Is it forbidden to talk about manual testing strategies?

[Uncwilly]

Originally Posted by Dobri

If the forum expects from me to contribute immediately after being asked a specific question in a gray zone of inquiry, I would prefer not to fall in the trap but rather take time and reflect on it, use computer tools, and have a verified response in the future.

But, you brought up the topic and so you are being asked about it. You don't need to answer immediately. You could say, let me get back to you. But you didn't. You jumped into this "gray zone" with 2 left feet. There was no trap. You were handed rope and you tired the snare around your own foot. I suspect a moderator will give you "time [to] reflect on it". Go back and try to learn the lessons that were pointed out in the the other thread. This thread you have studiously chosen not to address the issues brought up.

Sit on your hands for a while.

[Dobri]

Originally Posted by S485122

Like so many you confuse "I will defend your right to say it", with "I am under obligation to help you say it and publish it". Most forumites will defend your right to see patterns without any rational ground. But that doesn't mean we are under an obligation to support numerological rants. I suppose there are forums where you will get away with numerology. Why not post there.

Just post some unacceptable content (according to the rules you accepted when getting a forum ID and posting here) and see what will happen to your login ...

Jacob

I am aware that this is a borderline topic. My intention is to treat the matter as a combinatorial problem and this is the reason not to share now more work in progress that would only open a new can of worms rather than gain acceptance.

[drkirkby]

Originally Posted by Dobri

Your dog would serve as an excellent assistant to randomly select exponents.
Manual testing is concerned with Category 4 assignments and they take much longer to test indeed (years per exponent).

Manual testing is not limited to category 4. According to the rules at
https://www.mersenne.org/thresholds/
category 2 is available for manual testing, although I admit my attempt to get a category 2 exponent for manual testing failed.

However, if you know the exponent you want, and have a semi-reasonable computer, it should be possible to get the server to assign it to that computer. Just as a test, I did a quick search for a Mersenne exponent, with a twin-prime below it. I aimed for a category 1 assignment. Sure enough, I just reserved M108201493 by putting the appropriate entry in worktodo.add, then executing mprime -c.

Exponents around 333 million don't take years to test. I have only been on here since January, and tested M332646233 in about 3-months of very sporadic testing. Sometimes it was not running at all, and sometimes on only one core. Doing M332646233 was enough to convince me that the exponents are too time-consuming, but it is not years - even on a fairly modest computer it would not take years. I reckon I could do them at a rate of about 1 every 10 days, but still 10x as long as first-time tests of the smallest exponents.

BTW, I will unreserve M108201493. I'm fairly confident the twin-prime thing is just coincidence, and the exponent is much bigger than I would want to test. I've got a few under 105 million to test, so are not going to bother with one over 108 million.

[Dobri]

Originally Posted by drkirkby

... but it is not years - even on a fairly modest computer it would not take years.

It seems that there are some users who still use i3 and i5 CPUs for LL testing of large exponents.
There was one user who three years ago completed a 6-year test of an exponent in the 100M-digit range.

I myself chose to experiment with 999,996,073 in an i7 10th Gen machine and the estimate is that it would take 1,945 days for the PRP to be completed. So I cannot fit within the 1-year time frame and will apply for an extension, then continue further the test till the end even if the exponent is re-assigned.

This way I am also testing the SSD memory and so far the PRP test was able to recover from one Gebricz/double-check error.

For comparison, I never had any errors when using hard disks.

[kriesel]

Originally Posted by Dobri

Your dog would serve as an excellent assistant to randomly select exponents.
Manual testing is concerned with Category 4 assignments and they take much longer to test indeed (years per exponent).
Anyway, using empirical strategies or not, the amount of work done by the volunteers would remain the same, so the GIMPS performance would remain unaffected.

Drkirkby already debunked somewhat the first of your provably false claims highlighted above. Manual testing is specified as applicable to Cat 2, 3 or 4.

Since you did not specify first test vs DC, I checked both, referring to https://www.mersenne.org/thresholds/
Cat4 minimum first test 7/18/21: above 112840494
add 100K: 112940494
assigned exponent above that: PRP=aid,1,2,112941749,-1,77,2
Gpuowl V6.11-380 ETA:

2021-07-18 10:58:39 asr2/radeonvii4 Proof using power 9
2021-07-18 10:58:44 asr2/radeonvii4 112941749 OK     4000   0.00%;  886 us/it; ETA 1d 03:48; ef6f162ded98f08d (check 2.00s)

Cat4 minimum DC 7/18/21: above 63748134
add 100K: 63848134
assigned exponent above that: PRP=aid,1,2,63848203,-1,74,0
Gpuowl V6.11-380 ETA:

2021-07-18 10:16:53 asr2/radeonvii3 63848203 OK     2000   0.00%;  660 us/it; ETA 0d 11:42; a7b1720661156407 (check 0.79s)

