Getting others to do the work on exponents I like (was: Trial Factoring Progress)

[tuckerkao]

Originally Posted by James Heinrich

I'm not everyone
What is your ultimate output (e.g. Factor= worktodo lines)? What exponents do you want to include/exclude based on which criteria (exponent range, current TF, etc).

It'd be much nicer if an additional column can be integrated on the right side to show the "PRP Test Status" as Yes or No.

https://www.mersenne.org/report_fact...99&tftobits=72

factor=168554021,76,77
factor=168554039,76,77
factor=168554053,76,77
..............
factor=168554861,76,77
factor=168554963,76,77
factor=168554993,76,77

exclude M168554723


https://www.mersenne.org/report_fact...99&tftobits=72 <- I'd like to see that M168641923 with a "Yes" on the PRP Status column of the data page.

With all the M168,***,*23 exponents showing, the PRP Test Verified of Yes or No can be really useful for me to find the next target to work on - https://www.mersenne.org/report_fact...99&tftobits=72

~[Exponents]~ PRP: Verified
M168306023 -:- Yes
M168316123 -:- No
M168326623 -:- Yes
M168336323 -:- No
M168346723 -:- Yes
M168366823 -:- Yes
M168386423 -:- Yes
M168396223 -:- No
M168396523 -:- Yes
M168396923 -:- No

[James Heinrich]

Originally Posted by tuckerkao

It'd be much nicer if an additional column can be integrated on the right side to show the "PRP Test Status" as Yes or No.
I'd like to see that M168641923 with a "Yes" on the PRP Status column of the data page.
the PRP Test Verified of Yes or No can be really useful for me to find the next target to work on

I have modified the report to give you options to include/exclude based on (Assigned, Has P-1, Has LL, Has PRP). I think that should meet your requirements, if not let me know what's missing.

[tuckerkao]

Originally Posted by James Heinrich

if not let me know what's missing.

It has basically met my needs. Only 1 minor bug that the LL test of M168288023 was not detected.

Thanks a lot, I can see the whole list very clearly - https://www.mersenne.org/report_fact...=79&bits_hi=84 - the total quantity of the verified M168,***,*23 is still very minority of all the available candidates.

[James Heinrich]

Originally Posted by tuckerkao

Only 1 minor bug that the LL test of M168288023 was not detected.

Thanks, fixed.

[tuckerkao]

Originally Posted by petrw1

Good to hear from you.

You mentioned that TF worked ended a few months ago but in fact that was only for the 74 bit range.
If you look here and notice where the yellow band is you will notice that eventually your range will be factored up to at least 79 bits, probably more because GPUs are SOOOO fast. You can expect to see the TF crew back in your range early in the New Year. Hopefully, you finish before they get there.

Here's the fast my best GPU is. Take a look when you get a free time.

1 M168M exponent was chosen for the trial factoring with my name displaying on the background Windows desktop of the screenshot below to prove it's my GPU.

[James Heinrich]

Originally Posted by tuckerkao

M168M exponent was chosen for the trial factoring with my name displaying on the background Windows desktop of the screenshot below to prove it's my GPU.

Even better, you should actually reserve the exponent for TF on mersenne.org

[tuckerkao]

The reason of the failure to generate the proof file of my only problematic PRP of M168286523 remained a mystery. Since then, I've not experienced this situation again, all later PRP results got certified including M168572023 which was finished earlier today.

The screenshot on the bottom of this post shows that my Ryzen 7950 CPU is capable passing 13 Gerbicz error checks on M168426023, so all good now.

Originally Posted by Uncwilly

Can do and makes sense to do are not always the same. We are more interested in what makes sense.

The definition of "What Makes Sense" term and its coverage change from time to time.

Example: The definitions of a man and/or a woman from the history of a century ago and the debates going on within the U.S. government about this topic during year 2022. Consider if someone owns a time travel device and can bring the lengthy debates of 2022 to the people lived back in 1922, how the 1922 people would react, probably a lot of similarities compare to the dozenal vs decimal conversions.

