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

[cmpute]

Thanks for pointing out this. I have been using gpuowl stage1 + prime95 stage2 for quite a while (a week) with a throughput of 2h per exponent for the wavefront using B1=1.2M. I will be using similar bounds for this task. I expect similar throughput (maybe 2.5h per exponent), though I'm still testing the first exponent. The machine I put on this will have ~80GB RAM for mprime, with multiple CERT tasks completed in the past year, it's very reliable.

I would use GPU for stage1 because I want higher B1. Most of these exponents (if not all) already have B1=1.48M, I will be using 1.5M.

[cmpute]

Originally Posted by James Heinrich

That's 6218 exponents (2230 if you only count unfactored ones, but you didn't say that). Why do they all need a new P-1? How long would it take you to do better P-1 on 6k exponents?

I didn't mean to have all of them, just the ones in the list of this post . That's only 60ish exponents. Is there something misunderstood on my side?

[James Heinrich]

Originally Posted by cmpute

I would use GPU for stage1 because I want higher B1. Most of these exponents (if not all) already have B1=1.48M, I will be using 1.5M.

No reason you can't do a higher B1 in Prime95 too. Feel free to adjust the tests-saved upwards from 1 to something a bit higher (e.g. 1.3) to get the B1 you want.

[cmpute]

Originally Posted by James Heinrich

No reason you can't do a higher B1 in Prime95 too. Feel free to adjust the tests-saved upwards from 1 to something a bit higher (e.g. 1.3) to get the B1 you want.

Yes. I just want faster P-1 to cope with the time of B2. Stage 1.with gpuowl and B1=1.2M has roughly the same time of the time spent in mprime for Stage 2 at wavefront.

[kriesel]

Originally Posted by cmpute

The machine I put on this will have ~80GB RAM for mprime, with multiple CERT tasks completed in the past year, it's very reliable.

Cert assignments are typically very fast, faster than a P-1 with adequate bounds and both stages on the same exponent, which is in turn much faster than a primality test. Completing Certs does not mean the hardware is highly reliable. Error rate seems to increase with duration and memory footprint. ECC ram helps somewhat.
Absence of GEC errors logged during a long PRP run means the hardware is reliable. Changing ram means the hardware reliability may have changed, for the worse, drastically. I had a rock solid 16GB system (0 GEC errors per first test wavefront PRP), that became unreliable (multiple GEC errors per first test wavefront PRP) so unusable for P-1 when "upgraded" to brand new 64GB ram.
If you haven't yet, check for

"errors":{"gerbicz":

followed by a nonzero value in recent past wavefront PRP runs in results.json.txt for the hardware as currently configured.
Have fun!

edit: to balance gpuowl stage 1 throughput and mprime stage 2 throughput, it's actually advantageous to have stage 1 take a LITTLE longer than stage 2. Set mprime worker to get wavefront P-1 work occasionally to do a stage 1 also via PrimeNet API work assignment to make up the difference & maintain a day or two's work queued. Automatic, easy as it gets.

[cmpute]

{"status":"C", "exponent":116509073, "worktype":"PRP-3", "res64":"CD1AA8C3C713E997", "residue-type":1, "res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fft-length":6553600, "shift-count":52791147, "error-code":"00000000", "security-code":"5D398D49", "program":{"name":"Prime95", "version":"30.10", "build":4, "port":8}, "timestamp":"2023-03-22 03:05:35", "errors":{"gerbicz":0}, "proof":{"version":2, "power":8, "hashsize":64, "md5":"047eb0f7d9622888b9541b785ea34a10"}, "user":----, "computer":"MServer", "aid":"A7C7B5C70C23D6016FC908FEC95256FA"}

[Mark Rose]

I'm currently processing with a B1 just under 5M and a B2 between 62M and 78M depending on the machine.

I'd aim for a higher B1 than 1.5M.

[Prime95]

Originally Posted by cmpute

I would like to take
Pfactor=1,2,100137221,-1,77,1
to
Pfactor=1,2,100250957,-1,77,1

I won't get an assign ID immediately because I plan to use GPU for stage 1, is that ok?

Done. You could register the assignments with prime95 (add 'em to worktodo.txt) and then comment them out while gpuowl does stage 1. Or not, it's unlikely someone will come along and do P-1 while you are working on them.

[Mark Rose]

Originally Posted by Prime95

Done. You could register the assignments with prime95 (add 'em to worktodo.txt) and then comment them out while gpuowl does stage 1. Or not, it's unlikely someone will come along and do P-1 while you are working on them.

I currently have 320 exponents reserved for P-1 in the 101.9 to 102 range. I'll finish them before powering down for the summer. I have already completed P-1 on the exponents I had reserved between 100 to 101.9.

[kriesel]

Originally Posted by Mark Rose

I'm currently processing with a B1 just under 5M and a B2 between 62M and 78M depending on the machine.

Is that what mprime would optimize the bounds to on your hardware? Those are very different from what I'm seeing, and your reported limits don't seem to me to capitalize much on the increased stage 2 efficiency; I'm seeing optimization to sub-1M stage 1 and >150M stage 2 (both stages in prime95, ~60G allowed ram):
UID: Kriesel/emu, M106970327 completed P-1, B1=563000, B2=176294580, Wi4: ...
UID: Kriesel/emu, M117360923 completed P-1, B1=828000, B2=266878920, Wi4: ...

[kriesel]

+40:
Pfactor=1,2,100304389,-1,77,1
to
Pfactor=1,2,100397459,-1,77,1

And these 20:
Pfactor=1,2,101995753,-1,78,1
to
Pfactor=1,2,103912801,-1,78,1