COMPLETE!!!! Thinking out loud about getting under 20M unfactored exponents - Page 92

petrw1 · 2021-12-12, 02:14

Quote:

Originally Posted by Luminescence

I also thought of TF in the lower ranges, there is no way they can keep up. 30.8 is leaving GPUs completely behind.

26.5M is in it's final moments and I don't like to see

Code:

Ranges remaining:
100 or more: 9

Mind if I take down 12.3M? Do you have any probability I should be aiming for?

Thanks ... go for it.

Statistically, Bounds of 3M/600M should give 123 factors.
You need 119; that's cutting it close.

Your two options then are higher bounds ... maybe 5M/1000M should give 144 factors.
Or if you or someone takes it to 73 bits you should get 25 factors via TF.

Luminescence · 2021-12-12, 02:28

Quote:

Originally Posted by petrw1

Thanks ... go for it.

Statistically, Bounds of 3M/600M should give 123 factors.
You need 119; that's cutting it close.

Your two options then are higher bounds ... maybe 5M/1000M should give 144 factors.
Or if you or someone takes it to 73 bits you should get 25 factors via TF.

Ok, thanks a lot! I'm gonna TF myself in parallel.

VBCurtis · 2021-12-12, 03:18

Quote:

Originally Posted by petrw1

Thanks ... go for it.

Statistically, Bounds of 3M/600M should give 123 factors.
You need 119; that's cutting it close.

Your two options then are higher bounds ... maybe 5M/1000M should give 144 factors.
Or if you or someone takes it to 73 bits you should get 25 factors via TF.

I don't think you're thinking big enough for B2. I'm running some 8.6M expos by agreement with masser, and with 8M/6000M stage 2 time is still lower than stage 1 time. B2 = 200*B1 isn't enough, if you have lots of RAM.
Perhaps those without 32GB+ ram should maybe not do the tough ranges, since they benefit so much from big memory? My timings are with 45GB assigned to P95 on a 64GB/socket machine.

Luminescence · 2021-12-12, 03:55

Quote:

Originally Posted by VBCurtis

I don't think you're thinking big enough for B2. I'm running some 8.6M expos by agreement with masser, and with 8M/6000M stage 2 time is still lower than stage 1 time. B2 = 200*B1 isn't enough, if you have lots of RAM.
Perhaps those without 32GB+ ram should maybe not do the tough ranges, since they benefit so much from big memory? My timings are with 45GB assigned to P95 on a 64GB/socket machine.

I've just played around a bit with the probability calculator. I'm gonna look at the runtime with 1.1M/1500M since I can get around 120GB out of three machines. Even with a 1200 B1/B2 ratio, stage 2 was still faster by a few minutes in 26.5M.

petrw1 · 2021-12-12, 04:15

Quote:

Originally Posted by VBCurtis

I don't think you're thinking big enough for B2. I'm running some 8.6M expos by agreement with masser, and with 8M/6000M stage 2 time is still lower than stage 1 time. B2 = 200*B1 isn't enough, if you have lots of RAM.
Perhaps those without 32GB+ ram should maybe not do the tough ranges, since they benefit so much from big memory? My timings are with 45GB assigned to P95 on a 64GB/socket machine.

I freely admit my B2 is likely low and I probably should have chosen a higher multiple.
I needed to pick a number for my spreadsheet to calculate the expected factors.

On my i5-7820X (8 core) with 24GB RAM allocated GIMPS chooses 256x.

3M/600M (200x) is 8.62%, 123 expected factors and 19.8 GhzDays
3M/1200M (400x) is 9.38%, 139 expected factors and 38 GhzDays

@Luminescence ... a little testing will tell you the B1/B2 you need for your desired percentages.

axn · 2021-12-12, 06:02

I'm guessing 1.5m/1.2g (800x) should comfortably get the job done.

@Luminescence, how much RAM do you have in a single machine (if different amounts, what's the largest one)?

Luminescence · 2021-12-12, 06:48

Quote:

Originally Posted by axn

I'm guessing 1.5m/1.2g (800x) should comfortably get the job done.