Running trial factoring of a specific M168M exponent from 2⁸² to 2⁸³ will take only around 10 hours and 45 minutes on my fastest GPU today. However the time duration requires to run the PRP test of a specific M168M exponent using Radeon VII or Radeon RX 7900 XTX have not been shortened by the same ratio.

[Uncwilly]

[James Heinrich]

Originally Posted by tuckerkao

Running trial factoring of a specific M168M exponent from 2⁸² to 2⁸³ will take only around 10 hours and 45 minutes on my fastest GPU today.

You have already TF'd this exponent to 84, which is 6 bitlevels over the target value of 78, for roughly a 7% chance of a factor, spending about 23000 GHz-days. For the same effort you could have taken dozens, maybe even hundreds, of similar exponents from their current to their target TF level for a significantly higher chance of finding a factor (and finding a factor means saving at least one PRP test).

You seem very keen to toe a line on P-1 bounds (despite Prime95's better judgement) and then totally ignore the recommendations on TF. Your hardware of course, but what you're doing is not optimal for GIMPS.

[tuckerkao]

Originally Posted by James Heinrich

You have already TF'd this exponent to 84, which is 6 bitlevels over the target value of 78, for roughly a 7% chance of a factor, spending about 23000 GHz-days. For the same effort you could have taken dozens, maybe even hundreds, of similar exponents from their current to their target TF level for a significantly higher chance of finding a factor (and finding a factor means saving at least one PRP test).

You seem very keen to toe a line on P-1 bounds (despite Prime95's better judgement) and then totally ignore the recommendations on TF. Your hardware of course, but what you're doing is not optimal for GIMPS.

My only problem is that my fingers are not fast enough to keep up with the TF speeds of those 2⁷⁴ to 2⁷⁶ entries on 1 Nvidia Geforce 4090 and 1 Nvidia Geforce 3080. I need at least 1 to 2 TF entries of 2⁷⁹ to 2⁸⁰ per day to keep my worktodo.txt not so long. Maybe you can help me on how to reserve all the M168M exponents that still not yet reached 2⁷⁶ within several mouse and keyboard clicks.

I'm not sure how to find the methods that SRBase had quickly mass reserved everything. This has been the reason why I mostly only work on the M168M exponents, so poaching don't keep happening because total numbers of active M168M assignments of other users are usually less than 10 besides my own.

The exponents that have been brought up to 2⁷⁹ or higher are mostly in the format of M168,***,*23, so they stand out from other exponents in the same range: https://www.mersenne.org/report_fact...=79&bits_hi=84

I also don't know how Kriesel uploads the PRP proof files from GPUOwl, so I just use Prime95 as it's automatic.

[James Heinrich]

Originally Posted by tuckerkao

I need at least 1 to 2 TF entries of 2⁷⁹ to 2⁸⁰ per day to keep my worktodo.txt not so long.

No, you don't need any assignments to 80, you need assignments to 78, no higher. If you can do 2 to 80 in one day, you can do 8 to 78 with the same effort. There's no need to fetch and report results every day, just queue up 60 at a time and report results once a week.

Originally Posted by tuckerkao

I mostly only work on the M168M exponents, so poaching don't keep happening.

Poaching is what happens when someone works on assignments without reserving them, as you apparently often do. Please reserve everything you want to work on.

It's very easy to get registered assignments:
1) Go to https://www.mersenne.org/manual_gpu_assignment/
2) Choose how many assignments you want (based on what you described, you need 50-60 per week)
3) Enter exponent range (168000000 to 169000000 if that's the range you like)
4) For 168M range, enter 78 in the factor-to field. Not higher.
5) Click "Get Assignments", and there you go. TF them all, report results, and repeat once a week.

Originally Posted by tuckerkao

I also don't know how Kriesel uploads the PRP proof files from GPUOwl, so I just use Prime95 as it's automatic.

Just put the .proof files in your Prime95 folder and it will upload them automatically.