What percentage of users and/or total GHz days are being used for TF, P-1, DC, PRP, etc.?

piforbreakfast · 2022-06-09, 04:06

I've been wondering for a while why some users seem to put thousands of GHz days into TF or P-1 instead of PRP. What percentage of registered users and/or GHz days are currently being used primarily for TF, P-1, PRP, etc.?

Uncwilly · 2022-06-09, 04:59

If you have a bunch of GPU resources, TF makes sense. If you have access to machines with a bunch of memory, P-1 might makes sense. Also, if you want to contribute and don't wish to deal with the headache that might follow if you find a prime on a machine that you might not have full rights to run Prime95 on, TF, P-1, ECM, and even DC's cound make sense.

Another thing. If you have a bunch of accounts that you use with G oogle Colab, running TF on the GPU's and P-1 on the CPU's would prevent issues that might arise if a prime were found.

Prime95 · 2022-06-09, 05:15

Comparing GHz-days across different work types is problematic. The CPU credit formula was developed in 2008. Back then GPUs couldn't do TF. Nowadays, GPUs are blazingly fast at TF and give tons of CPU credit. Recently, the P-1 algorithm improved radically for smaller exponents. I'm doing P-1 on small exponents on a CPU and getting tens of thousands of GHZ-days credit every day, whereas if I did PRP on the same CPU I'd only get a few dozen GHz-days credit each day.

Each user is encouraged to do the work type that interests them the most.

kriesel · 2022-06-09, 07:41

There are lots of users so probably lots of reasons. Among them would be suitability of available hardware, and some sorts of personal preferences, subproject goals, and more.

Offhand I don't know where to find percentages of cycles or assignments or GHD/day or whatever measures among the many different assignment types.
There is a lot of fairly rapid variation in specific activities. See https://www.mersenne.org/primenet/graphs.php
You could grab a snapshot of https://www.mersenne.org/report_recent_results/ and postprocess it somehow to produce percentages and post your conclusions for some short sample period.

petrw1 · 2022-06-09, 15:00

If you're willing to do some tabulating combine as you see fit this report for every work type:

https://www.mersenne.org/report_top_...=500&type=1001

mathwiz · 2022-06-09, 21:55

Quote:

Originally Posted by Prime95

Comparing GHz-days across different work types is problematic. The CPU credit formula was developed in 2008.

Has any effort been made to update the credit formula? What would (or could) it look like today? And would it be best to prioritize work that either tests or factors exponents at the wavefront?

piforbreakfast · 2022-06-11, 20:50

Quote:

Originally Posted by Uncwilly

If you have a bunch of GPU resources, TF makes sense. If you have access to machines with a bunch of memory, P-1 might makes sense. Also, if you want to contribute and don't wish to deal with the headache that might follow if you find a prime on a machine that you might not have full rights to run Prime95 on, TF, P-1, ECM, and even DC's cound make sense.

Another thing. If you have a bunch of accounts that you use with G oogle Colab, running TF on the GPU's and P-1 on the CPU's would prevent issues that might arise if a prime were found.

So are there a lot of users running Prime95 on other people's computers without the people knowing it?

slandrum · 2022-06-11, 22:24

Depends on what you specifically mean by "other people's computers".

Some of us run on rented/loaned machines (i.e. cloud computing).

I absolutely would not run it one my work machines without permission, and I don't have permission to use those for non-work purposes.

VBCurtis · 2022-06-12, 00:49

Work-owned machines are the primary example, yes. Lots of folks run P95 without explicit permission from employer.

Uncwilly · 2022-06-12, 04:44

Quote:

Originally Posted by slandrum

Some of us run on rented/loaned machines (i.e. cloud computing).

This is one of the 2 examples I was referring to. The loaned/free cloud computing could pose a compliction if there is a prime found.

Quote:

Originally Posted by VBCurtis

Work-owned machines are the primary example, yes. Lots of folks run P95 without explicit permission from employer.

