Prepare the way! SRBase is coming for the next bit level.

[petrw1]

Originally Posted by drkirkby

1. Do P-1 with say B1=1000000, B2=10000000
2. Keep the save files.
3. Extend this to B1=1000001, B2=10000001 and get more GHz for doing virtually nothing.
4. Extend this to B1=1000002, B2=10000002 and get more GHz for doing virtually nothing.

What changed?
The following assumes save files exist.
When I did tests about 3 years ago:
If I increased B1 it continued Stage 1 where it left off and did a complete Stage 2.
If I increased B2 it did a complete Stage 2.

This was with Prime95 Pminus1.

Is this not the case anymore with PFactor; or GPUOWL; or the new version of Prime95?

[kriesel]

Originally Posted by petrw1

What changed?
The following assumes save files exist.
When I did tests about 3 years ago:
If I increased B1 it continued Stage 1 where it left off and did a complete Stage 2.
If I increased B2 it did a complete Stage 2.

This was with Prime95 Pminus1.

Is this not the case anymore with PFactor; or GPUOWL; or the new version of Prime95?

Drkirkby sometimes types and posts before he checks for facts or identifies assumptions. As do most of us occasionally.

My understanding is prime95 has some bounds extension capability. In undoc.txt:

By default P-1 work does not delete the save files when the work unit completes.
This lets you run P-1 to a higher bound at a later date.  You can force
the program to delete save files by adding this line to prime.txt:
    KeepPminus1SaveFiles=0

I've not used it.

Gpuowl won't extend bounds after the initial P-1 run is completed, even with all save files present. It won't complete a partial run if the bounds are changed midstream, producing an error instead.

[Viliam Furik]

Originally Posted by chalsall

It's about as useless as what you currently have your GPU TF'ing clients doing. But, at least, that sub-project has a tangible completion goal.

I wanted to disagree, only to come to the conclusion, that I must agree. That's pure logic - with the same starting set of conditions, it always results in the same outcome...

Factoring of Mersenne prime candidates mostly far beyond the current wavefront helps the project a lot, but only in the long run. Factoring of known composite Mersenne numbers far beyond the current PRP-CF wavefront helps the PRP-CF "project" a lot, but only in the far longer run, as most of the numbers will have to wait years, decades, centuries, even millennia for complete factorization. Some of them (with lower bound is defined by the maximum theoretical amount of computational power and theoretical unknown, potentially unexisting, polynomial-time factorization method, and the time the universe will live - assuming the numbers don't have to be factored by the human race only unless there aren't and never will be other civilisations) will never be computed.

[drkirkby]

Originally Posted by kriesel

For a little history, see https://mersenneforum.org/showthread...inc#post561290
Also the BOINC project list shows many occurrences of NVIDIA and/or AMD GPU support, including SRBase.

George Woltman created prime95 and GIMPS in early 1996.
BOINC's initial release was in 2002.
As I recall SRBase was created quite recently by comparison (~4 years ago? 7?)
GIMPS is not a byproduct, created out of, something GIMPS predates by ~20 years.

You totally misinterpreted what I meant. My wording was very poor. I was not trying to imply GIMPS was a by-product of SRBase.

IF the aim of SRBase was to find large Mersenne primes, then they are clearly going about it the wrong way. It would be far more sensible to trial factor a bit ahead of the wavefront, not 100s' of millions ahead, which will probably not need a primality test for many tens or perhaps even >100 years.

But looking on the SRBase website
https://srbase.my-firewall.org/sr5/
the aim of the project is in an area of mathematics I don't know about - Sierpinski / Riesel Bases. So I'm wondering if the results they provide to GIMPS are just a by-product of their main aim. Is it a case of

• They need to trial-factor these huge primes for the work on Sierpinski / Riesel bases.
• They realise that any factors found are useful to GIMPS, so they very kindly pass along the factors.

[kriesel]

Originally Posted by drkirkby

Is GIMPS not just a by-product of the real interest for SRBase?

I responded to what you ACTUALLY wrote. (In all its absurdity, given the timeline.) When you mean something else, you should write THAT something else. Take the time to post accurately. Posting misleading nonsense will be countered, by one user or another, or a moderator. It's more efficient to post accurately the first time, than to apologize or clarify later.

[Uncwilly]

Originally Posted by drkirkby

the aim of the project is in an area of mathematics I don't know about - Sierpinski / Riesel Bases. So I'm wondering if the results they provide to GIMPS are just a by-product of their main aim.

Maybe read up on SR and at least get an idea before going off on the tangent. Saying, "I don't know about quantum chromodynamics, but, maybe this" is no good either. To join in on the conversation intelligently requires some knowledge. Knowledge comes via input, like reading.

[VBCurtis]

Originally Posted by Uncwilly

To join in on the conversation intelligently requires some knowledge.

This here is the hurdle for drkirby. Blather on in posts until corrected, then claim that's not what he meant / he's misunderstood / we math people take things far too literally / state he's too busy to be bothered to read the forum (but not too busy to post EVERY (*%^%^(* DAY).

What is so hard about "Read more, post less"? Try it. Try it some more.

[rebirther]

Originally Posted by drkirkby

You totally misinterpreted what I meant. My wording was very poor. I was not trying to imply GIMPS was a by-product of SRBase.

IF the aim of SRBase was to find large Mersenne primes, then they are clearly going about it the wrong way. It would be far more sensible to trial factor a bit ahead of the wavefront, not 100s' of millions ahead, which will probably not need a primality test for many tens or perhaps even >100 years.

But looking on the SRBase website
https://srbase.my-firewall.org/sr5/
the aim of the project is in an area of mathematics I don't know about - Sierpinski / Riesel Bases. So I'm wondering if the results they provide to GIMPS are just a by-product of their main aim. Is it a case of

• They need to trial-factor these huge primes for the work on Sierpinski / Riesel bases.
• They realise that any factors found are useful to GIMPS, so they very kindly pass along the factors.

The main goal of SRBase was to solve all bases up to 1030 on Riesel and Sierpinski side. Trial factoring was only helping GIMPS.

[rebirther]

Originally Posted by chalsall

Hey... rebirther...

If we could please "press the reset button" between us...?

Might some of your compute clients be interested in helping out Wayne with his To Below 2K project?

It's about as useless as what you currently have your GPU TF'ing clients doing. But, at least, that sub-project has a tangible completion goal.

Absolutely no slight intended with that statement. Everyone's free to do whatever brings them joy with their own money, kit, time, and electrons.

Tell me more about the project. What apps are running, config, results. Iam planning later to get into other areas but only with GPUs while CPUs are used for the main goal of SRBase. Currently the hardware market is in a bad shape and graphic cards are rare or expensive. I hope it will be better soon with newer cards so we can increasing the output a lot.

[chalsall]

Originally Posted by rebirther

Tell me more about the project. What apps are running, config, results.

It's just TF'ing on low ranges, to get each 0.1M range below 2,000 unfactored candidates. Please see this thread where Wayne lays out the idea. Coordination happens there, so no "toes are stepped on".

I'm sure Wayne et al would appreciate the assistance. But please keep in mind that although the bit levels are low, the runtimes are high.

[rebirther]

Originally Posted by chalsall

It's just TF'ing on low ranges, to get each 0.1M range below 2,000 unfactored candidates. Please see this thread where Wayne lays out the idea. Coordination happens there, so no "toes are stepped on".

I'm sure Wayne et al would appreciate the assistance. But please keep in mind that although the bit levels are low, the runtimes are high.

If I can see the first post Iam currently in this 73-74bit range.