mersenneforum.org - Newer milestone thread

mersenneforum.org (https://www.mersenneforum.org/index.php)

- Data (https://www.mersenneforum.org/forumdisplay.php?f=21)

- - Newer milestone thread (https://www.mersenneforum.org/showthread.php?t=13871)

[QUOTE=NBtarheel_33;322401]It was November 8, 2008 - 4 years, 1-1/2 months ago - when [URL="http://www.mersenneforum.org/showpost.php?p=148293&postcount=30"]the first 50M exponent[/URL] was handed out for LL testing by PrimeNet. Note that back then, you could still grab exponents in the high 20Ms that needed a first LL. We've made (thanks in no small part to GPUto72) significant progress on the first LL tail.[/QUOTE]

[QUOTE=garo;322502]I've been following this project for over 10 years and at that time the leading wave of LL was progressing at about 4M per year and DC at 2M per year. We reached 50M a bit ahead of schedule because of the v5 switch over and exponents being assigned out of order during the transition.[/QUOTE]
The "4000/year" is something of a rough fluke, and if it constituted a valid "law of GIMPS progress" would convey the depressing notion that the expected rate of prime discovery was doomed to decrease indefinitely.

Fortunately, we have reached a milestone where a more reasonable and encouraging "law" kicks in: the wavefronts increase by say 10% per year.
This would yield (on average) a prime every four years, and require a
reasonable 33% in computing power per year.

How this may be achieved deserves a new post (or thread), but meanwhile, surely everyone agrees that the way new LLs get dished out ATM is somewhat shambolic!

:davieddy:

I'm eagerly waiting for mfaktc 0.20 as well. :smile:

[QUOTE=ckdo;256576]March 3, 2011: The trial factoring wavefront hits 82M.
March 24, 2011: The trial factoring wavefront hits 83M.

That's 27 days for 1M, and 21 days for 1M, respectively.[/QUOTE]

January 23, 2013: The trial factoring wavefront hits 80M (again).
January 25, 2013: The trial factoring wavefront hits 81M (again).
January 28, 2013: The trial factoring wavefront hits 82M (again).

It's about time to increase P-1 depth on all exponents, or some such, I'd say.

[QUOTE=ckdo;326622]January 23, 2013: The trial factoring wavefront hits 80M (again).
January 25, 2013: The trial factoring wavefront hits 81M (again).
January 28, 2013: The trial factoring wavefront hits 82M (again).

It's about time to increase P-1 depth on all exponents, or some such, I'd say.[/QUOTE]

What were the depths on the TF assignments? 69 in '11 and 70 in '13? So now these get handed off to GPU-2-7x, right?

[QUOTE=NBtarheel_33;326623]What were the depths on the TF assignments? 69 in '11 and 70 in '13? So now these get handed off to GPU-2-7x, right?[/QUOTE]
I am not very convinced, according with [URL="http://www.gpu72.com/reports/available/"]this report[/URL], of the fact that GPU72 handles out 82M assignment. PrimeNet activity summary reports also do not show too much things going on on eighties... That may be some babbling of PrimeNet, or maybe some guys requested few high TF assignments...

If so, then the front reached 1G exponents long ago, they are all factored to 65 bits or so.

[edit: correction: from[URL="http://www.mersenne.info/trial_factored_tabular_delta_30/1/0/"] this[/URL], indeed it seems like the front of the (horizontal) "Gauss bell" (moving down) reached the eighties in the last month. :whistle:]

Let me try to put this another way: Between 62M and 83M, there's currently 352,694 exponents waiting to be P-1 tested. Another 89,884 exponents are assigned for TF, and I assume that's TF to P-1 depth. So soon we will have 440,000+ exponents waiting to be P-1 tested, which is enough for several years - decades, eventually.

Meanwhile, GPU72 is rather busy taking the 61M range to 2^73, P-1 done or not.

It seems extremely reasonable, at least to me, to have the TF wave take everything to 2^72 (at least) starting at 62M, rather than shell out assignments in the mid-80s.

If the gap gets this big again, add another bit. Once enough bits have been added, consider skipping P-1 entirely, or at least running it at much relaxed bounds.

The TF wave is being run (mostly) by CPUs. For them, it is provably suboptimal to go to the higher depths. When GPU72 gets to that range (or if Primenet gets automated GPU support), it will go deeper.

Meanwhile, GPU72's work is creating well-TFed exponents which can be assigned for further P-1 / LL work. They're staying ahead of the LL wave (not sure about a P-1 wave).

Someday, all these TF-ing CPUs are going to run out of work -- maybe we can hand them small ECMs? But that day is ways off.

[QUOTE=axn;326638]Meanwhile, GPU72's work is creating well-TFed exponents which can be assigned for further P-1 / LL work. They're staying ahead of the LL wave (not sure about a P-1 wave).[/QUOTE]

We are staying ahead (to 73) of both the P-1 and LL waves. Unfortunately we should really be going to 74, but we don't have enough firepower at the moment, and it's better to take everything to 73 than only half to 74.

Also, unfortunately, the aggregate (both GPU72's and the standard Primenet) P-1 effort is not keeping ahead of the LL wave, so a few candidates are assigned without P-1 done. This means several candidates are getting sub-optimal or no P-1 effort before the LL test.

Edit: Also, to be clear, GPU72 is not assigning anything in the 70M or above ranges.

[QUOTE=chalsall;326669]
Also, unfortunately, the aggregate (both GPU72's and the standard Primenet) P-1 effort is not keeping ahead of the LL wave, so a few candidates are assigned without P-1 done. This means several candidates are getting sub-optimal or no P-1 effort before the LL test.[/QUOTE]
I would guess that most serious LL-testers would not find doing an adequate P-1 beforehand too much of a hardship.

D