OBD levels progress history

[kriesel]

Originally Posted by clowns789

The problem with levels 27-30 that you outlined is that it encourages the two P-1 stages to be separate and thus many of the second stages will likely be done on different systems than the first stages. I believe after Level 26 the project should split into a "wide" search for those who want smaller GPU work units to factor more exponents to lower bit depths, and a "deep" search with P-1 and PRP.

I think P-1 stages could be assigned separately, although it shouldn't be encouraged by the way the level system is structured; in other words, it shouldn't encourage participants to run 27 first stages before doing any second stages.

Well, first of all that assumes problems created by a level of success and interest at large work units that has not been much in evidence lately. Certainly not other than yours or mine.

Dozens of P-1 stage 1 would imply dozens of TF to 92 first.
I expect < 10 TF to 92 bits in 2022, perhaps <5.
And probably no more P-1 stage 1 or 2 than 4 in 2022.

The OBD subproject was at level 21.something IIRC when you took one exponent to 91 bits. There are still millions of OBD prime exponents available at 84 bits or lower. Other participants are running some on lower bit levels, despite the current levels system ignoring those.

PRP of OBD is such a large job that I anticipate none done through 2026. Probably longer. (If Mihai added sufficiently bigger fft lengths next month to gpuowl v7.x, it would take to about 2027 to complete OBD PRP / P-1 combined on Radeon VII. And who would dedicate several years of a fast GPU to one PRP? Who would dedicate a powerful multicore CPU to Mlucas on OBD for decades?)

[clowns789]

Originally Posted by kriesel

Well, first of all that assumes problems created by a level of success and interest at large work units that has not been much in evidence lately.

Fair enough. Certainly when the level system was created 17 years ago, P-1 and PRP were not considered, not to mention the latter didn't exist!

Luckily, we have some time to figure it out, and the mlucas benchmarking should help. As you've mentioned before, this work will likely aid in future programming efforts even if it is ruled that P-1 and PRP should be done consecutively on one machine.

[kriesel]

As of today, and since 2022-01-13, level 22 1/2.

[kriesel]

Today, level 22.21 -> 23.06
Level 23 = 23 p > 3321928095 have been TF to at least 2^89.

[kriesel]

As of today, level 23 1/2 passed

[kriesel]

As of today, 24.03

[kriesel]

It's been at this level for a while. The rate of progress has slowed for multiple reasons, including that each level is twice the work of the preceding, no one else has been working on OBD TF for a while, and much of my TF throughput has been shifted away from OBD, to MM127 mmff.