PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
21) Message boards : Wieferich and Wall-Sun-Sun Prime Search : Any idea on WU times CPU vs. GPU? (Message 145636)
Posted 8 days ago by Profile Michael GoetzProject donor
Maybe off-topic, but when is it planned to start exactly?


As soon as possible after the current challenge ends. We want to give people as much time as possible to run WW before they switch their GPUs over to GFN for the December challenge. If there's no unexpected problems, it will probably happen within a few days of the challenge ending. Or a few hours. It depends on if there are any unforeseen problems.
22) Message boards : Number crunching : Magellan 500th Anniversary Challenge (Message 145567)
Posted 10 days ago by Profile Michael GoetzProject donor
Somebody here or on Discord wanted to know when we passed n=20M on the challenge tasks. Woodall has now passed 20M. The highest that's been sent to a user so far is 20010019.

The highest Cullen task sent out so far is n=19537253.
23) Message boards : Number crunching : Magellan 500th Anniversary Challenge (Message 145562)
Posted 11 days ago by Profile Michael GoetzProject donor
am I seeing things, am I doing it wrong?


He’s got a full rack of servers in his apartment, with custom BOINC software and a custom version of LLR to get them to run without disk drives. They talk to the PrimeGrid server through some small gateway computers. Those gateways are what you see in his host list. They are not the computers doing the actual crunching.

He is also the person who wrote our LLR2 software.

24) Message boards : News : 321 Sieve is over (Message 145538)
Posted 12 days ago by Profile Michael GoetzProject donor
Should the 321 sv label on the front page be deleted?


Eventually.
25) Message boards : Proth Prime Search : pps_sr2sieve no longer recognised ? (Message 145533)
Posted 12 days ago by Profile Michael GoetzProject donor
My PC went weird and when I re-opened BOINC I got this ?

PrimeGrid: Notice from BOINC
Your app_config.xml file refers to an unknown application 'pps_sr2sieve'. Known applications: 'genefer18', 'llrMEGA', 'llrDIV', 'llrTPS'
22/11/2020 00:21:31


You can safely ignore that message.
26) Message boards : Generalized Fermat Prime Search : DO YOU FEEL LUCKY? (Message 145469)
Posted 13 days ago by Profile Michael GoetzProject donor
How soon will I know...


Sorry to burst the bubble, but you didn't hit the jackpot.

Generally speaking, you would know pretty quickly. Probably before the double checker even finished running his task.

We would see that a DYFL task returned a "probable prime" result, and would immediately start testing for primality. We have two ways of doing that, one on a GPU, and one on a CPU, and depending on what hardware we have available at the time, it might take a day or more for that test. But once that's done, you would be notified by email.


27) Message boards : Generalized Fermat Prime Search : DO YOU FEEL LUCKY? (Message 145394)
Posted 14 days ago by Profile Michael GoetzProject donor
I'm going to say something because I'm the person that originally came up with the rules PrimeGrid currently uses for determining the "optimal depth" for the GFN sieves.

We had a problem with this particular sieve. Originally, we had a sieve program that spit out all the factors for candidates under b=100M. That was fine and good, until we learned that the 100M was a completely arbitrary number. The code contained a line that was something like this:

if (b < 100000000) { printf ("%i | %i^^i+1\n", f, b, n); }


The program was actually finding factors up to b=2B -- 20 times higher than what it was outputting. The program was telling us about only 5% of the factors it was finding.

Subsequently, we changed the output to print all of the factors, but that produced a new problem.

All of a sudden, the exact same sieve program, running on the exact same computer, was magically 20 times more efficient. It was, after all, producing 20 times as many factors. Overnight, the optimal sieving point for every GFN sieve went up by a factor of 20.

That simply made no sense at all.

While we were technically "finding more factors" than before, those extra factors weren't actually useful to anyone, at least not any time soon. As far as actually removing candidates so we don't have to run them through Genefer, the extra factors changed nothing. Maybe our great great great grandchildren might benefit from those factors, but we wouldn't.

While it did make sense to record the factors -- and why shouldn't we, disk space is cheap -- it just didn't make sense to include those factors in our decision about when to stop sieving.

I suggested that we therefore guess at the number of candidates we would test, with Genefer, over the next five years. We extrapolate where b will be in 5 years, and count the candidates between the current b and our extrapolated b. Then we look at the number of factors the sieve is finding between those two b values. If the sieve is finding more than one factor in that range in the time that Genefer takes to test one candidate, then we keep sieving.

This is different than other sieves because with all other sieves, as far as I know, when you increase the range over which the sieve is running, the sieve slows down. With the GFN sieve, the speed is mostly independent of the range you're sieving, which effectively makes the speed of the sieve completely arbitrary. As a result, the normal logic for computing the optimal sieve point doesn't work for the GFN sieve.
28) Message boards : Number crunching : macOS's Rosetta 2 only loses about 20% performance emulating x86 (Message 145304)
Posted 16 days ago by Profile Michael GoetzProject donor
Right now, the only non-AVX app we have is PPS-Sieve


The AP27 app can run without AVX; it has an SSE mode which can be activated using either the -sse2 or -sse41 flags. But that requires app_config.xml, which can't be accessed on Android unless the phone is rooted. (and that's assuming the Android client actually reads config files)
Could be fine for ARM Macs though, as SSE will likely be emulated in Rosetta 2 using ARM's NEON instruction set.


Of course a non-AVX version of the apps can be made for ARM processors, but it would be much slower. Do you have an old computer from about 10 years ago that predates the Sandy Bridge AVX CPUs? It will be like running on that.

Phone/pads are dead in the water because BOINC doesn't run on iOS and we have no Android developers. Even when we *did* have an app that was suitable for Android devices, we couldn't do it because we couldn't get anyone to work on the app.

Unless something radical changes, the only question is whether we're going to support Apple's ARM CPU under MacOS.
29) Message boards : Proth Prime Search : DC Task (Message 145279)
Posted 17 days ago by Profile Michael GoetzProject donor
The admins are experimenting with LLR2 on PPS. They unfortunately are only discussing such matters on the Discord rather than sharing the information on the forums.


I strongly disagree! There’s an entire thread here discussing LLR2, and as was mentioned, the DC task links was an idea that originated on the forums.

While it is true that switching PPS to LLR2 wasn’t announced in the forums, it wasn’t announced in Discord either. It should have been. We’ll try to do better.

Sorry, "such matters" was referring to the experiment of LLR2 on PPS, not the whole effort. The main thread that you referenced should have been updated more proactively as more changes have been made, and especially once users pointed it out in Discord.


I agreed with this. :)
30) Message boards : Number crunching : Magellan 500th Anniversary Challenge (Message 145276)
Posted 17 days ago by Profile Michael GoetzProject donor
Do I need to select only CUL-LLR and/or WOO-LLR and unselect PPS-SV for the 10-day duration to make sure I only participate in the challenge?


PPS-Sieve on the CPU should be unchecked. If you want to run PPS-Sieve on your GPU, that's fine, although it might impact CPU performance, depending on your computer.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.96, 3.70, 3.81
Generated 4 Dec 2020 | 11:15:11 UTC