Welcome to MilkyWay@home

CPU crunching - A few questions.

Message boards : Number crunching : CPU crunching - A few questions.
Message board moderation

To post messages, you must log in.

AuthorMessage
Virtual Boss*

Send message
Joined: 26 Jul 09
Posts: 11
Credit: 1,240,681
RAC: 0
Message 44333 - Posted: 27 Nov 2010, 8:36:13 UTC

First I will explain why I am asking these questions.

I posted in previous thread (Every new WU crashing!!) when 1 of my computers started crashing WU's.
I have since discovered another machine is crashing. (The OS is linux and when MW crashes it takes BOINC down with it)
Both machines are old and do not have SSE2.

All my computers recently got version 45.
I have seen in the forum some posts about a version 45 (SSE2) app but none of my comps have it. I have tried to find it but no luck so far.

Further testing after my last thread has revealed that the version 45 app I have works fine but slow on most of my comps but not on the two without SSE2.

Currently I am running the old 0.21 opti app because it is ~3x faster than version 45.

Q1 - Where do I find the Version 45 SSE2 app? (and is it faster?)

Q2 - Why is there two apps for version 45 if both need SSE2?

Q3 - If Verssion 45 wont run on a cpu that doesn't have SSE2, why isn't the (SSE2) variant the default?.

Thanks in advance for any answers.

Bruce
ID: 44333 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 44334 - Posted: 27 Nov 2010, 8:54:57 UTC - in response to Message 44333.  
Last modified: 27 Nov 2010, 9:03:29 UTC

I've stopped trying to run any CPU Milkyway on any on my PCs despite being in a Milkyway competition. Most of my PCs are 'old' AND they all seem to drag out the finish time and I have aborted WUs despite having crunched in excess of 20 hours but creeping up to 50+ more to go. I used to be able to run Milkyway CPU WUs on these which certainly took less than 20 hours. Sorry I can't answer your questions and hope you find your answers so we can all run MW CPU WUs.

ID: 44334 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 44372 - Posted: 27 Nov 2010, 21:42:56 UTC - in response to Message 44333.  

Further testing after my last thread has revealed that the version 45 app I have works fine but slow on most of my comps but not on the two without SSE2.
The problem it looks like you're having on the old systems is unrelated to SSE2. It looks like a BOINC problem, and you also have a very old version of BOINC installed (5.10.24). Try updating it to the latest version.

Q1 - Where do I find the Version 45 SSE2 app? (and is it faster?)
You should get it by default if you have SSE2. The SSE2 one is much, much faster than the non-SSE2 one (somewhere over 2x faster).

Q2 - Why is there two apps for version 45 if both need SSE2?
Both don't.

Q3 - If Verssion 45 wont run on a cpu that doesn't have SSE2, why isn't the (SSE2) variant the default?.
I consider the SSE2 version to be the default since the vast majority of systems have it and it's much faster. There however is still a non-SSE2 version of the separation. The N-body does require SSE2 because the x87 FPU is a pain and it's harder / in some cases impossible to get consistent results from it.
ID: 44372 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Virtual Boss*

Send message
Joined: 26 Jul 09
Posts: 11
Credit: 1,240,681
RAC: 0
Message 44377 - Posted: 28 Nov 2010, 1:41:19 UTC - in response to Message 44372.  

Thanks for the info Matt.

I shall take a bigger hammer to my comps to try to get them to behave. :)

The problem it looks like you're having on the old systems is unrelated to SSE2. It looks like a BOINC problem, and you also have a very old version of BOINC installed (5.10.24). Try updating it to the latest version.


Hmmm, My P4m (has SSE2) is working OK with the old boinc, but it only got the non-SSE2 app from the server. That was what made me jump to my conclusions.

Some more deleting/installing/tweaking/testing about to start.

Bruce
ID: 44377 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Chris
Avatar

Send message
Joined: 3 Oct 10
Posts: 42
Credit: 320,242
RAC: 0
Message 46907 - Posted: 2 Apr 2011, 2:29:20 UTC - in response to Message 44334.  

Some of my CPU work units state 20-40 hours but they rarely take over 8 hours. I am not sure why. My guess is the time prediction algorithm is not quite accurate. Or perhaps because PCs can store solutions to the most commonly done problems in the cache therefore the more work they do, perhaps the larger their lookup table becomes and the faster they go. I dont know. I'm guessing.
ID: 46907 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
sandor

Send message
Joined: 7 May 10
Posts: 8
Credit: 39,603,036
RAC: 0
Message 46909 - Posted: 2 Apr 2011, 3:04:56 UTC - in response to Message 46907.  

Some of my CPU work units state 20-40 hours but they rarely take over 8 hours. I am not sure why. My guess is the time prediction algorithm is not quite accurate. Or perhaps because PCs can store solutions to the most commonly done problems in the cache therefore the more work they do, perhaps the larger their lookup table becomes and the faster they go. I dont know. I'm guessing.


The CPU WU's use as many Cores as you have available, at least the .40 ones do. So that is probably why there's a discrepancy between the estimated time and the actual time. I get 5+ hours estimated for a .40, but <1 hour actual and it uses all 6(5.85 actually) of my Cores.
ID: 46909 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 46910 - Posted: 2 Apr 2011, 3:08:11 UTC - in response to Message 46907.  

Some of my CPU work units state 20-40 hours but they rarely take over 8 hours. I am not sure why. My guess is the time prediction algorithm is not quite accurate.
The N-body time estimates are almost a complete guess. It's hard to predict.
ID: 46910 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : CPU crunching - A few questions.

©2024 Astroinformatics Group