[QUOTE=rebirther;405093]If you want to finish it faster we could add some ranges into BOINC.[/QUOTE]

These tests would be too quick for BOINC. Besides, BOINC will be busy with R63. :grin:

[QUOTE=KEP;405092]On an Intel(R) Core(TM) i5-2300 CPU @2.8GHz it takes about 100 days to test a 1G range to n=2500 using -f0

This will leave about 75000 k's remaining per G, wich will have to be sieved and tested, but I think that such an approach will make you do a 1G range per core, in about 4 months total AVX CPU time.

Since I'm always using -f0 and n=2500 for b=3, I can't tell you how long it will take if you use -f30 (wich Gary has recommended in the past).

On a sidenote. I have begun working the range from:

k=40,000,000,002 to k=63,064,644,936

Here is the overall plan:

Step 1: Test all 11,532,322,468 k's to n=2500 using Starting base script with -f0 and using PFGW version 3.7.9
Step 2: Sieve remaining k's from n=2501 to n=7500 and test using PFGW version 3.7.9
Step 3: Sieve remaining k's from n=7501 to n=25000 and test using PFGW version 3.7.9

I expect step 1 to complete somewhere around February and afterwards that step 2 and 3 will complete in march and april 2016 (maybe sooner)[/QUOTE]

How many cores do you have for step 1? My cores won't be available until August (unless my luck with S328/R328 improves), so I'm just planning ahead.

Also, Puzzle-Peter estimated 8 to 10 weeks per range of 1G on a single core. Which is right?

[QUOTE=rogue;405105]How many cores do you have for step 1? My cores won't be available until August (unless my luck with S328/R328 improves), so I'm just planning ahead.

Also, Puzzle-Peter estimated 8 to 10 weeks per range of 1G on a single core. Which is right?[/QUOTE]

This was from memory, I might be off quite a bit. But I noticed that KEP uses -f0 whereas I use -f30. Trial factoring to 30% of the default value does not use a lot of time, but it quickly eliminates all those candidates with small factors.

[QUOTE=rogue;405105]How many cores do you have for step 1? My cores won't be available until August (unless my luck with S328/R328 improves), so I'm just planning ahead.

Also, Puzzle-Peter estimated 8 to 10 weeks per range of 1G on a single core. Which is right?[/QUOTE]

I'm for the next 4 weeks going to use 3 cores on my Sandy Bridge (the mentioned computer) and from around august 5th 2015 I'm going to add 4 cores from my Quad core Hasswell. So from August 5th, I'm going to have a total of 7 cores available to step 1.

I doubt that 8 to 10 weeks per range of 1G on a single core is correct, even when using -f30 and running from n=1 to n=25K, because on my systems it has always been much more inefficient to run with -f30 than it was to run according to the plan I have listed above. I guess that running with -f30 on an AVX quad core is propably realistical with 8-10 weeks, but I've no idea. Guess what, the range from k=43G-44G I could try on my Hasswell to use -f30 and see how long it actually will take on a single core. If PuzzlePeter is correct, we will know around 1st of November and if PuzzlePeter is wrong we will also know around 1st of November :smile:

[QUOTE=rebirther;405093]If you want to finish it faster we could add some ranges into BOINC.[/QUOTE]

No Reb, even though I like your hard work, I doesn't want to kill your server or burden you with loads of administration. You will at least not with the current configuration of your BOINC server and feeders, be able to keep up with the demand for work, simply because each WU will take about 0.03 (or less) seconds per k/n tests (at n=2500), so if your server has to be able to keep up, you have to send out WU's with thousands of tests per WU and that will give you other problems you might not desire to have. Maybe from n>25K BOINC can be utilized in some way, but that is still at least 1 or 2 years into the future, dependent on how many others join the struggle with getting R3 started :smile:

Take care

KEP

[QUOTE=Puzzle-Peter;405107]This was from memory, I might be off quite a bit. But I noticed that KEP uses -f0 whereas I use -f30. Trial factoring to 30% of the default value does not use a lot of time, but it quickly eliminates all those candidates with small factors.[/QUOTE]

I think you will be quite far off, but we will know once I get the range k=43,000,000,002 to k=44G started :smile:

[QUOTE=KEP;405111]No Reb, even though I like your hard work, I doesn't want to kill your server or burden you with loads of administration. You will at least not with the current configuration of your BOINC server and feeders, be able to keep up with the demand for work, simply because each WU will take about 0.03 (or less) seconds per k/n tests (at n=2500), so if your server has to be able to keep up, you have to send out WU's with thousands of tests per WU and that will give you other problems you might not desire to have. Maybe from n>25K BOINC can be utilized in some way, but that is still at least 1 or 2 years into the future, dependent on how many others join the struggle with getting R3 started :smile:

Take care

KEP[/QUOTE]

Yeah, 15-25k would be ok :)

Kenneth,

Why do you continue making such large reservations? The files would be huge. I've asked before for you to reserve smaller ranges, complete them, reserve the next range, etc. I have reserved k=16-20G to n=25K. My suggestion is for you to reserve in 4G ranges max. You could reserve P=20G-24G, finish it, reserve P=24G-28G, finish it, etc. In other words, don't try to reserve half of the remaining k's at once. I think your estimates are way too optimistic because of the boredom factor.

