mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2015-04-27, 11:20   #67
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

100010110112 Posts
Default

239 TFLOP/s, an increase of 7.66% from the same 30-day period last year.
NBtarheel_33 is offline   Reply With Quote
Old 2015-04-27, 16:24   #68
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

63128 Posts
Default

Quote:
Originally Posted by NBtarheel_33 View Post
239 TFLOP/s, an increase of 7.66% from the same 30-day period last year.
I'm wondering if I'm skewing the results with the triple-checks I'm doing.

Now, I'm not saying that little old Madpoo could make a big difference, but I have been re-checking a bunch of smaller exponents (and a few big ones), and before that I hadn't really been doing any LL work at all.

Since the beginning of the year I've checked in 10630 results for a total of 58727 GHz Days. 35340 GHz-days of that just this month alone. 12338 GHz-days in the past 7 day period.

A good chunk of the total results were from me triple-checking all exponents below 1M. Very little GHz-days involved there (128 GHz-days) relative to everything else.

The project probably isn't too interesting for most folks (triple checking work where the same user did the first/second checks), so you may or may not want to consider the throughput of it as equal to actual first/second checks, factoring, etc.

So, adjust mentally as needed.
Madpoo is offline   Reply With Quote
Old 2015-04-28, 19:42   #69
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5×223 Posts
Default

Quote:
Originally Posted by Madpoo View Post
I'm wondering if I'm skewing the results with the triple-checks I'm doing.

Now, I'm not saying that little old Madpoo could make a big difference, but I have been re-checking a bunch of smaller exponents (and a few big ones), and before that I hadn't really been doing any LL work at all.

Since the beginning of the year I've checked in 10630 results for a total of 58727 GHz Days. 35340 GHz-days of that just this month alone. 12338 GHz-days in the past 7 day period.

A good chunk of the total results were from me triple-checking all exponents below 1M. Very little GHz-days involved there (128 GHz-days) relative to everything else.

The project probably isn't too interesting for most folks (triple checking work where the same user did the first/second checks), so you may or may not want to consider the throughput of it as equal to actual first/second checks, factoring, etc.

So, adjust mentally as needed.
The amazing thing is that I don't think that even that much work would skew the numbers that badly. Let's round things off and say that you've contributed an extra 60,000 GHz-days in the first four months (or 120 days) of 2015. That would account for a total of 120 TFLOPS over 120 * 86,400 = 10,368,000 seconds, or 11.574 TFLOP/s sustained over those four months. Over the last month, you've sustained 70.68 / (30 * 86,400) = 27.269 TFLOP/s. That *might* be inflating things a little, but even without it, we'd still be around 215-220 TFLOP/s.

And the fact of the matter is that triple-checking those exponents is very much worthwhile. Many of them were likely checked (and double-checked) on systems that were one-tenth or one-twentieth as powerful as my smartphone, with very early versions of Prime95. It is comforting (and bodes well for the validity of the results of much larger tests) to see modern systems running modern versions of Prime95 confirm these results.

Also, you have to think of it this way: suppose your extra 27+ TFLOP/s had not been available this past month. Your work would have had to have been completed by systems that would have been pulled away from other work. Thanks to your extra resources and willingness to contribute them, GIMPS benefited from your triple-checks without having to sacrifice newer results.

Did I read you correctly that *every* prime exponent below 1M was LL tested in a total of 128 GHz-days? That is amazing. I suppose all of our work will be triple-checked with a similar (almost absurd) speed advantage in the years to come. I bet GIMPS' first twenty years of work could be re-done in 3-5 years.
NBtarheel_33 is offline   Reply With Quote
Old 2015-04-28, 22:48   #70
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

2×1,637 Posts
Default

Quote:
Originally Posted by NBtarheel_33 View Post
Did I read you correctly that *every* prime exponent below 1M was LL tested in a total of 128 GHz-days? That is amazing. I suppose all of our work will be triple-checked with a similar (almost absurd) speed advantage in the years to come. I bet GIMPS' first twenty years of work could be re-done in 3-5 years.
Not every exponent below 1M, but every exponent that had merely been double-checked. There were a good few that had been triple-checked already.

