Advanced search

Forums : Technical Support : Linux/Windows speed difference?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · Next

AuthorMessage
Profile [B^S] Acmefrog
Volunteer tester
Avatar

Send message
Joined: 8 Jun 07
Posts: 175
Credit: 446,074
RAC: 0
Message 858 - Posted: 6 Jul 2007, 19:15:25 UTC

Same here. I haven't seen any real difference between 1.18 and 1.19 on either my Intel or AMD. Once I get home I can compare better.
ID: 858 · Report as offensive     Reply Quote
Profile Truck Target
Avatar

Send message
Joined: 23 Jun 07
Posts: 8
Credit: 173,417
RAC: 0
Message 859 - Posted: 6 Jul 2007, 19:48:31 UTC

Same thing here. On 3 of my 4 Windows machines the WU times are basically the same with 1.19 as with 1.18. The 4th is just slow and hasn't reached a 1.19 WU yet.
ID: 859 · Report as offensive     Reply Quote
Profile [B^S] Acmefrog
Volunteer tester
Avatar

Send message
Joined: 8 Jun 07
Posts: 175
Credit: 446,074
RAC: 0
Message 860 - Posted: 6 Jul 2007, 21:45:35 UTC

I will say this that some WUs are taking about 30 minutes and others about half that on the same machine using the same WU version. I didn't know we had 2 different length WUs out there---or maybe they are the same and the changes only affected some of the 1.19 WUs.
ID: 860 · Report as offensive     Reply Quote
Profile Cori
Volunteer tester
Avatar

Send message
Joined: 22 May 07
Posts: 52
Credit: 34,836
RAC: 0
Message 861 - Posted: 6 Jul 2007, 21:58:59 UTC
Last modified: 6 Jul 2007, 21:59:39 UTC

I think we really had two different WU lengths with 1.18: the shorter WUs needed only about 1,200 seconds on my P4 lappy while the longer ones took about 2,900-3,800 seconds.
Now with 1.19 my test WU needed nearly 3,900 seconds.

Lovely greetings, Cori

ID: 861 · Report as offensive     Reply Quote
Profile Jayargh
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 25 Jun 07
Posts: 508
Credit: 2,282,158
RAC: 0
Message 862 - Posted: 6 Jul 2007, 22:30:41 UTC - in response to Message 856.  

I can't do anything about the memory leaks at this point, since I'm not proficient enough in Fortran 90 to go through another person's code and check for them.


Maybe a fellow colleague or advertise on campus for help.....or get more funding j/k
ID: 862 · Report as offensive     Reply Quote
Profile [B^S] Gamma^Ray
Volunteer tester
Avatar

Send message
Joined: 7 Jun 07
Posts: 47
Credit: 70,587
RAC: 0
Message 863 - Posted: 7 Jul 2007, 0:36:47 UTC

Im still seeing the two different lengths work units for me :

Jul 2007 0:31:25 UTC Over Success Done 1,558.59 4.75 pending
Jul 2007 23:48:55 UTC Over Success Done 639.73 1.95 pending
Jul 2007 0:31:25 UTC Over Success Done 1,381.97 4.21 pending
Jul 2007 0:10:58 UTC Over Success Done 1,307.20 3.98 pending
Jul 2007 23:28:24 UTC Over Success Done 674.97 2.06 pending

G^R

Windows-XP-Pro, AMD 3800X2, 5.10.28
ID: 863 · Report as offensive     Reply Quote
Profile [B^S] Doug Worrall
Volunteer tester
Avatar

Send message
Joined: 20 Jun 07
Posts: 29
Credit: 12,130
RAC: 0
Message 864 - Posted: 7 Jul 2007, 10:22:37 UTC - in response to Message 863.  
Last modified: 7 Jul 2007, 10:24:02 UTC

Im still seeing the two different lengths work units for me :

Jul 2007 0:31:25 UTC Over Success Done 1,558.59 4.75 pending
Jul 2007 23:48:55 UTC Over Success Done 639.73 1.95 pending
Jul 2007 0:31:25 UTC Over Success Done 1,381.97 4.21 pending
Jul 2007 0:10:58 UTC Over Success Done 1,307.20 3.98 pending
Jul 2007 23:28:24 UTC Over Success Done 674.97 2.06 pending

G^R

Basically same here
PClinuxOS 93A Big Daddy 2007 Final Release
:CPU type P4/GenuineIntel
Intel(R) Celeron(R) CPU 2.40GHz [Family 15 Model 4 Stepping 1][fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe constant_tsc up pni monitor ds_cpl cid xtpr]
Number of CPUs 1
2.6.19.2
Memory 749.02 MB
Cache 256 KB
Swap space 1592.34 MB
Total disk space 18.62 GB
Free Disk Space 16.46 GB
Measured floating point speed 1076.89 million ops/sec
Measured integer speed 2034.68 million ops/sec




ID: 864 · Report as offensive     Reply Quote
Profile Saenger
Volunteer tester
Avatar

Send message
Joined: 22 May 07
Posts: 110
Credit: 282,157
RAC: 0
Message 865 - Posted: 7 Jul 2007, 18:52:29 UTC

I see this different levels as well for my puter. Most WUs are around 1100 sec, some (~10%) around 450 sec, just take a look at my results.

