Let's Optimize P-1 for low exponents. TL;DR in post #1. More in posts 60 and 61.

[petrw1]

Originally Posted by axn

TF bit depth -- in this case, the ECM equivalent.

I thought I had to leave off B2 for Stage 2 and let mprime choose it.
Are we choosing an excessively large TF to influence the B2 choice?

[DrobinsonPE]

Please reserve the 3.8M range for me.

I am going to start feeding my other slow computer with that range. Even though 3.8M is closer to 4M, I am going to start with B1 at 8.7M instead of 6.2M. About half the 3.8M range is TF to 69. Would it make sense to bring the exponents up to TF 70 first or just focus on P-1? My GPU can go from 69-70 in about 30 minutes per exponent.

[petrw1]

Originally Posted by DrobinsonPE

Please reserve the 3.8M range for me.

I am going to start feeding my other slow computer with that range. Even though 3.8M is closer to 4M, I am going to start with B1 at 8.7M instead of 6.2M. About half the 3.8M range is TF to 69. Would it make sense to bring the exponents up to TF 70 first or just focus on P-1? My GPU can go from 69-70 in about 30 minutes per exponent.

I'd say if your GPU is "bored" let it TF to 70 here.
It looks like there are enough to TF70 that your GPU can stay ahead of your CPU.

Have fun.

[petrw1]

Stage 1 at 5M is almost half of 10M and less than half of 2M for relative bounds.
So my 8-core is moving there now.
200K and 1.2M is on hold for now....maybe I'll get smarter at setting parms by then.

[axn]

Originally Posted by petrw1

I thought I had to leave off B2 for Stage 2 and let mprime choose it.

I believe, if you specify TF depth, it will ignore the given B2 and compute the optimal

Originally Posted by petrw1

Are we choosing an excessively large TF to influence the B2 choice?

We're trying to account for the ECM done on the exponent to more accurately do the probability/bounds calculation. Only considering the actual TF done is not accurate.
Now... ECM can't really be directly equated to TF, but since P95 bounds calculation only takes a TF depth as input, we try to give an equivalent. For that particular exponent, I guess George has estimated that ECM done is equivalent to about 117 bits of TF (i.e. it is unlikely, but not impossible, that there are any smaller undiscovered factors left).

For my runs in the 1.0M range, I'm using a TF-equivalent of 88 bits, even though actual TF is 71.

[Zhangrc]

Does this project involve TF?
I saw some exponents at 38.1M with TF done to 2^71, and it would be fairly easy to TF all of them to 2^73.
May I reserve that range for TF?

[petrw1]

Originally Posted by Zhangrc

Does this project involve TF?
I saw some exponents at 38.1M with TF done to 2^71, and it would be fairly easy to TF all of them to 2^73.
May I reserve that range for TF?

Yes that is mentioned in post 61.
Are you using GPU72/MISFIT?
If you do you will get them soon.
@Chris is bringing in TF work in the low 40's and soon in the 30's.

When he reads this post I expect him to bring in 38.1 soon enough.

[petrw1]

Originally Posted by Zhangrc

Does this project involve TF?
I saw some exponents at 38.1M with TF done to 2^71, and it would be fairly easy to TF all of them to 2^73.
May I reserve that range for TF?

@Chris has brought it into GPU72 for you.
Does that work for you?

[masser]

I don't know how to list this in the reserved ranges in the first post, but I have been and will continue to work on the "already factored" exponents between 27.1M and 34.2M. This includes P-1 and TF work types. My plan is to build up a large set of factored exponents that have been a) TF'd to 71 bits and b) have had some (targeting 5% factor probability) P-1 done. The goal will be to eventually check the cofactors for a world-record size PRP. The odds (of finding a PRP) are decidedly NOT in my favor, but it's interesting work. If others become interested in helping, maybe a group project (later, much later) could find a record-size PRP.

[Prime95]

I'll do 70K - 90K (or 0.07M - 0.09M).

I suggest Wayne copies reservations to the first post and deletes the individual post that asked for the range.
We may need a separate list of completed ranges to make selecting the next range easier.

[tha]

Originally Posted by Prime95

We may need a separate list of completed ranges to make selecting the next range easier.

It would be nice if the mersenne.ca progress visualization could be reworked into a tool for the new project stating the number of exponents that meet the new criteria for B1 and B2, amount of factors expected to be found if brought up to new criteria etc.