PFGW with the -f30 factoring switch and no sieving is the way to go up to n=25K on this base. On my 4G range, I just have one 4-core old slower machine set to run it non-stop with no sieving (i.e. factoring to 30%). It will take 4-6 months (1G per core) and I don't ever have to touch the machine except to verify that it is still running. Compare that to you taking 4 months on a modern AVX machine for a 1G range to n=25K. There is almost no difference and I'm running an ancient machine. Doing it the way you describe is too cumbersome. Sieving n<25K on base 3 saves very little time. Running it all at once to n=25K also saves you having to send me so many different files of different n-ranges.

Regardless, if you want to test to n=2500, sieve the rest, and test that, go ahead. Just please reserve smaller ranges.


Reb,

For BOINC, I would prefer that we stick with testing n>25K on base 3. The files and primes just get too big otherwise. It will reduce both yours and my admin effort.


Gary

Since most of the time used would be for I/O, not CPU, has anyone considered running more instances than cores at a machine? Also, since there is so much I/O does the use of an SSD improve the speed? Consider the following experiment:

On a machine with 4 cores:
a) Time running 4 ranges of 100,000 k concurrently
b) Time running 5 ranges of 100,000 k concurrently

If (b) takes 25% more time than (a), then no gain. If (b) takes < 20% more time than (a), it might be work considering.

Run that test on a machine with an HDD and on a machine with an SSD and see if (b) improves dramatically over (a) on the machine with the SSD. If (b) take < 10% more than (a), it might be interesting to run even more instances to see where the drop off is.

[QUOTE=gd_barnes;405135]Kenneth,

Why do you continue making such large reservations? The files would be huge. I've asked before for you to reserve smaller ranges, complete them, reserve the next range, etc. I have reserved k=16-20G to n=25K. My suggestion is for you to reserve in 4G ranges max. You could reserve P=20G-24G, finish it, reserve P=24G-28G, finish it, etc. In other words, don't try to reserve half of the remaining k's at once. I think your estimates are way too optimistic because of the boredom factor.

PFGW with the -f30 factoring switch and no sieving is the way to go up to n=25K on this base. On my 4G range, I just have one 4-core old slower machine set to run it non-stop with no sieving (i.e. factoring to 30%). It will take 4-6 months (1G per core) and I don't ever have to touch the machine except to verify that it is still running. Compare that to you taking 4 months on a modern AVX machine for a 1G range to n=25K. There is almost no difference and I'm running an ancient machine. Doing it the way you describe is too cumbersome. Sieving n<25K on base 3 saves very little time. Running it all at once to n=25K also saves you having to send me so many different files of different n-ranges.

Regardless, if you want to test to n=2500, sieve the rest, and test that, go ahead. Just please reserve smaller ranges.


Reb,

For BOINC, I would prefer that we stick with testing n>25K on base 3. The files and primes just get too big otherwise. It will reduce both yours and my admin effort.


Gary[/QUOTE]

Well, I've tried telling numerous times, that it is more efficient on my systems to run these large ranges. Power isn't free nor cheap and I work hard for my money, so for me, efficiency is the way forward. We however seems to disagree on the right approach and therefor I decline doing further work. So to be clear, only R3 8G-13G is reserved to n=100K and after that I withdraw my ressources.

I just don't understand how you or PuzzlePeter can complete a 1G range on a single core in 16 weeks of computation time. Last night I did test k=43,000,000,002 to k=43,000,100,000 using both -f0 and -f30 on my Hasswell. Using -f0, I can run to n=2500 a range of 75M (37,5M k's base 3) each day, versus 19.8M (9.9M k's base 3) a day running with -f30. So even though no one seems to believe that running with -f0 to n=2500 and sieve and test according to my (now cancelled) suggested approach, it is more efficient by a factor of 3.8 times. I just don't think that it will take me 149 CPU days to go from n=2501 to n=25K.

However best of luck to the future hunters.

[QUOTE=KEP;405159]Well, I've tried telling numerous times, that it is more efficient on my systems to run these large ranges. Power isn't free nor cheap and I work hard for my money, so for me, efficiency is the way forward. We however seems to disagree on the right approach and therefor I decline doing further work. So to be clear, only R3 8G-13G is reserved to n=100K and after that I withdraw my ressources.

I just don't understand how you or PuzzlePeter can complete a 1G range on a single core in 16 weeks of computation time. Last night I did test k=43,000,000,002 to k=43,000,100,000 using both -f0 and -f30 on my Hasswell. Using -f0, I can run to n=2500 a range of 75M (37,5M k's base 3) each day, versus 19.8M (9.9M k's base 3) a day running with -f30. So even though no one seems to believe that running with -f0 to n=2500 and sieve and test according to my (now cancelled) suggested approach, it is more efficient by a factor of 3.8 times. I just don't think that it will take me 149 CPU days to go from n=2501 to n=25K.

However best of luck to the future hunters.[/QUOTE]

KEP, please don't be so hasty. Gary, if someone wants to take on a large amount of work, we need to know if they can complete it in a reasonable amount of time with their resources. He is right that boredom is a factor, but at the same time seeing the light at the end of the tunnel allows one to keep focused on a task until it is completed.

Between the two of you, I think you can work out how you can submit results on a large reservation in a way that will not burden Gary.

IMO, we need some concrete numbers so that we know the actual level of effort that you or anyone else is signing up for on this base, which is partly why I posted what I did earlier today. I am still running some tests to determine the optimal settings. One result I can say is that -f30 is not optimal when searching to n=25000. It is 10% slower than -f10. I will have more complete results later.