Strategic P-1 for DC (requires 24GB or more memory for prime95)

[Prime95]

Some double-check exponents could use more P-1.

Requirements:
1) A machine with at least 32GB of RAM.
2) Version 30.8 of prime95
3) Configure prime95 to use 24GB or more.

Reserve some exponents from the list below (post your reservation so the exponents can be removed from the list). My quad core machine is doing one P-1 every ~70 minutes.

[Prime95]

Choose from:

Pfactor=60C91062A9B68629577154EFB5797907,1,2,104887421,-1,77,1
Pfactor=D80C65484A97180704E24962B2A009FF,1,2,104887493,-1,77,1
Pfactor=6727B2066DA6F550AF0C500B20EED9CB,1,2,104887571,-1,77,1
Pfactor=0D2186B36BF42B091F631C03C1029BC5,1,2,104888029,-1,77,1
Pfactor=7CA0B57DB8E02203FBF5DBE1345CAF4E,1,2,104888101,-1,78,1
Pfactor=B1B8B1B879794ABE708F81C9452CED36,1,2,104888177,-1,78,1
Pfactor=85EE49B64998B0F2BEAD2310998F921D,1,2,104888507,-1,78,1
Pfactor=A1B65E943FAC701166192564F663491F,1,2,104889061,-1,78,1
Pfactor=89EFCAB7A3F6134BDF71A6D06E676364,1,2,104889157,-1,78,1
Pfactor=349E21325D249D8F10EF3896F3E74CF4,1,2,104889373,-1,78,1
Pfactor=07E89EC97B863AB02C62CBBF59F03A03,1,2,104889503,-1,78,1
Pfactor=E559177EA9A6B3D840B799FD8BE4D1DF,1,2,104889559,-1,78,1
Pfactor=EB004F60BCE6207AEAF6672FC96961AA,1,2,104889619,-1,78,1
Pfactor=ACDA78FD291423B0268272A111CDD79A,1,2,104890099,-1,78,1
Pfactor=5DF14872C79F40167E07FA2D5A2D3E88,1,2,104890427,-1,78,1
Pfactor=D0A65BA34A521D987E8BAD0B9833658D,1,2,104890451,-1,78,1
Pfactor=4D594C23F181FFB97FA2750BB0E7B405,1,2,104891327,-1,77,1
Pfactor=B67A90931550A4F40E9E5D08CDAC77F6,1,2,104891359,-1,77,1
Pfactor=DD8B63BFB5E931B3ADCEF8667EEC6A9B,1,2,104891387,-1,77,1
Pfactor=15732EE9E3E8E99E765A013525872A93,1,2,104891441,-1,77,1
Pfactor=7C10ED8DC58CC1B3FABF6010411479BA,1,2,104891497,-1,77,1
Pfactor=48CD531964C77C5405E06AB5FABAB37A,1,2,104891561,-1,78,1
Pfactor=40378536A734631036A82852A308107F,1,2,104891719,-1,77,1
Pfactor=5478C3816B61CF01597E915BE88458AB,1,2,104891807,-1,77,1
Pfactor=1EBF2BC93DAE8C60C5CC9708533B0440,1,2,104892097,-1,77,1
Pfactor=A0EAC22A284467482FE30E2BDE830D46,1,2,104892197,-1,77,1
Pfactor=23D657AB0B2B43F310607C711E899F10,1,2,104892223,-1,77,1
Pfactor=779DED02309A289B1C9033F7E6BC814B,1,2,104892289,-1,78,1
Pfactor=FE4D325F83DA43C759078ABBF6010236,1,2,104892311,-1,77,1
Pfactor=8DC697BB9B1A992ECC992D31C7D121A9,1,2,104892451,-1,77,1
Pfactor=C85BCF396D35F9B16BF95E4D6F69AF64,1,2,104892493,-1,77,1
Pfactor=C112F3947C59ECBD891ECA87FA142EB7,1,2,104892607,-1,77,1
Pfactor=E32C029CB6E506448B7C0157CE0194E6,1,2,104892647,-1,77,1
Pfactor=5A94AF7891E754D95147575DA26AD178,1,2,104892679,-1,77,1
