When did PrimeNet begin rubberstamping results?

[Mark Rose]

Originally Posted by Aramis Wyler

I wonder if my cpu is more than 5 years old. Since it's an AMD, I didn't have to replace it last time I replaced my motherboard, and that was a long time ago. It's a phenom II x4, black box edition, but not overclocked (I don't think).

Those came out just over four years ago. I'm still running an Athlon II X4 640 at home. I built it cheap exactly 3 years ago.

[Brian-E]

My machine will be 7 years old in March. It runs mprime (usually DCs) while I am sitting at it, and I switch it off at other times. I am aware that running mprime does cause it to use more energy while I am using it than it would otherwise, but I don't think I'm being particularly irresponsible.

On principle I hate the idea that we should be considering banning anyone from taking part in the project provided that they are generally turning in reliable results and doing work which has been assigned to them. The open-to-all nature of GIMPS is one of its strongest features as a project in my opinion.

[lycorn]

Originally Posted by Brian-E

On principle I hate the idea that we should be considering banning anyone from taking part in the project provided that they are generally turning in reliable results and doing work which has been assigned to them. The open-to-all nature of GIMPS is one of its strongest features as a project in my opinion.

I fully agree with you. I took the liberty to stress what I see as an important commitment for all participants, though.

[chalsall]

Originally Posted by Brian-E

On principle I hate the idea that we should be considering banning anyone from taking part in the project provided that they are generally turning in reliable results and doing work which has been assigned to them. The open-to-all nature of GIMPS is one of its strongest features as a project in my opinion.

OK, I will concede the argument; I'm clearly in the minority.

But, for the record:

1. I was not suggesting banning anyone.

1.1. Under my suggestion v4 software would still be allowed to complete the assignments they had been given.

1.2. Anyone with older hardware would still be welcome to participate and be given new assignments, so long as they updated their software. "Fire and forget" is fine, but asking someone to update once after more than five years does not seem that unreasonable.

2. I continue to be concerned about how much energy / carbon emissions this project consumes / generates.

2.1. For some this doesn't matter; for example those who get their electricity from renewable sources, or benefit from the heat generated.

3. Many v4 clients are *not* reliable. matt_joe's machines, for example, often is assigned work which is then "sat on" for extended periods, or is worked on for a while and then abandoned.

4. I do think that v4 clients should *not* be given preferred assignments.

[petrw1]

Agreed...
Maybe though TF-LMH instead of DC. Then they only need to be "alive" for a few days or hours a day to complete work.

PS I can't speak for GPU software but only V24 of Prime95 allows TF for exponents <20,000 ... though one could ask "Why bother?".

[Uncwilly]

Originally Posted by chalsall

1.2. Anyone with older hardware would still be welcome to participate and be given new assignments, so long as they updated their software. "Fire and forget" is fine, but asking someone to update once after more than five years does not seem that unreasonable.

2. I continue to be concerned about how much energy / carbon emissions this project consumes / generates.
....
4. I do think that v4 clients should *not* be given preferred assignments.

1.2. Some of the machines in the past that I have used would not have benefited from later versions of Prime95. I have had machines that were slower with the more advanced versions. Model specific tweaks were eliminated, slightly larger code size extended beyond the first level cache, etc. (also, Athlons got less love back in the day.) Life cycle replacements of machines by "corporate entities" (businesses, non-profits, governments, etc.) is often ~5 years. I was recently (as in ~1 week ago) using an XPpro machine at my day job.

2. Many machines are on anyway and can serve as space heaters. If you want to worry about that too much, you can drive yourself crazy. Better to find your off-sets elsewhere (calculate the GHG load of a plane flight from your locale to Di$ney W0rld, etc.)

4. Why not give all v4 clients work that fits their reputation?
Those that continue to turn in valid results in a timely manner, let them have DC's near the bottom end (choice work).
Those that are a little slower, higher up DC's, near the middle of the pack. So they won't hold up milestones.
Those that are even slower, TF's from 79.3-100m, and when we get up in there with the LL wavefront, 100-120m.
Those that are buggy, no work.

[lycorn]

Originally Posted by chalsall

I do think that v4 clients should *not* be given preferred assignments.

That, yes, is definitely something to be implemented, IMHO.

[swl551]

1. I agree with all points made by Chasall on post #37
2. this conversation should be moved to its own thread as it is irrelevant to the original post.

[Prime95]

Originally Posted by Uncwilly

4. Why not give all v4 clients work that fits their reputation?

The unique identifier for each machine was a v5 invention. There is no (easy / foolproof) method of uniquely identifying the v4 computers.

In response to some other posts:

v4 computers and ANONYMOUS are not supposed to get preferred assignments. I've looked at the PHP code several times and I don't see where I went wrong -- but apparently I have a bug.

Also, as someone pointed out there are no speed improvements for upgrading these old machines. All optimization effort has been aimed at newer machines with AVX and FMA.

Let the machines run a few more years. I think in 3-5 years we can look into cutting these clients loose.

[NBtarheel_33]

Originally Posted by Prime95

v4 computers and ANONYMOUS are not supposed to get preferred assignments. I've looked at the PHP code several times and I don't see where I went wrong -- but apparently I have a bug.

No, I don't think it's a bug, just a case of "preferred today not necessarily being preferred yesterday". Keep in mind that the line between preferred and common exponents has been a moving target for at least the last year. Today's 30M exponent is last year's 25M exponent.

3-5 more years sounds perfectly reasonable. That would be a total of 8-10 years without updating. My guess is that many of these machines will have long dropped out of the project by then for sure, whether via obsolescence, breakdown, or replacement.

[NBtarheel_33]

Originally Posted by LaurV

I still dream that someone finds a missed prime, that would really be a hit!

That sure would make doublechecks a whole lot more popular and interesting. The DC and LL wavefronts would probably match in a matter of weeks.