| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project - Phase 2 Forum
Thread: CEP2 AMD 4P Optimization |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 58
|
|
| Author |
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7668 Status: Offline Project Badges: 
|
Even if you do not complete the entire WU in the 18 hour window, the result will be useful to the scientists as long as some of the steps are completed. Some WU's will self terminate before completing all the steps for some unknown reason, but are still valid. I agree you should let the units run and see how most of them do before making any rash decisions about the cpu's being inadequate. I have crunched some on an E5405 which is at least 7 to 8 years old.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
ok ... thanks for the encouragement .... i got a tun of clean up to do and a mountain of parts to sort ... from updating 3 servers and building 3 (2 for me 4 for you) ... see what i can find ... the biggest thing at the moment is RAM ... so you guys keep putting up ways to make things more efficient ... and let me see what i can come up with ...
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
just in case anyone is interested ...the image with streamlined kernels for multiprocessor systems is in this Thread ...
----------------------------------------http://hardforum.com/showthread.php?t=1769083 there has been talk of upgrading it to 14.04 LTS but nothing yet [Edit 1 times, last edit by Former Member at Dec 5, 2015 7:03:38 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I might get a reprimand from a Moderator for this a being "off topic"
but your machine isnt very optimized if it isnt reliable. with that being said: if you are useing a MBR (Master Boot Record) on your HDs they are not as reliable as they could be by using a GPT (guided partition table) once a new drive is initialized ... you got what you got ... so how do we fix that? there is a free utility that can be downloaded here http://www.rmprepusb.com/documents/release-2-0 on this page your are looking for ... Install_RMPrepUSB_Full_v2.1.730.zip Full version with installer 8195k v. 1 3 Nov 2015, 01:03 download and install it on a Winblows Box put a HD in this box that is to be "reinitialized" go to the start menu and select the (ALLDRIVES) version of this utility ... when it comes up ... there is a box at the top ... select the appropriate drive (BE CAREFUL! select the appropriate drive, or you will be sorry) and there is a big red button on the right bottom that says CLEAN ... (PUSH DA BUTON) clean it ... say yes to all the command line prompts when its done ... wait a few seconds ... a prompt that this "clean" drive has no drive letter assigned ... perfect!!! close the app then go into winblows Disk Manager ... select the appropriate drive ... a box will appear that will allow you to select a MBR or GPT ... BAM!!! treat it like it was a bran new drive after that  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I'm not going to be very precise on this (and other volunteers will correct me) but you'll catch the gist of the matter anyway. The setting you're referring to doesn't control the power state of your CPU but which amount of its cores to reserve for crunching. Setting it to 25% on a quad core it assigns 1 core to the science applications of the projects you've attach to, so if you're running more than one WU at a time only one will be really crunched and all the others will'be waiting for their turn, one at a time. what im taking away from what you are saying here is ... the % of cpu used is controlling the number of cores that is used in the system and has nothing to do with their P-state correct me if necssary |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
BOINC Manager doesn't influence P-states, correct.
The option controlling the number of cores of a multicore or multithread processor to use for crunching is On multiprocessors, at most use: N processors. In addition there's the option On multiprocessors, use: xxx% of processors. The option I was referring to, Use at most xxx% of CPU time, isn't addressed to multiprocessor systems only and works on single core processors too; there's a sligh difference between them but their purpose is the same. Taking a look at Device Profiles page on WCG site I see this option isn't there so I got confused with other project that actually has it. |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
That's new, no "Use at most xxx% of CPU time" on the WCG device profiles, as when I recollect it, it was near the top of the Custom > Basic options in slightly different wording... they must have removed it, else can't be bothered to look as with BOINCTasks or the BOINC Manager you get that setting too and former lets you control all your devices in one view.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hay:
---------------------------------------- i have been doing some reading and what i go out of it is ... (short answer) the RAM Drive thing wont work for a quad processor. ************************************* (painfully long answer) been reading up on the streamlined kernel for multiprocessors that i linked to above ... and there are 4 things that make it tick 1st the kernel is compiled so that it is NUMA-aware (from what i can tell this is just one little switch when the kernel is compiled) 2nd ... the app that is being used must be coded to use NUMA 3rd you have to be using a CPU that is NUMA capable so What is this "NUMA" thing anyhow??? well quad processors have banks of ram ... each bank is physically close to a processor ... NUMA Say ... use the ram that is close to the CPU, not the stuff WAaaaaa over there ... even if its all system ram ... so this speeds up the processing quite a bit .... so lets do some math CEP2 uses 700 megs disk space per core, that we want to put on a ram drive now lets say we have a 48 core Box so ... we need at least .... 33,600 Mg ram drive so 32 gigs wont get it ... so lets have 48 Gigs. and make a 33 gig Ram drive ... so our 33,600 Mg will fit if we do that!!! we are going to "block out" whole banks of ram that are going to be locked out from the CPU in proximity ... and the processors will have to make ram calls, to and from .... WAY WAY Far away "available" RAM this is going to kill CPU performance ... the reason that RAM Drives works on single CPU boxen is ... all system ram is local to the CPU ... so it doesnt matter i did say 4 things ... ill get back to you on that  maybe [Edit 2 times, last edit by Former Member at Dec 6, 2015 10:39:37 PM] |
|||
|
||||
|
OldChap
Veteran Cruncher UK Joined: Jun 5, 2009 Post Count: 978 Status: Offline Project Badges: 
|
Hmmmm...... Make ram calls to "distant" ram vs read/write to disk. I wonder which...?
----------------------------------------Do you really want to throw the cat among the pigeons? Take a good close look at Fatdog64 as an OS. Runs completely in ram as does boinc and wcg. In fact you have to learn quite a bit to get things to save to disk when you need that  [Edit 1 times, last edit by OldChap at Dec 6, 2015 11:21:56 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hmmmm...... Make ram calls to "distant" ram vs read/write to disk. I wonder which...? doesnt matter ... both are slower Do you really want to throw the cat among the pigeons? Yes ... thats the point School me Take a good close look at Fatdog64 as an OS. Runs completely in ram as does boinc and wcg. In fact you have to learn quite a bit to get things to save to disk when you need that 2 quick questions (just because i am to lazy to look it up) can you get the 32 bit Libraries for it? and does it support the "affinity-setting" (i believe thats what its called when a core is locked to a specific task) [Edit 2 times, last edit by Former Member at Dec 7, 2015 12:46:53 AM] |
||
|
|
|