mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-08-24, 09:14   #1
ZFR
 
ZFR's Avatar
 
Feb 2008
Bray, Ireland

3·52 Posts
Default Is it worth doing LLs on GPU now?

I generally switch between mfaktc and CUDALucas on my GPU - when I get bored with one, I change to the other. However, with the new changes coming to the way proofs will be done, is it still worth it to check new exponents on CUDALucas? Or should I just switch to DCing on CUDALucas (or factoring on mfaktc) now?
ZFR is offline   Reply With Quote
Old 2020-08-24, 11:35   #2
ATH
Einyen
 
ATH's Avatar
 
Dec 2003
Denmark

BA616 Posts
Default

Yes, it is better to do LL DC with CUDALucas instead of new LL tests.

Last fiddled with by ATH on 2020-08-24 at 11:35
ATH is offline   Reply With Quote
Old 2020-08-27, 01:28   #3
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

3·1,579 Posts
Default

What gpu model? Some can run gpuowl instead; P-1, LLDC with Jacobi check, or PRP with proof, and usually more GhzD/day than CUDALucas.
kriesel is online now   Reply With Quote
Old 2020-08-27, 07:18   #4
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

22×7×11×29 Posts
Default

Quote:
Originally Posted by ZFR View Post
Is it worth doing LLs on GPU now?
Yes, and it was always worth. The only argument is when to switch from TF to LL, and this depends on your GPU, you have to think how fast you can clear exponents by either completing 1 or 2 LL tests, versus finding a factor (i.e. TF-ing about 75 exponents or so).

Last fiddled with by LaurV on 2020-08-27 at 07:18
LaurV is offline   Reply With Quote
Old 2020-08-28, 08:11   #5
ZFR
 
ZFR's Avatar
 
Feb 2008
Bray, Ireland

3·52 Posts
Default

Quote:
Originally Posted by LaurV View Post
Yes, and it was always worth. The only argument is when to switch from TF to LL, and this depends on your GPU, you have to think how fast you can clear exponents by either completing 1 or 2 LL tests, versus finding a factor (i.e. TF-ing about 75 exponents or so).
Oh yes, I understand this. My question was: v30 of prime95 is going to include the new way of making proofs without DCing. Is it worth then doing new LLs on GPU that will have to be double-checked and take almost twice as long?
ZFR is offline   Reply With Quote
Old 2020-08-28, 10:07   #6
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

22·7·11·29 Posts
Default

Still don't know how that changes everything. If I have a GPU that can eliminate this amount of exponents per month/year whatever, doing TF, and that amount doing LL+DC, and it can't run gpuOwl, then nothing changes.

So, like kriesel said, which model? Unless you have some very old GPU, it may be able to run gpuOwl. And if so, also nothing changes from the TF vs LL/PRP point of view (the owl runs PRPs), except now you count against a single LL/PRP test plus small offset for CERT, instead of counting against two LLs. Also nothing changes for LLDC front, those still need to be DC, and DC is faster than PRP+CERT.

For the front LL, those will be obsolete anyhow, by the new PRP+CERT. So, if your card can run the Owl, forget about first time LL.
LaurV is offline   Reply With Quote
Old 2020-08-28, 10:20   #7
ATH
Einyen
 
ATH's Avatar
 
Dec 2003
Denmark

BA616 Posts
Default

Quote:
Originally Posted by LaurV View Post
For the front LL, those will be obsolete anyhow, by the new PRP+CERT. So, if your card can run the Owl, forget about first time LL.
This is the original question: When he tires of mfaktc and wants to run CUDALucas, should he run first time LL tests with CUDALucas or stick to LL DC on CUDALucas.

While he can run first time LL test if he wants, it is much better for the project to run LL DC on it, or PRP tests if the card can run gpuowl.

Last fiddled with by ATH on 2020-08-28 at 10:27
ATH is offline   Reply With Quote
Old 2020-08-28, 10:44   #8
ZFR
 
ZFR's Avatar
 
Feb 2008
Bray, Ireland

4B16 Posts
Default

Quote:
Originally Posted by ATH View Post
This is the original question: When he tires of mfaktc and wants to run CUDALucas, should he run first time LL tests with CUDALucas or stick to LL DC on CUDALucas.

While he can run first time LL test if he wants, it is much better for the project to run LL DC on it, or PRP tests if the card can run gpuowl.
Yes, pretty much. While I know that *any* tests I make move GIMPs forward, it was more of a question of the best way of spending those cycles.

The GPU is a Geforce GTX 970. I thought gpuowl was for AMD GPUs only?
ZFR is offline   Reply With Quote
Old 2020-08-28, 12:27   #9
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

22·7·11·29 Posts
Default

Nvidia cards can run both Cuda and OpenCL except that for the last, they are quite slow, except for the newest cards.
LaurV is offline   Reply With Quote
Old 2020-08-28, 14:04   #10
ZFR
 
ZFR's Avatar
 
Feb 2008
Bray, Ireland

1138 Posts
Default

Quote:
Originally Posted by LaurV View Post
Nvidia cards can run both Cuda and OpenCL except that for the last, they are quite slow, except for the newest cards.
I see. Once CERT is implemented into gpuowl I'll take a look and see how fast it is. In the meantime I'll stick with DCing and factoring.

Thanks, all.

EDIT: OK, I see latest gpuowl already has it.

Last fiddled with by ZFR on 2020-08-28 at 14:07
ZFR is offline   Reply With Quote
Old 2020-09-01, 20:13   #11
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

6338 Posts
Default

I'd be a lot more comfortable running gpuowl on my 2070 if I could get a better read on how far we are ahead of the wavefront on TF'ing. I hate to think about people on the gpu72 project massively cutting over to PRP with the Owl and then the whole project slowing down because CPUsers are doing month long PRPs with factors in the low to middling bit range.
Aramis Wyler is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Is it worth persisting with this machine? stebbo Hardware 6 2017-05-27 04:05
Is it worth keeping the relations? fivemack Factoring 2 2013-05-19 18:40
Successful TF worth more than unsuccessful TF?! NBtarheel_33 PrimeNet 5 2010-06-17 00:17
Does it worth it? victor Lounge 30 2009-05-30 21:53
Worth running on Atom? db597 Hardware 8 2009-01-28 09:45

All times are UTC. The time now is 21:21.

Mon Nov 30 21:21:58 UTC 2020 up 81 days, 18:32, 2 users, load averages: 2.53, 2.40, 2.39

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.