mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet > GPU to 72

Closed Thread
 
Thread Tools
Old 2020-04-05, 18:51   #177
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×3×1,499 Posts
Default

Quote:
Originally Posted by KEP View Post
...but untill there is adequate room between the wavefront of TF and the various FC and P-1 wavefronts, I would sure appreciate if we wavered that part of TF for the next time being.
Please trust me on this. I understand resource management in the temporal domain. It's what network engineers do...
chalsall is offline  
Old 2020-04-06, 02:45   #178
axn
 
axn's Avatar
 
Jun 2003

11F116 Posts
Default

Quote:
Originally Posted by chalsall View Post
Further, the opposite argument to yours can be made. If we have TF'ing resources that aren't going to be redirected to FC'ing, should they optimally TF the ranges years ahead of the wavefronts? Or should they instead TF past the optional only a few months ahead?

I would argue the latter, where people are willing.
I am retroactively in agreement with you , to wit
Quote:
Originally Posted by axn View Post
Unless the BOINC project hands out LL tests as well, there is no point in calculating optimum bit level using traditional metric.
<snip>

The optimum bit level is the maximum bit level which can still prepare enough exponents for the LL testers. If that means normal + 10 bits, so be it. Obviously we shouldn't be "abusing" it -- we should be good DC citizens and not do grossly inefficient calculations, so we should go with current max bit level for the best hardware (assuming BOINC can keep up with LL demand)
However, we should also be mindful of BOINC WU runtimes. Actually, on that front, I was thinking that maybe we should slice up a large bitrange by fractional bit depths / k range rather than by residue classes. Like 77-77.5, 77.5-78, etc. I think it would be easier to implement.
axn is offline  
Old 2020-04-06, 04:58   #179
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3×2,281 Posts
Default

The primenet database supports fractional bit levels. Manual results processing and perhaps some other PHP routines, probably not. I think prime95 worktodo lines recognize fractional bit levels. Not sure about gpuowl, mlucas, others.
Prime95 is offline  
Old 2020-04-06, 07:45   #180
axn
 
axn's Avatar
 
Jun 2003

3×1,531 Posts
Default

Quote:
Originally Posted by Prime95 View Post
The primenet database supports fractional bit levels.
That is actually good news.

But, to be clear, I'm not suggesting that we start doing fractional bit TFs from PrimeNet's perspective. The idea is still to do full bit levels, but break them up into fractional WU in BOINC world. Then report the final whole bit result to PrimeNet. This would need (hopefully) minor modifications to mfakc/o, and minor-to-medium changes in the SRBase backend.

EDIT:- The halfway mark between 77 and 78 bits is 77.585 bits.

Last fiddled with by axn on 2020-04-06 at 07:45
axn is offline  
Old 2020-04-06, 07:47   #181
pinhodecarlos
 
pinhodecarlos's Avatar
 
"Carlos Pinho"
Oct 2011
Milton Keynes, UK

3·1,523 Posts
Default

Quote:
Originally Posted by axn View Post
That is actually good news.

But, to be clear, I'm not suggesting that we start doing fractional bit TFs from PrimeNet's perspective. The idea is still to do full bit levels, but break them up into fractional WU in BOINC world. Then report the final whole bit result to PrimeNet. This would need (hopefully) minor modifications to mfakc/o, and minor-to-medium changes in the SRBase backend.

EDIT:- The halfway mark between 77 and 78 bits is 77.585 bits.
Love this idea. Short, medium and long tasks. I would put my low GPU to the lower bits.
pinhodecarlos is offline  
Old 2020-04-06, 09:01   #182
lalera
 
lalera's Avatar
 
Jul 2003

3·193 Posts
Default

hi,
i do not like the idea of fractional bitlevels
fractional bitlevels do not bring better performance
every time a wu is starting the program starts
for mfaktc long running tasks (several hours - not several days) are better for the gpu

mmff for double mersenne (based on mfaktc) can do fractional bitlevels
it makes sense because a bitlevel could take weeks or longer to complete
lalera is offline  
Old 2020-04-06, 13:27   #183
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·3·1,499 Posts
Default

Quote:
Originally Posted by pinhodecarlos View Post
Love this idea. Short, medium and long tasks. I would put my low GPU to the lower bits.
Yeah, I think it's a really good idea. I hope Reb can do it.