Pfactor=D892BDA762EF934CAA3823C43DCCEA7D,1,2,104892727,-1,77,1
Pfactor=EEF3543904A52FF0E59C636326C40753,1,2,104892757,-1,77,1
Pfactor=7B890736F1343FF06B79BE48AAA358F9,1,2,104892761,-1,77,1
Pfactor=08786F098E884B0E9A65FA48B698DDD3,1,2,104892829,-1,77,1
Pfactor=791CB9D63F9CF00CD7CFA7A3C83C91A7,1,2,104892859,-1,77,1
Pfactor=30888490A864C9E86FEEE61F82306E12,1,2,104893001,-1,77,1
Pfactor=A9F23B9028E6D1A2E086698952D97786,1,2,104893021,-1,77,1
Pfactor=385E374DF58F91735477EF6F5973E295,1,2,104893067,-1,77,1
Pfactor=2AC736D1827C96E6AC59344176033ADE,1,2,104893171,-1,78,1
Pfactor=AB9F8D18DFACE6A28131BB90B4A0D74A,1,2,104893183,-1,77,1
Pfactor=9C3CC76481DC313B38FC5812C55B7E10,1,2,104893219,-1,77,1
Pfactor=0051B162E14BF193A2F95B3AA7AD6BC5,1,2,104893259,-1,77,1
Pfactor=C8EB5C5DE11813A7AC5D75831166308C,1,2,104893309,-1,77,1
Pfactor=9A9292BA78D3B4640A3660A1C02BFAD9,1,2,104893339,-1,77,1
Pfactor=675A40636C68F812861D8FA2B5BB5132,1,2,104893361,-1,77,1
Pfactor=01036EF8139048ECAFF46476E87C3554,1,2,104893387,-1,77,1
Pfactor=D9591638662A529EDCD83F140CFCC0D8,1,2,104893489,-1,77,1
Pfactor=69852214F136E71C8145168D3C18B741,1,2,104893511,-1,77,1
Pfactor=017FD2414D4B2D14555F720A7AA54F71,1,2,104893517,-1,77,1
Pfactor=45E43254ADAB31B68D19A9686CAA32E1,1,2,104893693,-1,77,1
Pfactor=5C543096D4B10168D2262AB9F514635A,1,2,104893759,-1,77,1
Pfactor=54ADA47EC7415A48E453D1971E77532A,1,2,104893787,-1,77,1
Pfactor=CFC39F3CDC4F01A1DFF023C1F297FE94,1,2,104893829,-1,77,1
Pfactor=BCBED96757F37BDE46EB5E289FACABF1,1,2,104893891,-1,77,1
Pfactor=D18BEE3D415387A02E85879F6B5FEB11,1,2,104893931,-1,77,1
Pfactor=ECB82F1E1A6C684C57A27C00DC2629AF,1,2,104893933,-1,78,1
Pfactor=97B8B6227F9FD3FD99EBDB6336CCC418,1,2,104893969,-1,78,1
Pfactor=E7B2082978FC83EDC38C5A3EBD1D3301,1,2,104894341,-1,77,1
Pfactor=ADEEB595A19341AD7B7F7423B2AADE50,1,2,104894351,-1,77,1
Pfactor=286DDA94E4DF185B1323DC3994CCAA1E,1,2,104894353,-1,77,1
Pfactor=DA971F33A7B6D5F29EB678DE2017708D,1,2,104896999,-1,78,1
Pfactor=79DBC6F9CF7D1D69115F7A730629075C,1,2,104897179,-1,78,1
Pfactor=05D98AEE413339F815ED8BE52AE64FD5,1,2,104897729,-1,78,1
Pfactor=7581A28AF8D612FE8AE4B2918E61444A,1,2,104898763,-1,78,1
Pfactor=A9F6641A2925B20F3B892397F1A8D8ED,1,2,104898767,-1,78,1
Pfactor=B4B47161B2611FE26699F6215DC13DC2,1,2,104898809,-1,78,1
Pfactor=10F1D6585FA5FCC137DDF2E64D054939,1,2,104898923,-1,78,1
Pfactor=678B6E7915842751BAD62EC330A78866,1,2,104898979,-1,78,1
Pfactor=F6D66D8C0005EE1163ABC61C85B469F7,1,2,104899261,-1,78,1
Pfactor=197E8850A3A2EAA5A2EEA0F5AC0E1656,1,2,104899511,-1,78,1
