mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > Lone Mersenne Hunters > LMH > 100M

Reply
 
Thread Tools
Old 2015-12-27, 01:33   #595
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

127668 Posts
Default

I didn't ask what method you used to determine 400M; I asked how long P95 takes to do that stage 2, instead of GMP-ECM.

The fact you found GMP-ECM most efficient at 1/10th of its default B2, where I usually find GMP-ECM most efficient at default to 2x default B2, suggests to me that GMP-ECM is not a great choice for the numbers you're working on. That's why I asked how P95 does instead of GMP.
VBCurtis is offline   Reply With Quote
Old 2015-12-27, 01:37   #596
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2×3×937 Posts
Default

Quote:
Originally Posted by chalsall View Post
Sigh... Why do we not understand how hyperthreads work (repeatedly)?

It's a bit like being asked to teach a graduate course, but then have to entertain kindergarten children....
Chris-
Can you explain what you mean by this? If I understand your implication, it may be you who needs to freshen your understanding of HT, specifically that different programs respond quite differently to using HT "cores".
VBCurtis is offline   Reply With Quote
Old 2015-12-27, 01:43   #597
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

1088510 Posts
Default

Can a mod please move all of the off topic replies from 581 forward to a new thread please?
Uncwilly is offline   Reply With Quote
Old 2015-12-27, 02:01   #598
Gordon
 
Gordon's Avatar
 
Nov 2008

509 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
I didn't ask what method you used to determine 400M; I asked how long P95 takes to do that stage 2, instead of GMP-ECM.

The fact you found GMP-ECM most efficient at 1/10th of its default B2, where I usually find GMP-ECM most efficient at default to 2x default B2, suggests to me that GMP-ECM is not a great choice for the numbers you're working on. That's why I asked how P95 does instead of GMP.
Thought I had answered it, sorry.

P95 - 117 secs

GMP - 89 secs
Gordon is offline   Reply With Quote
Old 2015-12-27, 08:07   #599
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

10,273 Posts
Default

Quote:
Originally Posted by Gordon View Post
Task manager shows 8 threads, all at about 50% or so
Oh boy...

Edit: agree with Uncwilly that a mod should move out or delete the futile posts, and agree with RDS that some of us are worthless..

Last fiddled with by LaurV on 2015-12-27 at 08:09
LaurV is offline   Reply With Quote
Old 2015-12-27, 18:58   #600
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

22·5·7·79 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Can you explain what you mean by this? If I understand your implication, it may be you who needs to freshen your understanding of HT, specifically that different programs respond quite differently to using HT "cores".
Sure...

Those who are serious about their code (like George) work so close "to the metal" that getting 50% CPU utilization (with hyperthreading enabled) is exactly what we want.

I would argue that sub-optimal code (which does anything useful) will almost never get to 100% CPU utilization, even with hyperthreading enabled, because of memory bandwidth issues and lack of explicit affinity.

Happy to be counter argued....
chalsall is offline   Reply With Quote
Old 2015-12-27, 19:30   #601
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

2·52·163 Posts
Default

Quote:
Originally Posted by chalsall View Post

Those who are serious about their code (like George) work so close "to the metal" that getting 50% CPU utilization (with hyperthreading enabled) is exactly what we want.

Happy to be counter argued....

Code efficiency has nothing to do with Task Manager's CPU utilization number.

Task Manager simply measures the percentage of time the OS had a process available to assign to the logical CPU.

The most common source of confusion is that prime95 defaults to 4 workers on a hyperthreaded quad core. With only 4 active threads and 8 logical processors you get a 50% CPU utililization. We must then explain to newbies that since prime95 is super-efficient, the 4 physical CPUs are actually getting used to their maximum.

Gordon's original post of "Task manager shows 8 threads, all at about 50% or so" is a different case entirely. The way I read that statement, we have 8 workers and 8 logical cores and we should be seeing 100% logical CPU utilization. He must have some kind of problem with affinity settings that is preventing the OS from letting the workers occupy all 8 logical CPUs.
Prime95 is offline   Reply With Quote
Old 2015-12-27, 20:28   #602
Gordon
 
Gordon's Avatar
 
Nov 2008

509 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Code efficiency has nothing to do with Task Manager's CPU utilization number.

Task Manager simply measures the percentage of time the OS had a process available to assign to the logical CPU.

The most common source of confusion is that prime95 defaults to 4 workers on a hyperthreaded quad core. With only 4 active threads and 8 logical processors you get a 50% CPU utililization. We must then explain to newbies that since prime95 is super-efficient, the 4 physical CPUs are actually getting used to their maximum.

Gordon's original post of "Task manager shows 8 threads, all at about 50% or so" is a different case entirely. The way I read that statement, we have 8 workers and 8 logical cores and we should be seeing 100% logical CPU utilization. He must have some kind of problem with affinity settings that is preventing the OS from letting the workers occupy all 8 logical CPUs.
I was reporting what I saw when running 4 instances of gmp-ecm. Is that the os doing the threading or is gmp-ecm multi-processor aware?
Gordon is offline   Reply With Quote
Old 2015-12-27, 21:27   #603
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

22·5·7·79 Posts
Default

Quote:
Originally Posted by Gordon View Post
I was reporting what I saw when running 4 instances of gmp-ecm. Is that the os doing the threading or is gmp-ecm multi-processor aware?
We don't yet know exactly what is happening.

Further empirical testing would be useful.
chalsall is offline   Reply With Quote
Old 2015-12-28, 00:05   #604
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2·3·937 Posts
Default

GMP-ECM is not thread-aware. 4 GMP processes on 8 logical cores should read 50% across all 8 logical cores, as windows will bounce a process from core to core to try to "use" all available cores (I think). If some affinity mask is set, you might get 4 at 100% instead.

I believe GMP-ECM runs better with more processes than physical cores; on an old i7 laptop, 8 processes ran about 20% more curves per day than 4 processes. I never ran stage 1 and stage 2 separately, and I haven't found P95-ECM (using FFT's) to have a similar speedup; but you might still benefit from 4x P95 and 2x GMP stage 2 at the same time.

Chris- I definitely misunderstood your posting about HT, sorry.
VBCurtis is offline   Reply With Quote
Old 2015-12-28, 00:21   #605
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

101011001101002 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Chris- I definitely misunderstood your posting about HT, sorry.
Absolutely no problem. I often make mistakes, and want to be corrected.

Peace, brother!
chalsall is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
GPU72 / MISFIT use for 100M digit range? Uncwilly GPU to 72 64 2013-03-31 02:45
I want a 100M digit Mersenne that.... JuanTutors PrimeNet 8 2012-12-06 13:47
How far along are you in your 100M digit LL test? JuanTutors Lounge 6 2012-02-21 07:36
100M-digit n/k pairs __HRB__ Riesel Prime Search 0 2010-05-22 01:17
100M digit prime Unregistered Information & Answers 10 2010-03-24 20:16

All times are UTC. The time now is 07:43.


Fri Jan 27 07:43:18 UTC 2023 up 162 days, 5:11, 0 users, load averages: 0.65, 0.74, 0.74

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2023, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