P-1 discussion

[chalsall]

Originally Posted by Dubslow

I had (and have) nothing to say.

That is what is known as "taking the piss".

Your teachers don't take kindly to such behaviour.

[flashjh]

Originally Posted by James Heinrich

That seems.... fast. What CPU(s?) do you have in there? What bounds is it choosing?

Yeah, way fast... *who knows* what calculator I was using

So when I rechecked, it's actually about 19 days.

So, here are the bounds it's using:

 
[Jan 24 23:08] Assuming no factors below 2^77 and 2 primality tests saved if a factor is found.
[Jan 24 23:08] Optimal bounds are B1=4070000, B2=133292500
[Jan 24 23:08] Chance of finding a factor is an estimated 6.41%
[Jan 24 23:08] Using Core2 type-3 FFT length 22400K, Pass1=3584, Pass2=6400, 4 threads

I was hoping to get into stage 2 a little sooner, oh well. In three weeks, I'll see stage 2 info.

[kladner]

Originally Posted by chalsall

You are most welcome.

Now, might you continue the discussion you were wishing to hold when you asked for the record of the discussion to be moved?

My hope is that someone might shed light on RAM allocation in Stage 2 of P-1, and how it affects the choice of bounds and the number of Relative Primes. It already seems clear there is a marked difference between 8GB and 12GB machines. I see above that Dubslow is giving ~3.5-4.5GB to a single Stage 2, with 384 and 480 RP respectively resulting.

Then there's Jerry's monster machine, using 20779MB and 960 RP.

I suppose, in some ways the questions have been roughly answered. I remember somewhere Xyzzy mentioned certain thresholds of memory resulting in E=6 or E=12 B-S extension values.

[kladner]

Originally Posted by chalsall

You are most welcome.

Now, might you continue the discussion you were wishing to hold when you asked for the record of the discussion to be moved?

My hope is that someone might shed light on RAM allocation in Stage 2 of P-1, and how it affects the choice of bounds and the number of Relative Primes. It already seems clear there is a marked difference between 8GB and 12GB machines. I see above that Dubslow is giving ~3.5-4.5GB to a single Stage 2, with 384 and 480 RP respectively resulting.

EDIT: These RAM amounts are roughly the same as what I have allocated to 2 HighMemWorkers. I recently reduced that from 3, but so far, this has not changed the B-S extension values. I consistently hit E=6, but never higher, at least so far.

Then there's Jerry's monster machine, using 20779MB and 960 RP.

I suppose, in some ways the questions have been roughly answered. I remember somewhere Xyzzy mentioned certain thresholds of memory resulting in E=6 or E=12 B-S extension values.

[James Heinrich]

For what it's worth, on my good system I get:

Optimal P-1 factoring of M52686883 using up to 10000MB of memory.
Assuming no factors below 2^72 and 2 primality tests saved if a factor is found.
Optimal bounds are B1=520000, B2=10790000
Chance of finding a factor is an estimated 4.02%
Using Core2 type-3 FFT length 2800K, Pass1=448, Pass2=6400, 2 threads
Setting affinity to run helper thread 1 on logical CPU #8
Restarting worker with new memory settings.
Available memory is 9500MB.
Using 9497MB of memory.  Processing 412 relative primes (0 of 432 already processed).

Around 42M I can get 480 RPs into 9500MB; around 48M I can get 432 RPs into 9500MB; around 52M (as you see here) it can only get 412 of 432 into 9500MB. I really need to get that other 32GB of RAM for the machine.
For what it's worth, I get E=12 on everything in this range.

[kladner]

Thanks, James. I'll have to collect some results on my machine and make some comparisons. I take it from "...that other 32GB of RAM...." that you're currently scraping by with only 32GB.

I'm considering getting mine from 8 to 16GB at the moment. It's only $60-70. It does seem it would make some difference. I can still get the identical Corsair modules. I'm not sure if it would force slower timings, though.

[kladner]

I think that the change to MaxHighMemWorkers=2 has started to show results. The latest Stage 2 worker is now showing 480 relative primes. The per-worker allocation has gone up. Of course, this is with the Night time settings, since it is after 2300 local: my night break point.

[James Heinrich]

Originally Posted by kladner

I think that the change to MaxHighMemWorkers=2 has started to show results.

Just be warned: you'll end up with a backlog. With reasonably generous RAM allocation, stage2 always takes longer than stage1, so you'll never catch up.

[kladner]

Originally Posted by James Heinrich

Just be warned: you'll end up with a backlog. With reasonably generous RAM allocation, stage2 always takes longer than stage1, so you'll never catch up.

Thanks. Yes, indeed I am already seeing "Looking for work that uses less memory." At the current rate of backlog it seems that catch-up sessions will have to be allowed every few days to a week.

[bcp19]

Originally Posted by kladner

My hope is that someone might shed light on RAM allocation in Stage 2 of P-1, and how it affects the choice of bounds and the number of Relative Primes. It already seems clear there is a marked difference between 8GB and 12GB machines. I see above that Dubslow is giving ~3.5-4.5GB to a single Stage 2, with 384 and 480 RP respectively resulting.

EDIT: These RAM amounts are roughly the same as what I have allocated to 2 HighMemWorkers. I recently reduced that from 3, but so far, this has not changed the B-S extension values. I consistently hit E=6, but never higher, at least so far.

Then there's Jerry's monster machine, using 20779MB and 960 RP.

I suppose, in some ways the questions have been roughly answered. I remember somewhere Xyzzy mentioned certain thresholds of memory resulting in E=6 or E=12 B-S extension values.

I have 7000MB allocated to P95 on my 2400 and it's doing 52M exps and I notice on around 52.3 it gives E=6 and 52.9 it gives E=12. I figure I must be near the threshold with it jumping back and forth like that. Several 57M exp show E=6 as well. With the 7000 it can load about 248 into memory.

[kladner]

Originally Posted by bcp19

I have 7000MB allocated to P95 on my 2400 and it's doing 52M exps and I notice on around 52.3 it gives E=6 and 52.9 it gives E=12. I figure I must be near the threshold with it jumping back and forth like that. Several 57M exp show E=6 as well. With the 7000 it can load about 248 into memory.

Thanks very much! This is just the sort of thing I was curious about.