P-1 factoring anyone?

[Christenson]

I think "very generous" is 5-10 gigs

[kladner]

Originally Posted by Christenson

I think "very generous" is 5-10 gigs

OUCH! Is that facetious, or for real?

[delta_t]

You don't really need to devote that much, but some of us actually do give that much. I usually give P-1 6GB during the day and 9GB overnight. I only allow 2 instances of P-1 to run during the day, and 3 during the night, but each simultaneous run can get up to 3GB each, or the whole allotment if only one instance is running.

Memory=6144 during 7:00-23:00 else 9126
MaxHighMemWorkers=2 during 7:30-22:30 else 3

I do have to stop the high memory P-1 instances when I have to do some big GIS projects, but otherwise when the RAM isn't being used I let it have it's generous share. There are diminishing returns and I don't know where the optimum amount is, but if you have and can spare the memory, give it a little more I say, but I think Brain's suggestion is a good one, give at least 512 per P-1.

[Chuck]

Originally Posted by kladner

OUCH! Is that facetious, or for real?

[Oct 19 10:12] M56324143 stage 1 complete. 1918474 transforms. Time: 46271.881 sec.
[Oct 19 10:12] Starting stage 1 GCD - please be patient.
[Oct 19 10:13] Stage 1 GCD complete. Time: 71.033 sec.
[Oct 19 10:13] Available memory is 7916MB.
[Oct 19 10:13] Using 7388MB of memory.  Processing 288 relative primes (0 of 288 already processed).
[Oct 19 10:28] M56324143 stage 2 is 0.00% complete. Time: 847.733 sec.

It's for real on my machine. It does the Brent-Suyama extension to the 12th.

Chuck

[S34960zz]

Originally Posted by kladner

OUCH! Is that facetious, or for real?

...
[Oct 17 07:35:34] Optimal P-1 factoring of M57013367 using up to 14336MB of memory.
[Oct 17 07:35:34] Assuming no factors below 2^68 and 2 primality tests saved if a factor is found.
[Oct 17 07:35:35] Optimal bounds are B1=680000, B2=20400000
[Oct 17 07:35:35] Chance of finding a factor is an estimated 7.11%
[Oct 17 07:35:36] Using Core2 type-3 FFT length 3M, Pass1=1K, Pass2=3K
[Oct 17 07:45:01] M57013367 stage 1 is 1.019% complete. Time: 564848.944 ms.
...
[Oct 18 02:42:43] M57013367 stage 1 is 99.908% complete. Time: 730762.949 ms.
[Oct 18 02:43:48] M57013367 stage 1 complete. 1961800 transforms. Time: 68887372.793 ms.
[Oct 18 02:43:48] Starting stage 1 GCD - please be patient.
[Oct 18 02:45:34] Stage 1 GCD complete. Time: 105972.309 ms.
[Oct 18 02:45:34] Available memory is 14284MB.
[Oct 18 02:45:34] Using 12062MB of memory.  Processing 480 relative primes (0 of 480 already processed).
[Oct 18 03:11:11] M57013367 stage 2 is 0.000% complete. Time: 1536191.236 ms.
...
[Oct 19 03:18:30] M57013367 stage 2 is 99.261% complete. Time: 841249.614 ms.
[Oct 19 03:29:37] M57013367 stage 2 complete. 2192838 transforms. Time: 89035107.422 ms.
[Oct 19 03:29:37] Starting stage 2 GCD - please be patient.
[Oct 19 03:31:24] Stage 2 GCD complete. Time: 107675.838 ms.
[Oct 19 03:31:24] M57013367 completed P-1, B1=680000, B2=20400000, E=12, We4: C3BD3AD0
...

[kladner]

Wow! I had no clue. Thanks to all of you for educating me. My 8GB seems paltry by that standard. I thought I was being generous with 2GB for overnight runs of 2-3 P-1's. I guess I'll reset the night allocation in 64bit.

[Mr. P-1]

Originally Posted by kladner

What qualifies as 'very generous'?

I don't know. All I know is that it varies depending upon the size of the exponent, and that I don't have enough to get the B-S extension on the 45M exponents I'm testing.

The B-S extension finds more factors, but it also increases the running time, so the benefits are actually quite marginal. The main benefit of increase memory is reducing the number of passes, which reduces the running time, (to which P95 responds by increasing the limits, so finding more factors in about the same time), but again the benefit is marginal once you go past about 1GB.

To summarize: More memory is better, but not much better. If you have it, by all means use it. But if you're planning to build a new computer and intend it to be a P-1 engine, spend your money on a fast processor and fast memory, rather than lots of memory.

[kladner]

Originally Posted by Mr. P-1

But if you're planning to build a new computer and intend it to be a P-1 engine, spend your money on a fast processor and fast memory, rather than lots of memory.

That's good to know, but it's way down the road for me. I am running on my "new computer". I do have two more RAM slots, though. <G>

[Mr. P-1]

Originally Posted by kladner

That's good to know, but it's way down the road for me. I am running on my "new computer". I do have two more RAM slots, though. <G>

I didn't think anyone was actually planning such a build. My comment was really intended to convey just how insignificant all that extra memory really is. GIMPS needs more P-1ers, and nobody should be put off from doing this kind of work because they don't have multiple GB to allocate to it. 500MB per HighMemWorker is ample.

[kladner]

I do take your meaning, Mr. P-1. And I appreciate getting a clear picture of the relative importance of different aspects in a computer which make a difference.

I couldn't resist joking, though. I was gob smacked at the thought that 10GB was not excessive, even if it became clear that the returns don't exactly justify it unless those 10 gigs are lying fallow.

Still, I can throw a fair amount of RAM at P-1 when I'm not using the box.

[axn]

Originally Posted by Mr. P-1

The B-S extension finds more factors, but it also increases the running time, so the benefits are actually quite marginal.

Actually, it is not at all obvious that it _is_ beneficial. Period. The current P-1 code just makes WAGs about B/S efficiency. It is probably more beneficial if people don't give enough memory for B/S so that they can run thru more P-1 tests -- rather than some tests getting super P-1 and some getting none at all.