mersenneforum.org Future requests?
 Register FAQ Search Today's Posts Mark Forums Read

 2012-04-16, 03:37 #1 LaurV Romulan Interpreter     Jun 2011 Thailand 230728 Posts Future requests? I see we don't have a thread for requesting new features Well... my request is not exactly about GPU-2-72, but about the visualization page. When we go to the max zoom out on the table, there is a lot of information lost on the right, because the smaller expos (under 14M) are factored not too high, and they have to be on the table. This is valid not only for the status (tabular) view, but for changes table too. And it may affect lower levels of zoom too. So, my suggestion would be to implement either a possibility to "move" left-right with the columns (pan, or select from-to bits), or either implement a larger table (that would go out of screen, but then use firefox's "pan" functions), or either have a checkbox on that page, to "ignore first 10M exponents", or better 15M, 20M, or either to have an "intermediary" zoom level which ignores first row. I keep in mind that the things on the upper ranges of exponents are still evolving (lowest bound is raising) and also less and less people are interested on the lower expos which are already "cleared", "factored", etc. Last fiddled with by chalsall on 2012-07-27 at 14:13
2012-04-16, 15:36   #2
chalsall
If I May

"Chris Halsall"
Sep 2002

23·433 Posts

Quote:
 Originally Posted by LaurV So, my suggestion would be to implement either a possibility to "move" left-right with the columns (pan, or select from-to bits), or either implement a larger table (that would go out of screen, but then use firefox's "pan" functions), or either have a checkbox on that page, to "ignore first 10M exponents", or better 15M, 20M, or either to have an "intermediary" zoom level which ignores first row. I keep in mind that the things on the upper ranges of exponents are still evolving (lowest bound is raising) and also less and less people are interested on the lower expos which are already "cleared", "factored", etc.
A reasonable and logical request.

In all honesty, I haven't spent much time with the Mersenne.info site/code for several months. But I should do so.

Let me get back to you....

 2012-04-16, 16:09 #3 Dubslow Basketry That Evening!     "Bunslow the Bold" Jun 2011 40
2012-04-16, 16:20   #4
chalsall
If I May

"Chris Halsall"
Sep 2002

23×433 Posts

Quote:
 Originally Posted by Dubslow I've been thinking it would nice to be able to select ranges other than those with boundaries/multiples at powers of 10; that is to say, be able to view 36-46M instead of just 30-40M and 40M-50M; that is to say (again) something like what you did with the "Workers's Progress" graphs on GPU272.
Yes... That request was also noted, and added to the todo list. It is also reasonable and logical.

 2012-04-16, 19:03 #5 mognuts     Sep 2008 Bromley, England 32·5 Posts Is it possible to generate a downloadable worktodo.txt file instead of having to cut and paste the individual lines generated on the 'get assignments' page?
2012-04-16, 22:53   #6
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

10010101000012 Posts

Quote:
 Originally Posted by mognuts Is it possible to generate a downloadable worktodo.txt file instead of having to cut and paste the individual lines generated on the 'get assignments' page?
If this was mersenne.facebook I'd be looking for the "Like" button.

Last fiddled with by petrw1 on 2012-04-16 at 22:57

2012-04-16, 23:03   #7
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

476910 Posts

Quote:
 Originally Posted by LaurV When we go to the max zoom out on the table, there is a lot of information lost on the right, because the smaller expos (under 14M) are factored not too high, and ...
gee for a split second I thought you were going to end your above sentence with:

Quote:
 "so I'm going to GPU them all to 65 bits just so you don't have to bother changing the report."
oh well wishful thinking.

2012-04-16, 23:35   #8
chalsall
If I May

"Chris Halsall"
Sep 2002

23×433 Posts

Quote:
 Originally Posted by petrw1 gee for a split second I thought you were going to end your above sentence with: oh well wishful thinking.
What is even funnier (or, perhaps, sadder) is I and a few others compiled our own versions of mprime to remove the sanity check, and took the last few hundred low candidates up to 60 bits using CPUs just for the hell of it a couple of years ago....

2012-04-16, 23:47   #9
chalsall
If I May

"Chris Halsall"
Sep 2002

26E716 Posts

Quote:
Originally Posted by petrw1
Quote:
 Originally Posted by mognuts Is it possible to generate a downloadable worktodo.txt file instead of having to cut and paste the individual lines generated on the 'get assignments' page?
If this was mersenne.facebook I'd be looking for the "Like" button.
OK... I'm hearing that this is a desired feature...

