mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing > GpuOwl

Reply
 
Thread Tools
Old 2021-06-17, 15:44   #1
themoon123
 
Apr 2021

23 Posts
Default How to do P-1 on a PRP assignment that has no P-1 work done?

I just got a RX 6800 XT today and will be running PRP on it mostly. I got a PRP assignment from the manual page but it hasn't gone through a P-1 test yet.
How do i do a P-1 test before the PRP test like prime95 does?

I'm running gpuowl 7.2 on Windows 10.
themoon123 is offline   Reply With Quote
Old 2021-06-17, 16:02   #2
firejuggler
 
firejuggler's Avatar
 
"Vincent"
Apr 2010
Over the rainbow

2·32·149 Posts
Default

Pminus1=N/A,1,2,23562083,-1,1200000,36000000,70
aka
AID,k,base,exponent,minus/plus, B1,B2, how far factored.


but if you specify how far factored, P95 will calculate the best bound with the B1 you provided

Last fiddled with by firejuggler on 2021-06-17 at 16:46
firejuggler is online now   Reply With Quote
Old 2021-06-17, 16:20   #3
Viliam Furik
 
Viliam Furik's Avatar
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

2·353 Posts
Default

Quote:
Originally Posted by themoon123 View Post
I just got a RX 6800 XT today and will be running PRP on it mostly. I got a PRP assignment from the manual page but it hasn't gone through a P-1 test yet.
How do i do a P-1 test before the PRP test like prime95 does?

I'm running gpuowl 7.2 on Windows 10.
GpuOwl, version 7.2, is doing P-1 automatically at the same time as PRP because the inner maths of P-1 and PRP allow it. For how to set the worktodo line such that it does this, consult Kriesel's reference material. Specifically, look for worktodo entry formats.

Could you please compare your performance in v7.2 and v6.11? I would like to see those results. The first few iteration outputs should be sufficient from each version. A screenshot is great, but simply giving me numbers for us/it (microseconds per iteration) for each version is enough. Thanks in advance
Viliam Furik is offline   Reply With Quote
Old 2021-06-17, 16:21   #4
themoon123
 
Apr 2021

23 Posts
Default

Sorry, how do i calculate which bounds b1 and b2 should be? Is there some automated way to do p-1 on an exponent before PRP?
themoon123 is offline   Reply With Quote
Old 2021-06-17, 17:35   #5
Viliam Furik
 
Viliam Furik's Avatar
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

2·353 Posts
Default

Quote:
Originally Posted by themoon123 View Post
Sorry, how do i calculate which bounds b1 and b2 should be? Is there some automated way to do p-1 on an exponent before PRP?
If you put the following form of the worktodo line into your gpuOwl worktodo, it will do P-1 automatically.

PRP=AID,1,2,exponent,-1,tflevel,1

The one at the end is important, it tells the gpuOwl that P-1 hasn't been done yet. AID is optional, but if you are getting work from PrimeNet (i.e. mersenne.org page), you should have one.
Viliam Furik is offline   Reply With Quote
Old 2021-06-17, 17:59   #6
themoon123
 
Apr 2021

23 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
If you put the following form of the worktodo line into your gpuOwl worktodo, it will do P-1 automatically.

PRP=AID,1,2,exponent,-1,tflevel,1

The one at the end is important, it tells the gpuOwl that P-1 hasn't been done yet. AID is optional, but if you are getting work from PrimeNet (i.e. mersenne.org page), you should have one.
Ah.. Thank you! The assignment i got has a 2 at the end, what does that mean?

Last fiddled with by themoon123 on 2021-06-17 at 18:24
themoon123 is offline   Reply With Quote
Old 2021-06-17, 18:37   #7
themoon123
 
Apr 2021

10002 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
GpuOwl, version 7.2, is doing P-1 automatically at the same time as PRP because the inner maths of P-1 and PRP allow it. For how to set the worktodo line such that it does this, consult Kriesel's reference material. Specifically, look for worktodo entry formats.

Could you please compare your performance in v7.2 and v6.11? I would like to see those results. The first few iteration outputs should be sufficient from each version. A screenshot is great, but simply giving me numbers for us/it (microseconds per iteration) for each version is enough. Thanks in advance
I tested both 6.11 and 7.2 and got the following:

6.11: 1002 us/it
7.2: 1017 us/it
themoon123 is offline   Reply With Quote
Old 2021-06-17, 18:46   #8
Viliam Furik
 
Viliam Furik's Avatar
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

2·353 Posts
Default

Quote:
Originally Posted by themoon123 View Post
Ah.. Thank you! The assignment i got has a 2 at the end, what does that mean?
2 means, as far as I know, that if a factor is found by P-1, 2 tests will be saved. That's a remnant after LL testing, where this is true. For PRP, with proof generation and certification, this is no longer true, because only 1 test is being run; the certification is only a tiny fraction of the entire test.

For the gpuOwl 1 is thus an OK choice.

