Considering current hardware on the status page

[cheesehead]

Originally Posted by petrw1

That being said, I think it would be a major effort to convert the current standings and a major mind shift to think in terms of new units ...

< snip >

On the other hand, converting the Status chart from P400 (boat anchors in waiting) to something like P3000 would be much easier to implement,

Why is multiplying by 1.04 a major effort, but multiplying by 7.5 (or whatever) much easier?

much easier to implement, to wrap our heads around and to keep current in the future assuming it continues to be updated every year or so as hardware changes.

Why is multiplication by 1.04, done once and never again, a major effort, but multiplying by a different factor every year or so is much easier?

[cheesehead]

Originally Posted by S485122

But that is exactly my point !

So, are you objecting to replacement of "P90 CPU-year" by "petaflop", or not objecting to that?

[S485122]

Originally Posted by cheesehead

So, are you objecting to replacement of "P90 CPU-year" by "petaflop", or not objecting to that?

I am not objecting to a Petaflops measurement ... if trial factoring and especially P-1 factoring would be given due credit and not only when a factor is found... If the new unit would be used using the old rules, there is no need to change it.

In my view there is a drawback : Petaflops are not entirely hardware independent, depending on the microcode on the processorand on the program, more or less flops are needed for the same mathematical treatment. This means that an improvement in hardware or in the program changes the credit while the same “work” is performed. One gets less credit per time unit because the program needs less flops to achieve the same result. This is compensated by the greater speed of the computers, but it gives a greater weight to “old” results and thus less incentive to join the search for credit hungry people. To carry this argument to extremes : suppose one would build an ASIC that could compute one LL iteration in one cycle, it would need N-2 flops to complete the LL test of N: a 40M exponent would earn 0,000 000 040 Petaflops. If the ASIC would be clocked at 4GHz, it would earn 4 Gigaflops in a second. A 4GHz P4 needs a bit more than 25 days to earn 8,5 P90 years or about 9 Petaflops, this is roughly equivalent to 4Megaflops a second. The first machine would be turning out 100 results a second compared to one result every 25 days for the second and both machines would earn the same credit !

[jasong]

/me puts on his dunce cap so that people know they need to check my data before trusting it.

Correct me if I'm wrong, but as the exponent in 2[exponent]n[/exponent]-1 increases by 50%, the time to do a test approximately doubles. I've always assumed that, unless the L2 cache is way out of it's league with a number, that if number B takes twice as long as number A on one machine, than it'll take twice as long on another machine.

Is this not, at least partially, correct?

[petrw1]

Originally Posted by cheesehead

Why is multiplying by 1.04 a major effort, but multiplying by 7.5 (or whatever) much easier?

Why is multiplication by 1.04, done once and never again, a major effort, but multiplying by a different factor every year or so is much easier?

"...a major effort...": for George or whoever manages the Statistics to convet all the reports to the new unit of measure.


"...easier to wrap our heads around...": (almost) no one has a P400 any more. I assume this column was added intentionally when the current PC was a P400 and most members had one. But not any more. This makes the 5.5 multiplier irrelevant to most of us. If for that same reason this column was kept current (i.e. P3000 now) with the corresponding multiplier and I have a 3Ghz machine then this column tells me instantly how many P90 years of work I will be doing every year without having to look at another chart. And if I have a 2.4 Ghz I still have a pretty good estimation of where I stand.

An analogy would be if your local car dealer advertised that if you buy the 2007 Toyota Camry your Gas Mileage will be 50% better than your 1976 Toyota Corona. Yes, I could look up the 1976 Mileage and convert but I could relate better to: "... your gas mileage will be 10% better than your 2005 Toyota Camry..." because odds are much better that if I am looking at the 2007 Camry I probably currently own a similar but slightly older model.

[cheesehead]

Originally Posted by petrw1

"...a major effort...": for George or whoever manages the Statistics to convet all the reports to the new unit of measure.

... but it's exactly, precisely, the same effort to multiply all by 7.5 as it is to multiply all by 1.04 -- and it's a new unit either way.

So, again, why unfairly portray one as being much more laborious than the other, when in fact the relative laboriousness is equal the first time, then reversed as soon as the 7.5 is replaced by the next CPU-to-CPU multiplier?

It's less effort to convert to petaflops once than to do multiple conversions every few years.

[petrw1]

1. I guess all I am saying is that if the P90 Credits/Points units changes then George has to change the Status page, and the Top producers reports, and, and, and ... anything that refers to P90 years.

But if we stick with P90 for scoring (because it is merely a scoresheet) even if just for historical reasons then we save George work.

2. Tell me if I am wrong here, but I assumed the P400 column was added later when that was the then current hardware so members could understand the relative P90 years contribution of their current P400 hardware. If this is wrong then my entire case for changing it is probably invalid too. If, on the other hand, my assumption is correct (or close) then the same reason that P400 was added then could justify updating that column to something more current now.

=================================

Anyway to point 1. above, right from my first post (and I started this crazy topic) I said it makes sense to stay with P90 years. Mind you "90" years from now when our grandchildren on GIMPS are running PCs of speeds we cannot even imagine now and are looking for QUADRILLION digits primes and have P90 Yrs points of a size that they themselves are 10 million digits long and that give them a headache when they try to view the Top Producers reports via the nano-nano virtual display chip implanted in their cornea ... then maybe George's grandson will want to chance from P90 years to ... to something that even dwarfs petaflops ...

And to point 2. above, I never expected it to generate such intense discussion. But, hey it earned me a "little blurb" above my avatar so it can't be a bad thing.

Respectfully ... and finally ...

[Paulie]

This has been discussed before, see this thread from 2003:
http://www.mersenneforum.org/showthread.php?t=16

The new server will fix everything. :

[petrw1]

Originally Posted by Paulie

This has been discussed before, see this thread from 2003:
http://www.mersenneforum.org/showthread.php?t=16

Oops, sorry I should have searched before I brought this "dead hourse" back to the land of the living.

I did find it interesting in that thread that George himself stated:

" ...I pretty much convinced myself to change to GHz-days ... "

mind you, that was 2002. A lot can happen in over 5 years. Wonder where his thinking on this topic is at now?

BUT I also like the suggestion that P90 be kept as a kind of historical honor.

Maybe it's time to bury this horse too.

[Paulie]

It's worth talking about, but I think in the context of the new server, not modifying the existing one.