Pfactor=72C1E61D9B98B13E7CA70C99F143F774,1,2,104899577,-1,78,1
Pfactor=6BC6115AB9D96DE91031DF18F3860B8A,1,2,104899787,-1,78,1
Pfactor=C9235D0C80CCA836B762C2DC61722437,1,2,104900093,-1,78,1
Pfactor=91538908F66131C34CCD337A95522516,1,2,104900261,-1,78,1
Pfactor=3EC1C162176C74C353B3C1775982FEED,1,2,104900639,-1,78,1
Pfactor=F29320650C6E6F073B1465F8198DA513,1,2,104900701,-1,78,1
Pfactor=D38860CB365B02C726C07FB56103CD30,1,2,104900707,-1,78,1
Pfactor=EFC24F05433B2B5B6E39FD7E39392C68,1,2,104900921,-1,78,1
Pfactor=AB601DB24BF2D31541106D93F32A6A43,1,2,104900987,-1,78,1
Pfactor=274FE659B534A464C086551E71E6349A,1,2,104901617,-1,78,1
Pfactor=B1F913FF2BEBCADE4039A6DADBAD4436,1,2,104901631,-1,78,1
Pfactor=988A35B6886CB404B8228F1DF3A59F41,1,2,104902079,-1,78,1
Pfactor=8C9B098DD5948CEC06A012C99BD19114,1,2,104902397,-1,78,1
Pfactor=42EC6D41283B67C059D21A28B39E46DD,1,2,104902459,-1,78,1
Pfactor=1F6435354AAC889F864496C8BA5C161B,1,2,104902607,-1,78,1
Pfactor=4340D24A75E5A72BDB4FF51A598C480C,1,2,104902879,-1,78,1
Pfactor=0D7C7DB9DD6137461CB14A410029DC8F,1,2,104904269,-1,78,1
Pfactor=110F3797BDF9D5930D27194AF3F1413B,1,2,104904307,-1,78,1
Pfactor=F0C89D4F4008AA8CCA9CEAAAA4D6E1AC,1,2,104905211,-1,78,1
Pfactor=D8AFE707D10DD08157018E5FAA5B4562,1,2,104906143,-1,78,1
Pfactor=413BEECCA267DA7CB595C493D77C7B31,1,2,104906261,-1,78,1
Pfactor=D3295FC127F1482F991B9173B42466A1,1,2,104906437,-1,78,1
Pfactor=F0FF6213F1A861773290F3C7C48F2A3C,1,2,104908169,-1,78,1
Pfactor=FFEC8497092C99133F517435A6654FB5,1,2,104908723,-1,78,1
Pfactor=66414D4DA16A4BF62D4F7B68377AE21B,1,2,104909683,-1,78,1
Pfactor=8289B99F180DFE2D1E4FBA68C202A370,1,2,104909807,-1,78,1
Pfactor=8522FF16C0A27DFCADCB1AB9CCFC82E9,1,2,104909869,-1,78,1
Pfactor=74A05DA30FA49AA8A598843DBF9831B6,1,2,104910467,-1,78,1
Pfactor=8B2D32BC2FEA77970B30E2FD8587FBE5,1,2,104910527,-1,78,1
Pfactor=123BD572A5AB48FADEE6A731380E6818,1,2,104911171,-1,78,1
Pfactor=E5835C559F4E34865FC10F1D49D87122,1,2,104911259,-1,78,1
Pfactor=C1F460DCBFB6619E26AAFCC53739A7DE,1,2,104911273,-1,78,1
Pfactor=8DC67F32B37E52D17535C258DE5653A0,1,2,104911493,-1,78,1
Pfactor=4D27AB23E72D896AFAC0DE36F536E806,1,2,104911511,-1,78,1
Pfactor=A8EDF3EBA8C9BC5308A958962545D71B,1,2,104911549,-1,78,1
Pfactor=C654B737FC8B9DEED07BCFDCF6DEB0A3,1,2,104911591,-1,78,1
Pfactor=6F189B867F3E1F94FD50A52C9BD3111E,1,2,104911889,-1,78,1
Pfactor=9B8D0651F8A578A5D179749DE8AA993D,1,2,104911969,-1,78,1
Pfactor=53DF6A9FF369595C9652F76CB83C9520,1,2,104911979,-1,78,1
Pfactor=50D4088CC869EC3B88A262320ACEF757,1,2,104912099,-1,78,1
Pfactor=BB67AA1AE2BF97A55D9B339037DEBDEB,1,2,104912147,-1,78,1