Quote:
Originally Posted by themoon123 View Post
I tested both 6.11 and 7.2 and got the following:

6.11: 1002 us/it
7.2: 1017 us/it
Thanks! Really appreciated.
Viliam Furik is offline   Reply With Quote
Old 2021-06-17, 18:46   #9
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

10111001000112 Posts
Default

If P-1 is needed, gpuowl 7.x do PRP from the start, and P-1 stage 1 powers are assembled from PRP interim computations, which reduces stage 1 cost to about 10% of a standalone run. However, this greatly increases stage 1 memory requirements. (Gpuowl 7.x will run the stage 1 GCD in parallel with PRP, and if no factor is found in stage 1, pause the PRP to perform P-1 stage 2 computations, then if no factor is found in stage 2, continue the PRP to completion.)

Also combined P-1 and PRP runtime of V6.11-380 and close versions are for most FFT lengths faster than v7.x.
A detailed benchmarking versus FFT lengths and selected Gpuowl versions is available for Radeon VII as last attachment of https://www.mersenneforum.org/showpo...35&postcount=2.

Gpuowl provides default bounds, but they are larger than optimal for pure Mersenne prime hunting. Bounds can be specified on a per-worktodo-entry basis. See https://www.mersenneforum.org/showpo...9&postcount=20 for choosing bounds.


Code:
6.11: 1002 us/it
7.2: 1017 us/it
For what exponent or fft length? Factory standard clocking on the GPU?

Last fiddled with by kriesel on 2021-06-17 at 18:50
kriesel is online now   Reply With Quote
Old 2021-06-18, 05:55   #10
themoon123
 
Apr 2021

23 Posts
Default

Quote:
Originally Posted by kriesel View Post
If P-1 is needed, gpuowl 7.x do PRP from the start, and P-1 stage 1 powers are assembled from PRP interim computations, which reduces stage 1 cost to about 10% of a standalone run. However, this greatly increases stage 1 memory requirements. (Gpuowl 7.x will run the stage 1 GCD in parallel with PRP, and if no factor is found in stage 1, pause the PRP to perform P-1 stage 2 computations, then if no factor is found in stage 2, continue the PRP to completion.)

Also combined P-1 and PRP runtime of V6.11-380 and close versions are for most FFT lengths faster than v7.x.
A detailed benchmarking versus FFT lengths and selected Gpuowl versions is available for Radeon VII as last attachment of https://www.mersenneforum.org/showpo...35&postcount=2.

Gpuowl provides default bounds, but they are larger than optimal for pure Mersenne prime hunting. Bounds can be specified on a per-worktodo-entry basis. See https://www.mersenneforum.org/showpo...9&postcount=20 for choosing bounds.


Code:
6.11: 1002 us/it
7.2: 1017 us/it
For what exponent or fft length? Factory standard clocking on the GPU?
I got these numbers on an ASUS TUF OC 6800 XT with factory overclocking. The exponent i'm testing is 111482183.

So ideally i would get the bounds i should use from here right? https://www.mersenne.ca/exponent/111482183 under GPU72 bounds?

Last fiddled with by themoon123 on 2021-06-18 at 06:03
themoon123 is offline   Reply With Quote
Old 2021-06-18, 08:45   #11
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

231208 Posts
Default

If you are not happy with the default bounds (which for the most cases are good enough, but you may want to adjust them a bit higher if you are in the game for finding factors, or a bit lower, as anteposters said, if your goal is to find a large prime), or if you just wanna dirty your hands with P-1 bulk factoring, then you should better take your bounds from here, according with how many factors you want to find per 100 exponents, the power of your hardware (in GHzDays per day) and your tolerance for pain (i.e. patience).

Note that if you P-1 in bulk in a range, to find factors, then you don't need to compute for every exponent. Put some exponent in the middle of the range you are targeting, select your desired probabilities, and then go with those B1/B2 for the whole range.

Play a bit with the form first to see what's what (for example, yo don't need to fill all fields, if you let fields empty, the best values will be suggested for you), it is very easy to understand, and useful, (i.e. it saves your time and resources, and it helps the project better than if you pick the limits from the thin air, or from the "neighbors" ). If it is unclear, ask.

Last fiddled with by LaurV on 2021-06-18 at 08:53
LaurV is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Best type of work for lowest exponent work? cappy95833 Software 5 2019-11-21 04:24
Work Assignment Question Fred PrimeNet 4 2016-04-23 21:47
How can one prevent assignment of new work? Dorian PrimeNet 13 2012-04-11 21:52
How to calculate work/effort for PRP work? James Heinrich PrimeNet 0 2011-06-28 19:29
Old Assignment Primeinator PrimeNet 4 2009-06-04 22:35

All times are UTC. The time now is 00:58.


Fri Dec 3 00:58:45 UTC 2021 up 132 days, 19:27, 1 user, load averages: 1.28, 1.40, 1.42

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.