mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-12-03, 23:51   #1
SethTro
 
SethTro's Avatar
 
"Seth"
Apr 2019

2×101 Posts
Default TF data

I'm working on adding a verifiable component to TF
See:
https://mersenneforum.org/showpost.p...&postcount=199
TF-PROOF.md

I made the needed changes to mfaktc because that's what I'm using.

Is there data about what program (gpuowl, mfaktc, mfakto, prime95, ???) is submitting the most TF results?
Which program(s) are submitting JSON results?
What is the JSON / manual results mix?

Thanks
SethTro is online now   Reply With Quote
Old 2019-12-08, 06:51   #2
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

63158 Posts
Default

Quote:
Originally Posted by SethTro View Post
I'm working on adding a verifiable component to TF
See:
https://mersenneforum.org/showpost.p...&postcount=199
TF-PROOF.md

I made the needed changes to mfaktc because that's what I'm using.

Is there data about what program (gpuowl, mfaktc, mfakto, prime95, ???) is submitting the most TF results?
Which program(s) are submitting JSON results?
What is the JSON / manual results mix?

Thanks
I don't have any hard #'s for you, but TF work is almost entirely done by GPU's so that leaves out Prime95. gpuOwl is probably just doing PRP tests, so I'd guess it's mostly mfakt* programs doing the TF work.

As far as I know, only gpuOwl and Prime95 are doing JSON results so far. Ernst has done work on getting mlucas updated to use the JSON format.
Madpoo is offline   Reply With Quote
Old 2019-12-09, 02:36   #3
SethTro
 
SethTro's Avatar
 
"Seth"
Apr 2019

2·101 Posts
Default

That's great news for me. Thanks for the input.
SethTro is online now   Reply With Quote
Old 2019-12-09, 07:56   #4
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

11·431 Posts
Default

Quote:
Originally Posted by Madpoo View Post
gpuOwl is probably just doing PRP tests, so I'd guess it's mostly mfakt* programs doing the TF work.
gpuowl mostly PRP, some P-1, and rarely some old versions doing LL DC. I think gpuowl TF is the rarest.
Mfactor may be rejoining the TF fray.
kriesel is offline   Reply With Quote
Old 2020-04-29, 01:32   #5
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

11×431 Posts
Default

Quote:
Originally Posted by Madpoo View Post
I don't have any hard #'s for you, but TF work is almost entirely done by GPU's so that leaves out Prime95. gpuOwl is probably just doing PRP tests, so I'd guess it's mostly mfakt* programs doing the TF work.
Is there any way a mere user can get output from which to gauge the mfaktc / mfakto / other mix of factoring? I looked around a bit on the mersenne.org web pages & menus and did not find a way.

Last fiddled with by kriesel on 2020-04-29 at 01:35
kriesel is offline   Reply With Quote
Old 2020-04-30, 03:55   #6
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

327710 Posts
Default

Quote:
Originally Posted by kriesel View Post
Is there any way a mere user can get output from which to gauge the mfaktc / mfakto / other mix of factoring? I looked around a bit on the mersenne.org web pages & menus and did not find a way.
There's not really any good way to get that info besides looking at the raw messages turned into the server and pulling out the client information. Which is a pain.

But since GPU work is so much better at factoring, I think it's a very safe assumption that almost all TF work is done by one of those.

For example, while it's hard to break down results by specific non-Prime95 versions, I can give you a breakdown of Prime95 and non-Prime95 TF results. This is for the month of April to date:
Non-Prime95 (turned in by the manual result pages): 1,220,682
Prime95 (various builds): 74,167

That makes non-Prime95 ~ 94.3% of all TF work done this month so far. I'm actually surprised it's that low... I thought it'd be closer to 98%.
Madpoo is offline   Reply With Quote
Old 2020-04-30, 04:08   #7
axn
 
axn's Avatar
 
Jun 2003

7·683 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Non-Prime95 (turned in by the manual result pages): 1,220,682
Prime95 (various builds): 74,167

That makes non-Prime95 ~ 94.3% of all TF work done this month so far. I'm actually surprised it's that low... I thought it'd be closer to 98%.
Do you have the bit levels to go with these? I'd bet that once that is accounted for, the effort will be 99% GPU and < 1% CPU TF.
axn is offline   Reply With Quote
Old 2020-04-30, 12:23   #8
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

11×431 Posts
Default

Quote:
Originally Posted by Madpoo View Post
That makes non-Prime95 ~ 94.3% of all TF work done this month so far. I'm actually surprised it's that low... I thought it'd be closer to 98%.
Ideally it would be around 99.9% Your 94.3% figure suggests we could increase primality testing rate further. Cpus should not be doing ANY TF, if we could keep the first-primality-testing wavefront swept clean by gpus.
Do you have a sense of how much of the cpu TF relates to 100Mdigit, or doublechecking, or random manual assignments ahead of the wavefront?
It's less important to distinguish TF from mfaktc from mfakto from gpuowl, or prime95 vs mprime, than gpu from cpu, But a script that totalizes tf bit level completion counts and ghzdays counts per application name for a given data sample input seems very feasible to me.
Once ghzdays/bit-level is folded in, it's likely the gpu/cpu TF work ratio shifts by ~8-16fold, since the top TF bit level for gpus is typically 3-4 higher than for cpus.
I'd be willing to attempt such a script in perl (considerably) later. Perhaps anonymized data could be made available. In the short term I could use my own output for test vectors.

Last fiddled with by kriesel on 2020-04-30 at 12:26
kriesel is offline   Reply With Quote
Old 2020-04-30, 23:35   #9
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

17×523 Posts
Default

I think that a lot of the CPU TF is the LMH high stuff. You can churn through a lot of exponents in the 200,000,000 range. Lots of small assignments.
Uncwilly is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
NFS filtering in the era of big data jasonp Msieve 36 2018-05-07 19:55
GPU TF vs DC/LL data bcp19 GPU to 72 0 2011-12-02 16:41
New data page kar_bon Riesel Prime Search 144 2008-10-21 10:27
Data available? Prime95 LMH > 100M 10 2007-06-22 23:55
Have I submitted bad data? (Or, well, about to anyway) Nazo Software 5 2005-08-06 05:44

All times are UTC. The time now is 03:09.

Tue Dec 1 03:09:24 UTC 2020 up 82 days, 20 mins, 1 user, load averages: 2.39, 1.86, 1.76

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.