@Luminescence, how much RAM do you have in a single machine (if different amounts, what's the largest one)?

Three with 128GB total (don't know how much barebone server Ubuntu 20.04 requires) and one with 64GB (Win10, daily machine)

axn · 2021-12-12, 07:01

Quote:

Originally Posted by Luminescence

Three with 128GB total (don't know how much barebone server Ubuntu 20.04 requires) and one with 64GB (Win10, daily machine)

Ok. I missed it when you had said that earlier as well (3 with 120GB).

Then probably 1m/2g (2000x) should be looked at. The windows machine (64GB) can do 1.5m/1.2g

LaurV · 2021-12-12, 13:11

We are, for a little while, back to tickling 11M exponents with P-1. Santa was nice this year.
Starting 11.2 and 11.0, in parallel.

alpertron · 2021-12-12, 13:16

After using 8 years a system based on i5 3470, this week I upgraded my desktop to i5 11400 with 32 GB of RAM.

I disabled the turbo boost because the CPU drains too many watts when running 100%, so the processor runs at 2.6 GHz.

I populated Prime95 with the the list of worst ranges of P-1 in 10-10.9M from https://www.mersenne.ca/pm1_worst.php. I used the bounds 2M / 1288M (B2 was selected by Prime95). Each number requires 53 minutes.

Up to this moment, Prime95 processed 51 numbers and it found only one factorization, M10938773, in step 1.

I'm using Prime95 30.8 build 4. It uses 20 GB.

petrw1 · 2021-12-12, 15:05

Quote:

Originally Posted by alpertron

After using 8 years a system based on i5 3470, this week I upgraded my desktop to i5 11400 with 32 GB of RAM.

I disabled the turbo boost because the CPU drains too many watts when running 100%, so the processor runs at 2.6 GHz.

I populated Prime95 with the the list of worst ranges of P-1 in 10-10.9M from https://www.mersenne.ca/pm1_worst.php. I used the bounds 2M / 1288M (B2 was selected by Prime95). Each number requires 53 minutes.

Up to this moment, Prime95 processed 51 numbers and it found only one factorization, M10938773, in step 1.

I'm using Prime95 30.8 build 4. It uses 20 GB.

Wow thanks ... I assure you the average will be much better than 51 per factor
Do I understand you are working on all of 10.0 to 10.9?

2021-12-12, 13:16	#1011
alpertron Aug 2002 Buenos Aires, Argentina 1,447 Posts	After using 8 years a system based on i5 3470, this week I upgraded my desktop to i5 11400 with 32 GB of RAM. I disabled the turbo boost because the CPU drains too many watts when running 100%, so the processor runs at 2.6 GHz. I populated Prime95 with the the list of worst ranges of P-1 in 10-10.9M from https://www.mersenne.ca/pm1_worst.php. I used the bounds 2M / 1288M (B2 was selected by Prime95). Each number requires 53 minutes. Up to this moment, Prime95 processed 51 numbers and it found only one factorization, M10938773, in step 1. I'm using Prime95 30.8 build 4. It uses 20 GB. Last fiddled with by alpertron on 2021-12-12 at 13:22

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Thinking of Joining GPU to 72	jschwar313	GPU to 72	3	2016-01-31 00:50
Thinking about lasieve5	Batalov	Factoring	6	2011-12-27 22:40
Thinking about buying a panda	jasong	jasong	1	2008-11-11 09:43
Loud thinking on irregular primes	devarajkandadai	Math	4	2007-07-25 03:01
Question on unfactored numbers...	WraithX	GMP-ECM	1	2006-03-19 22:16

2021-12-12, 06:02	#1007
axn Jun 2003 5×29×37 Posts	I'm guessing 1.5m/1.2g (800x) should comfortably get the job done. @Luminescence, how much RAM do you have in a single machine (if different amounts, what's the largest one)?

2021-12-12, 13:11	#1010
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 9962₁₀ Posts	We are, for a little while, back to tickling 11M exponents with P-1. Santa was nice this year. Starting 11.2 and 11.0, in parallel.