I also triple-checked any that had a pair of matching 64-bit residues and also an older style smaller residue. (8-16 bytes or whatever).

I don't remember for sure... it was somewhere in the neighborhood of 10K exponents? I posted about it somewhere here with the details.

I was tempted to keep going and do the same for everything between 1M and 2M but I wanted to get back to triple-checking the self-verified work. Who knows, maybe once that's done I might. Just kinda see where it goes.
Madpoo is offline   Reply With Quote
Old 2015-04-29, 00:09   #71
cuBerBruce
 
cuBerBruce's Avatar
 
Aug 2012
Mass., USA

2·3·53 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Not every exponent below 1M, but every exponent that had merely been double-checked. There were a good few that had been triple-checked already.

I also triple-checked any that had a pair of matching 64-bit residues and also an older style smaller residue. (8-16 bytes or whatever).

I don't remember for sure... it was somewhere in the neighborhood of 10K exponents? I posted about it somewhere here with the details.
Speaking of 10k exponents, I'll mention again that I once computed the 64-bit residues for all of the first 10,000 odd prime exponents, even those that had no LL test in GIMPS because of having been factored. So that's up to an exponent of 104,743. I used my own little C program using the GMP library.
cuBerBruce is offline   Reply With Quote
Old 2015-08-09, 07:30   #72
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5·223 Posts
Default New record highs for 30-day throughput

242 TFLOP/s with a potential 619 TFLOP/s over the last 30 days. Both of these are record highs.

With this post, I am honored to enter the realm of the four-digit post count.
NBtarheel_33 is offline   Reply With Quote
Old 2015-08-09, 17:53   #73
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

6BF16 Posts
Default

Quote:
Originally Posted by NBtarheel_33 View Post

With this post, I am honored to enter the realm of the four-digit post count.
It looks like you don't do an unnecessary amount of talking
TheMawn is offline   Reply With Quote
Old 2015-08-18, 06:04   #74
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5·223 Posts
Default

Quote:
Originally Posted by TheMawn View Post
It looks like you don't do an unnecessary amount of talking
Actually, it has just been the most unexpected events that have kept me busy throughout 2014 and most of 2015, that have diverted my attention from GIMPS and the forum. I am hoping to resume my usual level of GIMPS contributions and posting activity in the near future.
NBtarheel_33 is offline   Reply With Quote
Old 2015-09-18, 20:30   #75
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5·223 Posts
Default GIMPS breaks the 1/4-petaflop barrier!

251.375 TFLOP/s sustained over the last 30 days. Congratulations GIMPS!
NBtarheel_33 is offline   Reply With Quote
Old 2015-10-04, 10:30   #76
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5·223 Posts
Default

A new high: 262.046 TFLOP/s (30-day sustained). We are also close (within 1%) to GIMPS' first-ever breach of 1,000,000 GHz-days (or 1 Petahertz-day!) of work in a 7-day period.
NBtarheel_33 is offline   Reply With Quote
Old 2015-12-12, 20:05   #77
NBtarheel_33
 
NBtarheel_33's Avatar
 
"Nathan"
Jul 2008
Maryland, USA

5×223 Posts
Default

We have been setting new highs for 30-day sustained throughput right and left as of late. The number to beat is now 319+ TFLOP/s, oh so close to a third of a petaflop, which I reckon we will achieve before we bid 2015 farewell.
NBtarheel_33 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
another defined sequence MattcAnderson Miscellaneous Math 4 2017-07-05 08:05
Budget PC Throughput Rodrigo Hardware 14 2011-09-26 10:16
SZPTR - whare is it defined tichy Software 4 2010-12-22 09:02
how is the throughput calculated? ixfd64 PrimeNet 5 2008-05-21 13:39
My throughput does not compute... petrw1 Hardware 9 2007-08-13 14:38

All times are UTC. The time now is 18:40.

Mon Sep 21 18:40:00 UTC 2020 up 11 days, 15:50, 1 user, load averages: 1.57, 1.57, 1.58

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.