Oops - New Prime! (M49 related)

[chalsall]

Originally Posted by Madpoo

I'm still not liking the baseline=0 for the tflop graph. LOL

Nor I. I think it was better before since there was higher resolution of the deltas...

Since it's a linear graph, just imagine (in your minds eye) where the zero line is below the actual graph.

[AG5BPilot]

There's a really good article on page A3 of the New York Times this morning:

http://www.nytimes.com/2016/01/22/sc...imes.html?_r=0

[kladner]

Originally Posted by AG5BPilot

There's a really good article on page A3 of the New York Times this morning:

http://www.nytimes.com/2016/01/22/sc...imes.html?_r=0

Dr. Cooper said, however, that the computer would be set aside for posterity, like the ones that had made the three earlier discoveries.
“It’s kind of a dumb computer,” he said. “It doesn’t know it’s so popular.”

[only_human]

Originally Posted by AG5BPilot

There's a really good article on page A3 of the New York Times this morning:

http://www.nytimes.com/2016/01/22/sc...imes.html?_r=0

I agree. Simply explained, no obvious errors, extra details about about the computer lab setting the computer aside for posterity. Very nice.

[ATH]

I really wished we had researched that Curtis was running 4 numbers at once on that computer before the press release. Now everyone thinks the tests takes 1 months on any normal computer.

[firejuggler]

ah yes, but it has a nice side effect : if it take only 2 week instead of 1 month, then people will feel encouraged. and maybe stick a while longer

[airsquirrels]

Originally Posted by ATH

I really wished we had researched that Curtis was running 4 numbers at once on that computer before the press release. Now everyone thinks the tests takes 1 months on any normal computer.

Well it does the way prime95 launches out of the box defaults. Most users are going to get one worker configured per core.

[only_human]

Has the time to complete a LL test stayed roughly constant over the years? That is have the exponent size increases of typical tests been roughly matched by hardware improvements?

[Madpoo]

Originally Posted by ATH

I really wished we had researched that Curtis was running 4 numbers at once on that computer before the press release. Now everyone thinks the tests takes 1 months on any normal computer.

Well, that's probably how most computers would be running things. Wouldn't that basically be the default settings for a new install? One worker per core?

It's been so long since I've looked at the default setup, I actually don't know the answer.

[petrw1]

Originally Posted by only_human

Has the time to complete a LL test stayed roughly constant over the years? That is have the exponent size increases of typical tests been roughly matched by hardware improvements?

Seems YES: My first Pentium 4 from about 22 years ago took 1 month for a LL test in the current Leading Edge 33M range.

[Madpoo]

Originally Posted by only_human

Has the time to complete a LL test stayed roughly constant over the years? That is have the exponent size increases of typical tests been roughly matched by hardware improvements?

I would hazard a guess and say that's probably the case. Anecdotally speaking, I seem to recall that back in 1997'ish it was taking current desktops around a month to finish testing exponents in the 3M-5M range.

Does that sound about right? Well, that was my recollection anyway.

Might be an interesting analysis... a look at the average spread between assigned date and completion date, over time. Trouble is that only work since 2007/2008 when the new server code went into production has the original assignment date saved. Before that all we have is the completion date. So there's a paucity of data points beyond 8 years ago.