estimations based on benchmarks on my own GTX1070).

Only 10920P left! We're now also under 18.9M remaining candidates, meaning that the sieve "depth" is now and extra step deeper; 62,2%/81,1%!

Nice work everyone! Thanks to all sievers, big and small. I think we'll easily finish before the challenge. Just remember to finish all your ranges everyone :)