Over on SRBase there are already different classes of work available for their S/R projects. The Long is averaging ~41 hours at the moment.

GPU72 (currently going to 78 bits) is averaging 5 hours, so breaking the higher bit levels down probably won't actually be needed. Going to 81 bits would take a similar ~40 hours.
chalsall is offline  
Old 2020-04-06, 18:50   #184
rebirther
 
rebirther's Avatar
 
Sep 2011
Germany

93B16 Posts
Default

Quote:
Originally Posted by chalsall View Post
Yeah, I think it's a really good idea. I hope Reb can do it.

Over on SRBase there are already different classes of work available for their S/R projects. The Long is averaging ~41 hours at the moment.

GPU72 (currently going to 78 bits) is averaging 5 hours, so breaking the higher bit levels down probably won't actually be needed. Going to 81 bits would take a similar ~40 hours.

Sorry, no more subapps, I have to manage 60+ with SRBase, if there is a new app you need to create all apps again and for new projects or every sub app I must create new server daemons. After this 77-78 run I will go back to 73-74 and rushing step by step though the bit levels.
rebirther is offline  
Old 2020-04-06, 19:00   #185
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×3×1,499 Posts
Default

Quote:
Originally Posted by rebirther View Post
After this 77-78 run I will go back to 73-74 and rushing step by step though the bit levels.
Well, it's a good thing GPU72 has quite a bit of firepower of its own, willing to do what's actually immediately needed rather than just what's easy.

Edit: Actually, could you please change the name of your sub-project over on SRBase? You're using my Service Mark without permission. Perhaps GIMFS would be more appropriate.

Last fiddled with by chalsall on 2020-04-06 at 19:16
chalsall is offline  
Old 2020-04-06, 23:55   #186
kracker
ἀβουλία
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

2×13×83 Posts
Default

Quote:
Originally Posted by chalsall View Post
Well, it's a good thing GPU72 has quite a bit of firepower of its own, willing to do what's actually immediately needed rather than just what's easy.

Edit: Actually, could you please change the name of your sub-project over on SRBase? You're using my Service Mark without permission. Perhaps GIMFS would be more appropriate.
Suggestions and recommendations are nice to consider and follow, but you should be happy and appreciative of firepower - wherever it is put to use by other people for whatever reason they may have.
kracker is offline  
Old 2020-04-07, 00:32   #187
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×3×1,499 Posts
Default

Quote:
Originally Posted by kracker View Post
Suggestions and recommendations are nice to consider and follow, but you should be happy and appreciative of firepower - wherever it is put to use by other people for whatever reason they may have.
Sure. But like I said above, GPU72 is about helping GIMPS. It is ***not*** GIMFS.

I feel I have wasted several days of time, being misled about what SRBase was actually going to deliver. Now I need to figure out where the needed "heavy-lifting" firepower is going to come from...
chalsall is offline  
Closed Thread

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Chess World Championship Match -- 2013, 2014, 2016 Raman Chess 34 2016-12-01 01:59
mprime ETA and primenet "days to go" do not match blip Software 1 2015-11-20 16:43
less v4 reservations being made tha PrimeNet 8 2008-08-14 08:26
LL test doesn't match benchmark drew Hardware 12 2008-07-26 03:50
WE MADE IT!!!!!!!!!!!!!!!!!!!!!! eric_v Twin Prime Search 89 2007-01-23 15:33

All times are UTC. The time now is 20:07.

Fri Jun 5 20:07:17 UTC 2020 up 72 days, 17:40, 1 user, load averages: 2.21, 1.84, 1.65

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.