Pfactor=ECB0CBCA063875F9911F92F938FF2D64,1,2,104912351,-1,78,1
Pfactor=08AC6C408005CB39218F67BBFD1F6CF4,1,2,104914421,-1,78,1
Pfactor=4F0DC9AB6E3E0159C82A95BCC994FA05,1,2,104914729,-1,78,1
Pfactor=F53E880884514CC7BA3C8ADA9E93B184,1,2,104914987,-1,78,1
Pfactor=015F77BF233F755011E3E8EA853C5DC5,1,2,104915123,-1,78,1
Pfactor=0FB9808D80F660BA69ED4F4AC8FE85CE,1,2,104915329,-1,78,1
Pfactor=50658AACD8BA19D4B6CF00D18BF4F21F,1,2,104917471,-1,78,1
Pfactor=5D794C9BEE6CDB08F72E569B39A45295,1,2,104917643,-1,77,1
Pfactor=48A0E261131A2DED69EAC0548919BC5E,1,2,104918003,-1,78,1
Pfactor=EEDDAA03232420CEBF6C94B351CB9874,1,2,104918537,-1,77,1
Pfactor=02916DE31935B7622CB5B438F1749A15,1,2,104918591,-1,78,1
Pfactor=8B8E6441E4920F7725C90FDFD2EC161E,1,2,104920111,-1,78,1
Pfactor=0444C335D27BFAF21FC3B75C3DD8E90E,1,2,104920303,-1,78,1
Pfactor=BFA593D4156F4210EB3EAFABD700316D,1,2,104920367,-1,78,1
Pfactor=541F4A9916BEABA710D591E7289A5AFB,1,2,104920877,-1,78,1
Pfactor=D8A54FB9AD0F2D9A650A6369D6781D8D,1,2,104921923,-1,78,1
Pfactor=9D847AB487933CD5A1CF4E704CD2CA19,1,2,104921989,-1,78,1
Pfactor=B686D1A8F117C51D628CAE02EA7AC6ED,1,2,104922331,-1,78,1
Pfactor=F7423BDD23FCCD1F328B7E341AC98445,1,2,104922613,-1,78,1
Pfactor=951477C5CC75578E888326D6F9B2807A,1,2,104922773,-1,78,1
Pfactor=58FB813852F7C608EFE8A4FFB93845E8,1,2,104922859,-1,78,1
Pfactor=8321868D62C0DAA0550F6104E0D6BE9E,1,2,104923081,-1,78,1
Pfactor=869482CCE263FCC867BA77B08623C3A2,1,2,104923121,-1,78,1
Pfactor=488BFA4F547ECB48CA804DC265AD3D23,1,2,104923207,-1,78,1
Pfactor=6CA6FF59ACB543A8A133B12164BCA96A,1,2,104923229,-1,78,1
Pfactor=0C5BA146FA4DA1BC4E735F2CBA7D3F4A,1,2,104923307,-1,78,1
Pfactor=7169C5DC58E3E6C75DC1C4A7F67B51FB,1,2,104923333,-1,78,1
Pfactor=B57A87CB548F709240438FD37A8ED499,1,2,104924177,-1,78,1
Pfactor=BA91FF4483C21ED2EC0586CAED5ED55B,1,2,104924249,-1,78,1
Pfactor=1E152B1E2A4CB0F04A84601FE36F696F,1,2,104924299,-1,78,1
Pfactor=A3BE958073B5DF073126263FF2DCC6A4,1,2,104929067,-1,78,1
Pfactor=FE585A31D6136A1033B953E6BBD9221F,1,2,104929189,-1,77,1
Pfactor=E7E48CAFD795F24C398A737D346DEE83,1,2,104929379,-1,78,1
Pfactor=410E4A431E0EF4733B86A3C0EBC554B9,1,2,104929663,-1,77,1
Pfactor=B4CF31463D2FFC68C5EB3D0FB9FC2901,1,2,104929681,-1,78,1
Pfactor=3E03B9C069911CA2CF46D058CCFBC592,1,2,104934281,-1,78,1
Pfactor=5ED677B490A447343ADFEB67E4D0DAA6,1,2,104934413,-1,78,1
Pfactor=F5B0E5C1C6B0829499864ADB99728CD7,1,2,104934439,-1,78,1
Pfactor=1F13AAD36A11E91FD8E7FB7906CBCD9A,1,2,104946313,-1,77,1
Pfactor=37B23C5570EB673C30A2F93B7357FC77,1,2,104950583,-1,78,1
Pfactor=340AC708F7DD8FB851ED659E0D662F7A,1,2,104955359,-1,77,1