Would it be OK if this was a function from the View Assignments page? It would be much easier to implement there.

2012-04-17, 06:11   #10
LaurV
Romulan Interpreter

Jun 2011
Thailand

2×3×7×233 Posts

Quote:
 Originally Posted by petrw1 gee for a split second I thought you were going to end your above sentence with: Quote: "so I'm going to GPU them all to 65 bits just so you don't have to bother changing the report." oh well wishful thinking.
Quote:
 Originally Posted by chalsall What is even funnier (or, perhaps, sadder) is I and a few others compiled our own versions of mprime to remove the sanity check, and took the last few hundred low candidates up to 60 bits using CPUs just for the hell of it a couple of years ago....
Yeah, I would do it with pleasure! I am the crazy farang who took M1061 from 62 to 63 bits(**see Notes below) and the same guy who P-1 this exponent to those astronomical limits, see here (it took few months for P-1 with p95). I did it (the TF) with my own (slow multiplication) tool, who checks all 8kp+1 and 8kp+6p+1 numbers (prime or not), for natural k and p=1061, and the 63-to-64 is on the way too, but interrupted at about 15% some time ago. The same slow tool I use it for the stuff in this post (trying P-1 on GPU). I would do "lower expos to 64 and higher" with pleasure if I would have a FAST software tool. I HAVE THE HARDWARE! That is why I am "hitting" other forum members in posts like this one, and keep my eyes on the range.

Notes:
** that job I reported as "mfaktc-faked-by-LaurV", the "faked" part does not refers to the job (the job was done last year!) but it refers to the report itself, as Primenet doesn't know my tool, and I had to report it as mfaktc. And with this opportunity I found out that the formula to compute the credit has a "gap". (read as "has an error, George, are you there?"). Generally the credit doubles with every bit-step for a given exponent, with the exception of exponents lower then 2000, from 62 to 63 bits, where it has a "times 4" jump. So, the credit it gives for a 62-63 bit on M1061 is about ~900GHzDays, but it should be only ~450GD. The credit for 61 to 62 is correctly given as ~230GD, and so far correct for all lower or higher. The "gap" is only 62 to 63 bits. The same error appears in the mersenne-aries page, I believe James uses the same formulas. This should be a two-days job on a gtx580 with a "good mfaktc" program, but it took me over two weeks.

Last fiddled with by LaurV on 2012-04-17 at 06:34 Reason: links added

2012-04-17, 11:35   #11
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

24×3×73 Posts

Quote:
 Originally Posted by LaurV And with this opportunity I found out that the formula to compute the credit has a "gap". (read as "has an error, George, are you there?"). Generally the credit doubles with every bit-step for a given exponent, with the exception of exponents lower then 2000, from 62 to 63 bits, where it has a "times 4" jump. So, the credit it gives for a 62-63 bit on M1061 is about ~900GHzDays, but it should be only ~450GD. The credit for 61 to 62 is correctly given as ~230GD, and so far correct for all lower or higher. The "gap" is only 62 to 63 bits. The same error appears in the mersenne-aries page, I believe James uses the same formulas.
You can call it a "gap" if you like, but it's not an error. It's designed to reflect actual CPU TF performance (example showing theoretical 1GHz Core2 Duo, basis of GHz-days metric) at around the 62-64 transition. If you look at CPU TF benchmarks you'll notice a large jump in TF times around 62-63 bits (11ms -> 17ms in the Core2 example), but you'll also notice the GHz-days/day TF graph stays flatline at 1.0GHz-days/day despite the slowdown. GHz-days are architechture-specific performance and may not make sense in other contexts, such as using mfaktc which doesn't perform the same (but is analogous to the mfakto slowdown above 2^70).

The actual formula for George's code, which is also used on my site, includes this comment:
Code:
// bits    tf_timing
// ----    ---------
// <=62    0.00465
// 63-64   0.00743
// >=65    0.00707

 Similar Threads Thread Thread Starter Forum Replies Last Post Xyzzy Forum Feedback 104 2017-04-02 22:20 robert44444uk Prime Gap Searches 2 2017-01-17 07:57 WraithX FactorDB 32 2014-12-18 03:40 tului GPU Computing 15 2014-06-24 06:22 ixfd64 PrimeNet 44 2010-01-11 20:21

All times are UTC. The time now is 15:38.

Fri Oct 22 15:38:22 UTC 2021 up 91 days, 10:07, 1 user, load averages: 1.17, 1.41, 1.33