Yep, if the machine always sets on your desk at work and you are the exclusive user, running Prime95 can seem like a no harm, no foul kind of thing. Finding a new prime in that situation could really muck up the water.

piforbreakfast · 2022-06-13, 02:51

Quote:

Originally Posted by Uncwilly

This is one of the 2 examples I was referring to. The loaned/free cloud computing could pose a compliction if there is a prime found.
Yep, if the machine always sets on your desk at work and you are the exclusive user, running Prime95 can seem like a no harm, no foul kind of thing. Finding a new prime in that situation could really muck up the water.

I'm self-employed, so it's not easy to tick off the boss.

2022-06-09, 04:06	#1
piforbreakfast Oct 2020 Terre Haute, IN 122₁₀ Posts	What percentage of users and/or total GHz days are being used for TF, P-1, DC, PRP, etc.? I've been wondering for a while why some users seem to put thousands of GHz days into TF or P-1 instead of PRP. What percentage of registered users and/or GHz days are currently being used primarily for TF, P-1, PRP, etc.?

2022-06-09, 07:41	#4
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 1CBE₁₆ Posts	There are lots of users so probably lots of reasons. Among them would be suitability of available hardware, and some sorts of personal preferences, subproject goals, and more. Offhand I don't know where to find percentages of cycles or assignments or GHD/day or whatever measures among the many different assignment types. There is a lot of fairly rapid variation in specific activities. See https://www.mersenne.org/primenet/graphs.php You could grab a snapshot of https://www.mersenne.org/report_recent_results/ and postprocess it somehow to produce percentages and post your conclusions for some short sample period. Last fiddled with by kriesel on 2022-06-09 at 07:45

2022-06-11, 22:24	#8
slandrum Jan 2021 California 11·47 Posts	Depends on what you specifically mean by "other people's computers". Some of us run on rented/loaned machines (i.e. cloud computing). I absolutely would not run it one my work machines without permission, and I don't have permission to use those for non-work purposes. Last fiddled with by slandrum on 2022-06-11 at 22:25

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Total Ghz days decaying ?	tServo	PrimeNet	3	2018-08-19 12:03
Why does my GHz-days total go DOWN? ugh	billvau	Information & Answers	10	2017-03-01 01:12
Mechanics of DC instead of LL percentage	Fred	PrimeNet	1	2016-02-29 02:19
Bug in P-1 completion percentage report	NBtarheel_33	Software	3	2010-05-21 16:47
Modifying Percentage-Done Display?	Primeinator	Software	4	2005-03-15 05:32

2022-06-09, 04:59	#2
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 2³×1,361 Posts	If you have a bunch of GPU resources, TF makes sense. If you have access to machines with a bunch of memory, P-1 might makes sense. Also, if you want to contribute and don't wish to deal with the headache that might follow if you find a prime on a machine that you might not have full rights to run Prime95 on, TF, P-1, ECM, and even DC's cound make sense. Another thing. If you have a bunch of accounts that you use with G oogle Colab, running TF on the GPU's and P-1 on the CPU's would prevent issues that might arise if a prime were found.

2022-06-09, 05:15	#3
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 3·11·13·19 Posts	Comparing GHz-days across different work types is problematic. The CPU credit formula was developed in 2008. Back then GPUs couldn't do TF. Nowadays, GPUs are blazingly fast at TF and give tons of CPU credit. Recently, the P-1 algorithm improved radically for smaller exponents. I'm doing P-1 on small exponents on a CPU and getting tens of thousands of GHZ-days credit every day, whereas if I did PRP on the same CPU I'd only get a few dozen GHz-days credit each day. Each user is encouraged to do the work type that interests them the most.

2022-06-09, 15:00	#5
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 5275₁₀ Posts	If you're willing to do some tabulating combine as you see fit this report for every work type: https://www.mersenne.org/report_top_...=500&type=1001

2022-06-12, 00:49	#9
VBCurtis "Curtis" Feb 2005 Riverside, CA 2³×19×37 Posts	Work-owned machines are the primary example, yes. Lots of folks run P95 without explicit permission from employer.