[congsz]

Originally Posted by ScottieMckinley

How do I add them to my assignments?

Just add them to your "worktodo.txt" file.
An original "worktodo.txt" often looks like this

[Worker #1]
DoubleCheck=[some random assignment code],64590839,74,1
PRP=[some random assignment code],1,2,113700893,-1,79,0

And you can add the lines above between "[Worker #1]" and the second line of the file, like this:

[Worker #1]
Pfactor=[something]
Pfactor=[something]
Pfactor=[more thing]
DoubleCheck=[some random assignment code],64590839,74,1
PRP=[some random assignment code],1,2,113700893,-1,79,0

[kriesel]

Originally Posted by ScottieMckinley

How do I add them to my assignments?

Like any manually obtained assignment; stop prime95's computation (Test, Stop), copy paste into prime95's worktodo.txt with a text editor & save changes & exit, then Test, Continue in prime95; Advanced, Manual communication, check Send new expected completion dates to server, OK (which switches the reservations to you, if it can register them with the server). And post here what you've claimed. (Very similar process to the directions here.) Or for mprime, the numbered-menu equivalent.

I see you're new to the forum. You may find some of the reference info useful or interesting.


I'll work on this block:

Pfactor=121A68F7153A29C5ECA580A00C229F0D,1,2,63267409,-1,75,1
to
Pfactor=B93285D5FD03BA2CCE9D3517BFC27F21,1,2,63271763,-1,75,1

[petrw1]

Absolutely Stage 2 is fastest with all cores and lots of RAM. But I find Stage 1 is slower in that setup.

This requires some manual intervention but I get the best total thruput if I:
1. Run 4 workers but only 1 HIGHMEMWORKERS; so at any point 3 are doing Stage 1 and 1 Stage 2.
2. This quickly results in a Stage 2 backlog; cores having to skip all the Stage-1-done work.
3. So once a week or two I restart with all the Stage 2 work stacked up and 1 worker 4 cores.

I believe another option is to:
1. Run all your assignments Stage 1 only with 4 workers realizing the results will be reported.; make sure you save the Stage1 work files
2. Reque all the work for Stage2 with 1 worker; the Stage2 result will also be reported.

BTW on my 8 core I only run 4 workers x 2 cores for Stage 1. Even with a liquid cooler 4x2 give me Temps just over 80. 8x1 gives me Temps of 90; too hot for my liking.
Stage 2 with 1 worker x 8 cores give me low 70's.

[kriesel]

Originally Posted by petrw1

Absolutely Stage 2 is fastest with all cores and lots of RAM. But I find Stage 1 is slower in that setup.

Latency for a P-1 run is one thing; best aggregate throughput for the GIMPS project is quite another. On i5-1035G1 4-core & x2 HT, AVX512, 64 GiB ram installed, benchmark iteration timings are fastest for fft lengths suitable for P-1 retry on DC wavefront and somewhat above, with 1 core per worker. BUT, the overall productivity of net % of primality test time saved, is increased by ~25%, due to higher stage 2 efficiency, and consequent higher stage 1 and 2 bounds selected, with almost all ram available to the worker running stage 2 at the moment, while the benchmark iteration timings are only ~3-6% lower aggregate throughput, for 2 cores per worker, than for one core per worker, so overall, P-1 effectiveness appears to be optimized at 2 workers, 2 cores/worker. If running P-1 in both workers, stage 1 and 2 will eventually alternate between workers to about fully utilize the ram most of the time. (That's the configuration George settled on.)