Both the above test exponents are Cat 4, which you claimed take years. Their ETAs computed by gpuowl indicate they will be done tomorrow, and tonight, respectively. A 100Mdigit primality test is ~15 days on a Radeon VII GPU.
Not even exponents ~10⁹ require 6 months, much less years, on a Radeon VII.
And those ETAs are for GPUs that are operating at reduced power for power efficiency, and a model (Radeon VII) that's no longer in production. There exists at least one GPU model with ~2.5 times the primality test computing speed rating of a Radeon VII, which could complete the 112.9M Cat 4 primality test in hours.
Your claim of years per exponent is only rarely anecdotally supported with badly misguided allocation of resources to much higher exponents, if at all. Don't bother claiming GPUs are a rare exception; even EWMayer and Prime95, authors of CPU oriented GIMPS software, are doing substantial primality testing on GPUs. And Uncwilly has already debunked the CPU side of your misleading claim. Such severe inaccuracy in a posted easily checked claim, ~10⁴:1, casts considerable doubt on the rest of your posted material.

Attempting as you appear to do, to divert computational resources from where it is most productive (lowest available exponents), to where it is less productive (higher exponents selected via unproven or fallacious reasoning), based on a voluminously advocated yet fundamentally unsupported claim masked by unpersuasive allegations of grounding in statistical analysis of a very small data set, and similarly divert forum participants' and moderators' time with your too frequent, argumentative, and poorly reasoned posts, acts to reduce GIMPS productivity. Stop.

My cat does not make such erroneous claims as you do. Has the good sense to not say anything, when she has nothing germane & accurate & constructive to say.

You've steadfastly reduced perceptions of your credibility to a good approximation of zero, been determined a troll/spammer, and earned a spot on my mersenneforum ignore list setting. Few have, in years.


Moderators: how much more rope are you going to allow Dobri?

[Dobri]

What I wrote about can be easily verified in the server database.
Also, I do not troll or spam anyone. You will not find my posts anywhere else except in the post I created and in the Lounge.
Despite all of your negativity, you are not on my ignore list.
Wishing you all the best, I will not post anymore.

[RomanM]

«De tous ceux qui ont traité cette mautîè, c’est sans contredit M. de la Condamine qui l ’a fait avec plus de succès. Il est déjà venu à bout de persuader la meil­leure partie du monde raisonnable de la grand utilité de l’inoculation: quant aux autres, il serait inutil de vouloir employer la raison avec eux: puisqu’ils n ’agissent pas par principes. Il faut les conduire comme des enfants vers leur mieux...»
(Daniel Bernoulli, 1760)

[xilman]

Originally Posted by charybdis

Before you immediately accuse someone of malice, stop and think whether there might be any other explanation for what they've done.

One should always apply Hanlon's Razor in my view.

[axn]

I'm closing this thread. I will move some of the more productive posts into a separate thread. Link will be posted here

EDIT:- Move posts are available here: https://www.mersenneforum.org/showthread.php?t=27024

[Dobri]

Originally Posted by kriesel

...claimed manual assignments were cat 4 and took (required as a minimum) years to complete...

The exact sentence was
"Manual testing is concerned with Category 4 assignments and they take much longer to test indeed (years per exponent)."

It was written in the context of users with modest computational resources who would attempt to potentially explore the entire remaining 9-digit exponent range.

It was not intended to mean "(required as a minimum) years to complete." Obviously, users having the latest state-of-the-art computers could complete 332xxxxxx assignments for less than a year.

The previous thread was intended for volunteers who could be students thinking how to pay their student loan or hard working individuals who have to provide for their families and cannot afford expensive CPUs, and yet are genuinely interested in GIMPS.

For a modest computer, the weighted time average is years for the entire Category-4 range till M999999937.

Even Kriesel currently has some attempted assignments that are scheduled to be completed in 2022, 2023 and 2024 (whatever the reasons are, there is no need to speculate), see <https://www.mersenne.org/assignments...xtf=1&excert=1>.

[slandrum]

Originally Posted by Dobri

The previous thread was intended for volunteers who could be students thinking how to pay their student loan or hard working individuals who have to provide for their families and cannot afford expensive CPUs, and yet are genuinely interested in GIMPS.

I have a many years old laptop that was cheap when I bought it, came with Windows 7 (was force updated to Windows 10 years ago back when Microsoft forced that on millions of unsuspecting people), but it is not years to complete a cat 4 exponent. It would be on the order of 5-6 months to do a fresh cat4 FTC PRP assignment. If I was silly enough to go out and look for much higher exponents with that machine, it could take years. I see a lot of people using less capable equipment than my old cheap laptop, but I really would not expect such equipment to generally be what students would have at their hands.