| View previous topic :: View next topic |
| Author |
Message |
Tulatin
Enlightened Master
Joined: 11 Dec 2002
Posts: 14664
Location: Canada
|
 Posted: Mon Sep 15, 2003 10:21 pm Post subject: |
 |
|
| Indeed it is... maybe folding treats them the same. |
|
| Back to top |
|
 |
joshvee
Black Belt 4th Degree
Joined: 12 Jul 2003
Posts: 4411
Location: Australia
|
| Posted: Mon Sep 15, 2003 10:41 pm Post subject: |
|
|
well well just tried something which works....I noticed when running the graphical client that it always set the client as machine 1.
As I am now running 2 console clients, and one of them is machine 1 and the other is machine 2, I stopped machine 1 console client, and ran a graphical client instead (adding the flags, -local -advmethods -forceasm, to the shortcut) and it worked. now I have 1 console client and 1 graphical client running....now we wil see which one is faster once and for all....ok maybe not in the way u think coz it depends on the WU right? well I placed the graphical client in the console client machine 1 folder so it is taking over the 13min WU. Now I just have to wait a couple of frames and look at the log file to get the times........
Hey that reminds me, I noticed some of u are timing the WUs over say a few hours to work out the frame time speed. In the graphical client just open the log file and every frame has a time on it so just subtract 1 frame from the next to give u a frame time speed....every frame in a given WU is done at the same time. 
_________________
Last edited by joshvee on Tue Sep 16, 2003 12:26 am; edited 2 times in total |
|
| Back to top |
|
|
Tulatin
Enlightened Master
Joined: 11 Dec 2002
Posts: 14664
Location: Canada
|
| Posted: Mon Sep 15, 2003 10:43 pm Post subject: |
|
|
| Yep - the log file makes calculations VERY easy... |
|
| Back to top |
|
|
joshvee
Black Belt 4th Degree
Joined: 12 Jul 2003
Posts: 4411
Location: Australia
|
| Posted: Tue Sep 16, 2003 12:32 am Post subject: |
|
|
OK heres the results of my little test.
the console client did the frames of the WU in 13mins 22 seconds per frame. The graphical client did the same WU frames with the garphical window open in 13mins 35 seconds and with the window closed in 11mins 30 seconds.
quite a difference.
So at least on my Intel P4 2.8c the graphical client (window closed) is faster than the console client by 15 to 20%
_________________
|
|
| Back to top |
|
|
Aussie
Mobo-fu Master
Joined: 22 Nov 1999
Posts: 10244
Location: Mackay, Queensland, Australia
|
| Posted: Tue Sep 16, 2003 12:55 am Post subject: |
|
|
| Quote: | | every frame in a given WU is done at the same time. |
That is not the case in my experience, a similar time yes but not necessarily the exact same time.
In my experience "short term" evaluations of folding performance are nearly always proven incorrect over the long term.
Your observations are interesting none the less.
Aussie
_________________
Our lives begin to end the day we become silent about things that matter - Martin Luther King |
|
| Back to top |
|
|
joshvee
Black Belt 4th Degree
Joined: 12 Jul 2003
Posts: 4411
Location: Australia
|
| Posted: Tue Sep 16, 2003 4:36 am Post subject: |
|
|
Yes I noticed that to... they do vary by a few seconds to a few mins depending on what else is happening on the computer.
I have now redone the test to show more accurately what is going on. I stopped as many background functions as I could whilst still keeping the comp functional (got it down to 18 processes from 36) I then waited for the graphical to complete the WU and opened the log. The console client had compleated 49 frames when I got the graphical client to take over the same WU and averaged all the frames with the computer ONLY folding. so that means:
console : total time = 658.24mins / 49 frames = average 13.44 mins per frame.
Graphical : total time = 591.6mins / 51 frames = average 11.6 mins per frame
and this was on the same WU with the computer only folding.
_________________
|
|
| Back to top |
|
|
Aussie
Mobo-fu Master
Joined: 22 Nov 1999
Posts: 10244
Location: Mackay, Queensland, Australia
|
| Posted: Tue Sep 16, 2003 7:18 am Post subject: |
|
|
An interesting observation but inconclusive because both tests were not run under the same conditions.
The testing of the graphical client was carried out with a reduced number of processes running on the machine:-
| Quote: | | I stopped as many background functions as I could whilst still keeping the comp functional (got it down to 18 processes from 36) I then waited for the graphical to complete the WU |
Unless the console client was tested under those same conditions the test is inconclusive.
The results do suggest that your assessment may be correct because of the relatively large difference in times however more accurate testing of both clients under identical conditions would be required to prove the theory and accurately establish the differences if in fact there are any.
Questions that spring to mind are, was there only one client running at the time of each test or two and was hyperthreading enabled or disabled? Do the results vary if hyperthreading is enabled or disabled and /or one or two instances of the client are running? Do the results differ from one work unit to the next?
Trying to assess the benefits of optimisations or different methods of running the folding clients is time consuming to say the least.
Aussie
_________________
Our lives begin to end the day we become silent about things that matter - Martin Luther King |
|
| Back to top |
|
|
joshvee
Black Belt 4th Degree
Joined: 12 Jul 2003
Posts: 4411
Location: Australia
|
| Posted: Tue Sep 16, 2003 8:29 am Post subject: |
|
|
im doing as you suggest now...will post the results tomorrow
_________________
|
|
| Back to top |
|
|
joshvee
Black Belt 4th Degree
Joined: 12 Jul 2003
Posts: 4411
Location: Australia
|
| Posted: Tue Sep 16, 2003 2:08 pm Post subject: |
|
|
ok I've done 70% of the test and am running tests on 2 machines and have discovered something a bit annoying. From the results so far there seems to be no point running 2 clients on a hyperthreaded machine, 1 on CPU 0 and 1 on CPU 1....If you run 1 client on CPU 0 and 1 (the default setting) it runs twice as fast.  so whats the point of hyperthreading...another Intel marketing gimmic?
_________________
|
|
| Back to top |
|
|
Southwind25
Black Belt 5th Degree
Joined: 01 Oct 2002
Posts: 7339
Location: Kansas
|
| Posted: Tue Sep 16, 2003 3:37 pm Post subject: |
|
|
| Quote: | so whats the point of hyperthreading...another Intel marketing gimmic?
|
ummmmmmmmm could be  along with quad pumped........
_________________
 |
|
| Back to top |
|
|
|