At higher exponent, such as first-test wavefront, two cores/worker becomes closer to optimal in the benchmark timings, so the case for two workers then becomes stronger.

Experimenting with increasing allowed ram, prime95 imposed an upper limit of ~57.4GiB allowed ram for stage 2, for an attempt on ~67M on the 64GiB system.

On other systems, containing 128 GiB installed ram, prime95 accepted the highest attempted, 110GiB. At wavefront or somewhat higher these ran ok. At p~551M, it appears that around 27% of the way through stage 2, it began thrashing badly, not logging further progress for more than a day.

[Prime95]

BTW, thanks to all for getting us well ahead of the DC wavefront. A few expiring assignments below the wavefront will appear from time-to-time, but there should be no problem from now on keeping ahead of the DC wavefront.

[kriesel]

Originally Posted by lavalamp

Edit: Incidentally, I don't suppose the shiny new P-1 code would work on billion digit numbers too? I've wanted to run P-1 on some of those for a while but had no tools to do so.

No. Max fft size depends on processor type. For AVX512, it's 64M in prime95/mprime, which supports up to ~1169 Mbits in LL/PRP, and in P-1. Also the required memory would be quite large; scale up by exponent, roughly, or more accurately as FFT length required I think. So if running ~16-60GiB for 100M exponents, plan on ~512GiB-2TiB for OBD, polynomial_P-1 style. I haven't seen any indication George plans to extend the fft sizes any time soon, much less to 192M which is what OBD takes.

However, Mlucas 20.1 supports traditional bounds OBD P-1 of which johnny_jack and I have a few under way, and Mlucas also supports somewhat higher for Mersennes, and up to F33 P-1 which Ernst Mayer has under way. Think ~ a year per OBD exponent on medium speed machines. And at least 64 GiB ram, more is better, for OBD. ECC type and enabled is recommended for such long runs. Pre-qualifying hardware on smaller exponents with known factors to confirm reliability and determine run time scaling allowing extrapolation to Gdigit is recommended.

To do OBD P-1 on GPU would require modifying gpuowl to larger fft sizes than it currently supports.
And LOTS of on-GPU ram; 16GiB on radeon VII, Tesla P100, etc. is about enough for 1.46Gbit Mersennes in gpuowl v6.11, almost 1Gbit in V7.x. I don't know of any GPUs offering more than 48GiB yet. Except perhaps this, but that's out of nearly everyone's price range.

The preceding is partly a rehash of the exponent limits reference info post.

[mrh]

Originally Posted by kriesel

of how many cores, on what hardware? George suggests on Intel, 2 workers, so one can be knocking out stage 1 while the other is using almost all the ram on stage 2, alternating. At these modest exponents, fewer cores/worker tends to be more throughput, on hardware I've benchmarked.

This system has two Intel(R) Xeon(R) Gold 6146 cpus. So a lot of cores. I picked 8/256 because it didn't seem to interfere with the other work I'm using the machine for, and seemed to cost about only about 150 watts. that seemed like a reasonable amount of $/day to apply to this effort. (my electricity isn't exactly cheap ) Do you think it makes more sense to split the resources between two workers?

[kriesel]

Originally Posted by mrh

Do you think it makes more sense to split the resources between two workers?

Yes, but test to confirm. And beware of NUMA impact in any case.

[James Heinrich]

Claiming:
Pfactor=0F31F773E7803BCD9177A794225E942F,1,2,85349807,-1,77,1
to
Pfactor=0307B430913824CA2BEC59F394AB974E,1,2,85406911,-1,77,1

edit: range completed, found 2 factors:
M85351163 has a factor: 6582679500630930530062633 (P-1, B1=415000, B2=88044000)
M85405897 has a factor: 7091352346301817991540489 (P-1, B1=434000, B2=89378850)