The best work for my CPU

[davieddy]

Originally Posted by MacMagnus

A reduce by a factor of 200? Some lotteries have those odds, hehe.

I know it's difficult to tell whether I'm joking or not,
but (60/360)^3 ~ 1/216 if I'm not mistaken.

[TheMawn]

I highly doubt any BIOS setting will affect the way Primenet gives you results unless your "failsafe mode" is actually underclocking your CPU beyond the need for a heatsink...

It might be that the server is over-emphasizing trial factoring. Chalsall, who seems to be the go-to guy for the GPU computing side of things, recently expressed interest in getting all the exponents trial factored to 2⁷⁵ as opposed to the 2⁷⁴ everyone was kind of going for. If he's asked the server to assign TF to 75 bits, then by the server's standards, TF is WAAAAAAAAAAAAAAAAAAAAAAAY behind schedule, even though we're quite far ahead as far as factoring to 74 bits goes.

[davieddy]

Originally Posted by TheMawn

Chalsall, who seems to be the go-to guy for the GPU computing side of things, recently expressed interest in getting all the exponents trial factored to 2⁷⁵ as opposed to the 2⁷⁴ everyone was kind of going for.

Chalsall's mouth is bigger than his stomach and brain combined.

[kracker]

Originally Posted by TheMawn

Someone please correct me if I am wrong but is it possible that the server may assign TF just to test the CPU's reliability before giving it a LL test?

I believe you mean DC checks? If so, no I don't think so. TF will tell you nothing of cpu stability.

[chalsall]

Originally Posted by TheMawn

Chalsall, who seems to be the go-to guy for the GPU computing side of things, recently expressed interest in getting all the exponents trial factored to 2⁷⁵ as opposed to the 2⁷⁴ everyone was kind of going for.

I'm just one gear in a very complicated machine.

To be clear, I hope and expect that we can start going to 75 "bits" at 64M or so. But right now we're riding the wave going to 74.

[davieddy]

Originally Posted by chalsall

I'm just one gear in a very complicated machine.

To be clear, I hope and expect that we can start going to 75 "bits" at 64M or so. But right now we're riding the wave going to 74.

Incorrect.
You are just a spanner in the works of a very simple machine.

James has shown that a typical GPU could profitably TF to 75 before doing LL work.
As you agree, the limitation is currently the firepower available.

Up until 3 months ago, "we" were happily TFing everything >53M to 73 bits, and were getting comfortably ahead of the LL assignment wavefront.
This also left plenty of time for P-1, and could contemplate upping the "what makes sense" goal to 74 bits.
You argued that we were far enough ahead to do this immediately, even if the assignment front might catch up a bit before TF to 74 drew ahead again.

Despite my misgivings and BCP19's self-induced implosion, you elected (unilaterally) to "Suck it and see". Reasonable enough.

Realizing that TFing from 70 to 74 was not reaching the ~300+ per day, I suggested a "crackpot" reductio ad absurdum idea of backtracking TFto74 to 62M (thereby erasing your "TF is well ahead" argument).

You rose to the bait, and the result is farcical.

75 bits? Don't make me larf.
Maybe we can start another war in 2 years time.

David

[TheMawn]

Originally Posted by davieddy

Incorrect.
You are just a spanner in the works of a very simple machine.

James has shown that a typical GPU could profitably TF to 75 before doing LL work.
As you agree, the limitation is currently the firepower available.

Up until 3 months ago, "we" were happily TFing everything >53M to 73 bits, and were getting comfortably ahead of the LL assignment wavefront.
This also left plenty of time for P-1, and could contemplate upping the "what makes sense" goal to 74 bits.
You argued that we were far enough ahead to do this immediately, even if the assignment front might catch up a bit before TF to 74 drew ahead again.

Despite my misgivings and BCP19's self-induced implosion, you elected (unilaterally) to "Suck it and see". Reasonable enough.

Realizing that TFing from 70 to 74 was not reaching the ~300+ per day, I suggested a "crackpot" reductio ad absurdum idea of backtracking TFto74 to 62M (thereby erasing your "TF is well ahead" argument).

You rose to the bait, and the result is farcical.

75 bits? Don't make me larf.
Maybe we can start another war in 2 years time.

David

Holy crap. Let's just settle down... I've been trying to make sense of the info on GPU72.com and the PrimeNet summary, but it's been a bit difficult. I feel like we lack a consolidated summary of how many exponents are at which stage (i.e., 72, 73, 74, P-1, ready for LL, LLing and first LL finished).

Looking at 60M:

• Last week, 426 of 18,949 exponents were trial factored from 73 to 74 (and I think 6 were found to be composite). This gives 1,290 factored to 74.
• Presently, 13,910 exponents are being LL'ed. 14,476 exponents have not had any LL-done, which leaves 566 that are in need of LL but have not had any done yet.
• Of these 566, 153 are available for LL. This leaves 413. There are 409 assigned TF and 39 assigned P-1, and no work available in either category.

First off, it seems to me like the numbers aren't quite adding up. Second, it seems like a LOT of the 73 bits factored are being sent out for LL tests. Third, I can't tell how many exponents have had and/or need P-1.

This is all making it seem to me like Trial Factoring isn't keeping up.

[Uncwilly]

Originally Posted by TheMawn

This is all making it seem to me like Trial Factoring isn't keeping up.

Chris (Chasall) keeps track of what the GPUto72 throughput is. If it looks like the lead is diminishing too quickly, he will dial it back down by a bit. If it looks like TF is pulling way far ahead, he can up the bit level. It is quick for him to do.

[davieddy]

@ Mods.
Perhaps the last few posts could be moved elsewhere (excluding "Useless Posts"), where TheMawn and MacMagnus would be most welcome to continue the discussion.

Things are getting a bit "tasty" here, and I don't think the "front door" to Mersenneforum is the best place to continue it.

D

[kladner]

Despite my misgivings and BCP19's self-induced implosion, you elected (unilaterally) to "Suck it and see". Reasonable enough.

You modestly omit your helpful role in "pissing off Pete". Come on! Take pride in your work!

[chalsall]

Originally Posted by TheMawn

This is all making it seem to me like Trial Factoring isn't keeping up.

We are. Just (which is how it should be -- it means we're at the perfect balance of firepower / bit-depth).

Compare this report which shows over the last month we've TFed 8,963 candidates to 74 "bits", against this report which shows that approximately 8,906 candidates were LLed once.

Currently no candidate above 62M is being assigned without being TFed to at least 74 bits, and P-1'ed "well". In addition, we're taking a few candidates in the 60M and 61M ranges to 74 bits as the situation allows.