P-1 factoring anyone?

[James Heinrich]

Originally Posted by Prime95

Try this. Change your local.txt to:
Memory=500 during 5:00-22:00 else 1400

I (immediately after posting my previous post) just dropped the memory settings from local.txt and edited in the CPU dialog to 500/1000 and then Primenet happily gave me a P-1 assignment.

[petrw1]

I have been delinquent while pursuing other goals. I will return shortly with at least 2 PCs.

[petrw1]

The current assignment summary reports shows hundreds of P-1's in the 30-35M range. Why do I not get assigned those when I ask for a P-1 assignment? Instead I get 50M+ assignments.

[Prime95]

Originally Posted by petrw1

The current assignment summary reports shows hundreds of P-1's in the 30-35M range. Why do I not get assigned those when I ask for a P-1 assignment? Instead I get 50M+ assignments.

P-1 assignments are in the 50M+ range so that when they are finished the exponent can be handed out for its final level of trial factoring.

[petrw1]

On my way...

1 Q9550 core starting beginning of next week.
1 PIV 3.4 starting early next week.
1 Q6600 core starting mid next week.

Possibly more to follow shortly ...

[Kevin]

Is a PIII@1.1 GHZ with 400MB assigned (out of 512MB) powerful enough to be worth putting on P-1? It's a server box that's more or less dedicated to crunching until I have the time to make Samba work, so not very much system overhead or concern for system responsiveness.

[James Heinrich]

Pentium III's efficiency seems to drop off noticably at larger FFT sizes (1024K and larger is only about 60-70% as efficient as 4K-20K). I'd expect a 400MHz P-III to throughput about 0.08GHz-days per day; a current P-1 assignment is around 4.75GHz-days (maybe a little less with 400MB RAM); therefore a current P-1 assignment would take about 60 days...

[Kevin]

Originally Posted by James Heinrich

Pentium III's efficiency seems to drop off noticably at larger FFT sizes (1024K and larger is only about 60-70% as efficient as 4K-20K). I'd expect a 400MHz P-III to throughput about 0.08GHz-days per day; a current P-1 assignment is around 4.75GHz-days (maybe a little less with 400MB RAM); therefore a current P-1 assignment would take about 60 days...

I know whatever it does, it will do it slowly. But will it be considerably less efficient at P-1 than it will be at factoring, given that we have an abundance of people doing factoring and a shortage of people doing P-1?

[cheesehead]

Originally Posted by Kevin

I know whatever it does, it will do it slowly.

... and it will, nevertheless, be valuable.

A speed of 10,001 (overall GIMPS progress-units) is greater than a speed of 10,000.

But will it be considerably less efficient at P-1 than it will be at factoring, given that we have an abundance of people doing factoring and a shortage of people doing P-1?

Seems to me that those two things (P-1 efficiency, and shortage of P-1 participants) aren't really related. A PIII will be just as efficient when lots of other systems are doing P-1 as when no other systems are doing P-1.

Your P-1 would be valuable either way. What counts more IMO is whether you are happy with your participation. (Okay, you might think your happiness would be related to your system's P-1 efficiency. But it's easier for you to decide to be happy about your PIII's P-1 contribution regardless of its efficiency than it is to make your PIII run P-1 any more efficiently than it already does. Remember: a speed of 10,001 is greater than a speed of 10,000.)

OTOH, you might think your happiness is related to how much difference your choice of assignment makes to GIMPS. As this thread already shows, there is a certain bottleneck (P-1 assignments) that matters only to the relative differences in progress of different GIMPS assignment types (but not to overall GIMPS progress).

Your P-1 participation, regardless of CPU comparison, would help relieve that bottleneck! That, in turn, would help GIMPS make more LL-only assignments available to those participants who prefer LL-only assignments to other types, perhaps increasing the happiness of those whose happinesses are (* alas *) dependent upon receiving LL-only assignments.

[Kevin]

Originally Posted by cheesehead

Seems to me that those two things (P-1 efficiency, and shortage of P-1 participants) aren't really related. A PIII will be just as efficient when lots of other systems are doing P-1 as when no other systems are doing P-1.

No, they're related.

Ultimately, throughput is maximized if every system does the task it's (relatively) the best at compared to other machines. If a modern computer has a ratio of 1:1 between time for a P-1 test and time for a TF test, and my computer has a ratio of 10:1, I'd be greatly under-utilizing my system by doing P-1 testing. However, since there is a shortage of P-1 testing, I'm willing to "under-utilize" my system to a reasonable degree if it puts CPU time where it's needed more. So the lack of P-1 testing is just saying that I have a looser standard for what is "considerably less" efficient.

I think what James was talking about is on the right track, and seems to be indicating I'd be better off keeping it on TF.

[James Heinrich]

I'm 80% done writing a tool that should give a quick overview for expected throughput on various types of assignments (e.g. machine of type <specs> at <speed>GHz can expect to process [x]GHz-days/day at TF (to various bit levels), [x]GHz-days/day at P-1 (at various FFT sizes, [z]GHz-days/day at LL (at various FFT sizes, etc) -- overall giving you a picture of what the most efficient type of work for this particular machine is. I'm not sure that I want to get into all the ugly complexities of analyzing work mix efficient on multicore machines, but at least looking at a per-core should give some idea of what the best focus is for most efficient use of that type of machine.