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

[tuckerkao]

Originally Posted by chalsall

I miss my Father. He was a difficult man. But the discipline helped me learn how to stand alone.

Any big achievement starts from small advancement at first. Bill Gates didn't become a multi-billionaire overnight.

Once I receive my new PC and hook it up, I should be able to complete most of the entry assignments with ease. Finish 500+ PRPs of M168,***,*23 maybe seem to be the hardest.

What I've enjoyed the most is the badge system which has been widely available on the Steam client of my video games. Wouldn't it be nicer to have the achievements of all the difficulties integrated with the Mersenne GIMPS too?

Iron Badge: Completed a P-1 task on 1 new exponent
Copper Badge: Completed a P-1 task on 10 new exponents
Silver Badge: Completed a P-1 task on 50 new exponents
Gold Badge: Completed a P-1 task on 200 new exponents
Platinum Badge: Completed a P-1 task on 1000 new exponents

Iron Badge: Found 1 new factor
Copper Badge: Found 10 new factors
Silver Badge: Found 50 new factors
Gold Badge: Found 200 new factors
Platinum Badge: Found 1000 new factors

Special Badge: Found a factor that is greater than 2^100

Diety Badge: Discover a new Mersenne Prime

[Uncwilly]

Originally Posted by tuckerkao

Platinum Badge: Completed a P-1 task on 1000 new exponents

Platinum Badge: Found 1000 new factors

What about the 10,000's of those? And what about the number FTC's and DC's? And what about the amount of ECM work? And what about the P+1 work? What about PRP-CF's? What about Certs?

[tuckerkao]

I only gave several examples of the ideas. The total # of the badges are always expandable, thus having a new set of the harder badges add in as the expansion pack has been very common on that Steam gaming console. If an objective has only been achieved by a very small portion of the users or players, then they are listed under the Deity section.

As far as I know, there have been only several top guys finished 10,000+ of the PRPs or P-1 or factors found which is an extremely small portion of the total # of active Mersenne users.

[chalsall]

Originally Posted by tuckerkao

I only gave several examples of the ideas. The total # of the badges are always expandable, thus having a new set of the harder badges add in as the expansion pack has been very common on that Steam gaming console.

IMHO, at best, Graduate level Directed AI. More likely, an inexpensive set of Humans.

Lots of training left to do to face Humans at this level.

I could, of course, be incorrect with this assessment... That's how this game works.

[tuckerkao]

Originally Posted by chalsall

More likely, an inexpensive set of Humans.

If each of them finishes 1 or 2 PRPs from the wave-front, a new Mersenne Prime will probably be discovered in the matter of 1 month or 2.

A lot of gamers have fast GPUs, thus recommended trial factoring or even +1 bit can be done.

[kruoli]

Originally Posted by Viliam Furik

Who is Mrs. F?

She is probably 50 cents higher than Mrs. E.

[Zhangrc]

Originally Posted by kriesel

TF: +6 bits above recommended levels; 23281 vs 364 (~64. TIMES what made sense)
Even if that was on an RTX30xx, at its exceptionally high ~50:1 "exchange rate" TF/DP hourly productivity ratio, giving
22917/50 = 458 GHD DP, it could have accomplished 40.% of a PRP/GEC/proof, something useful instead, in that wasted time.

See https://www.mersenne.ca/cudalucas.php?model=801, which shows that TF should be done to 2^81, so only +3 to +4 bit levels.

Originally Posted by kriesel

Waste of resources = 49.2% of PRP duration. Do it twice, and enough has been wasted to about fully PRP/GEC/proof-gen such a Mersenne number.

But he might just be interested in finding a factor, which is not the general case.

Originally Posted by kriesel

That is very basic information that has been publicly available for years or decades

Things can change. the information can be modified and the bounds can be adjusted.
And every user has the right to do whatever work they want, as long as it doesn't interfere with others.

For example, I run TF on GPU and PRP on CPU separately, and I run into several problems when running gpuowl (Fristly it was compatibility problems, then I found it difficult to let it write save files every time before shutting down, and large amounts of work could be lost; however, mfaktc write save files very frequently, so very little work could be lost).
I only run TF on GPU, so you could just consider the ratio as infinity to one.
If I wanted to maximize my chance of finding a Mersenne number, I would have used all the time it costs for a PRP test on my CPU, to TF a single exponent on GPU (which means 2^84 to 2^85 for 109M exponents).

[RomanM]

Originally Posted by Viliam Furik

Who is Mrs. F?

She is The Goddess Who Must Not Be Named) On the Earth and Discworld (by Terry Pratchett), also known as (again see the First rule), as Mrs. L.

[Dobri]

Originally Posted by RomanM

She is The Goddess Who Must Not Be Named) On the Earth and Discworld (by Terry Pratchett), also known as (again see the First rule), as Mrs. L.

Fortuna, Goddess of Luck, the rival of Fate, the one who gives you a chance even if you have been factored out. Oops, I just named her...

[RomanM]

Here is her Temple. There are also Prophets and Archbishops, followers and sinners) Perhaps there is even a Pope?

[kriesel]

Originally Posted by Zhangrc

See https://www.mersenne.ca/cudalucas.php?model=801, which shows that TF should be done to 2^81, so only +3 to +4 bit levels.

But he might just be interested in finding a factor, which is not the general case.

Things can change. the information can be modified and the bounds can be adjusted.

That chart's labeled as CUDALucas. Gpuowl is the state of the GPU primality testing art, for which about 1 bit less would be justified due to its greater iteration speed. PRP/proof is too, which reducing tests from 2 to 1 accounts for another 1 bit less justified. So, ~5-6 bits again, if doing both TF and PRP/proof on RTX3080. A better approach would be to use the RTX3080 for optimized TF that it excels at, and let CPUs and other GPUs with less of a DP performance penalty do the P-1 factoring and primality testing that they do well at. That more effective blend does not justify TF as high as pure-RTX3080 work for every phase of an exponent. So that makes the TF overfactoring even more extreme. For example, even taking the CUDALucas-labeled curves as valid for recent Gpuowl versions, for new and old GPUs, 1-test (labeled 2nd LL) on RX6900XT, 168M, 77 bits. https://www.mersenne.ca/cudalucas.php?model=807
Or more likely Radeon VII at 76 bits. https://www.mersenne.ca/cudalucas.php?model=752
or GTX1080Ti 78 bits https://www.mersenne.ca/cudalucas.php?model=706
average of those 3, 77.

1-test on RTX3080 doing TF at 168M, 80 bits. (NOT 81 as for 2 tests saved = "1st LL") New LL assignments ('1st LL') are no longer issued. Same goes for GTX1650 https://www.mersenne.ca/cudalucas.php?model=760

Split the difference between TF and PRP/proof GPUs, at 78, for the case of users prudently distributing work to GPUs for what they each do better. So, 84 is 6 too high, for 168M, 64 TIMES too costly, for optimal factoring effort to advance determining primality testing status, which after all is the stated purpose of GIMPS.
It's Great Internet Mersenne PRIME Search, not GIMFS (Factor search). https://www.mersenne.org/
And yes, a great deal of departure from efficient pursuit of the project's goals is tolerated to hopefully retain more participation than it costs.
If a factor hunter, TF or P-1 on another exponent would have been a more effective use of RTX3080 cycles than orders-of-magnitude-too-high TF effort. There is certainly no shortage of exponents to adequately TF or P-1. Going to extremes, since the TF effort is exponential in bit depth, one can spend more clock time on TF than would be used on optimal TF + optimal P-1 + PRP/GEC/proof combined. That happens around 85 bits at 168M for RTX3080, and yet leaves the primality still often undetermined.