I don't see any difference between them, If you, Scott, can determine a predictable feature in the short WUs, you can give them less credit. And, more important, you can perhaps see why they take just that short time and perhaps change the setup a bit, so that all WUs will profit from it.
Grüße vom Sänger
ID: 865 · Report as offensive     Reply Quote
Profile Scott
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 1 Apr 07
Posts: 662
Credit: 13,742
RAC: 0
Message 868 - Posted: 8 Jul 2007, 1:10:11 UTC

Well this is disconcerting. I still have some tests running in my office, so I'll take a look at those on Monday and get the final word.

I'll be sorely disappointed if I mixed up the executables or something. =(
Scott Kruger
Project Administrator, Cosmology@Home
ID: 868 · Report as offensive     Reply Quote
Profile Wang Solutions
Avatar

Send message
Joined: 28 Jun 07
Posts: 24
Credit: 1,020,464
RAC: 0
Message 869 - Posted: 8 Jul 2007, 2:18:59 UTC

Two comments from me:

1) Yes, there are definitely two different sets of WUs, the short ones being always roughly half the time of the long ones. This applies regardless of processor and operating system.

2) CAMB 1.19 is roughly 15-20% faster than 1.18 on my AMD machines, particularly on an X2, but is no faster and in fact slightly slower on the Intel machines. This suggests to me something to do either with L2 cache utilisation, or integer calculations.

ID: 869 · Report as offensive     Reply Quote
Profile [B^S] Acmefrog
Volunteer tester
Avatar

Send message
Joined: 8 Jun 07
Posts: 175
Credit: 446,074
RAC: 0
Message 870 - Posted: 8 Jul 2007, 4:11:12 UTC - in response to Message 869.  
Last modified: 8 Jul 2007, 4:13:06 UTC

Two comments from me:

1) Yes, there are definitely two different sets of WUs, the short ones being always roughly half the time of the long ones. This applies regardless of processor and operating system.

2) CAMB 1.19 is roughly 15-20% faster than 1.18 on my AMD machines, particularly on an X2, but is no faster and in fact slightly slower on the Intel machines. This suggests to me something to do either with L2 cache utilisation, or integer calculations.


1) That's what I hope was happening.

2) Taking into consideration that my AMD is slightly faster than my intel, I haven't noticed any real difference between the two and how fast they complete a WU.

P.S. @Wang I like your profile. Amusing. :)
ID: 870 · Report as offensive     Reply Quote
Profile Wang Solutions
Avatar

Send message
Joined: 28 Jun 07
Posts: 24
Credit: 1,020,464
RAC: 0
Message 871 - Posted: 8 Jul 2007, 6:42:39 UTC - in response to Message 870.  



P.S. @Wang I like your profile. Amusing. :)


Thanks Acme :-)

The sad part is that it is true! ;)


ID: 871 · Report as offensive     Reply Quote
Profile [B^S] Gamma^Ray
Volunteer tester
Avatar

Send message
Joined: 7 Jun 07
Posts: 47
Credit: 70,587
RAC: 0
Message 872 - Posted: 8 Jul 2007, 7:38:33 UTC

Strange as on my AMD Dual Core, I haven't noticed any increase at all since 1.19. If anything perhaps a tiny increase in wu length, But to close to really call for sure on that one.

Don't worry Scott, Compared to other projects that I worked with going through the early phases of their Alpha, Yours is running smoother than alot of them, With less bugs off hand that I see. :)

G^R
Windows-XP-Pro, AMD 3800X2, 5.10.28
ID: 872 · Report as offensive     Reply Quote
Profile Cori
Volunteer tester
Avatar

Send message
Joined: 22 May 07
Posts: 52
Credit: 34,836
RAC: 0
Message 876 - Posted: 8 Jul 2007, 18:12:45 UTC - in response to Message 869.  

... but is no faster and in fact slightly slower on the Intel machines...

That is what I noticed as well. ;-)
Lovely greetings, Cori

ID: 876 · Report as offensive     Reply Quote
Profile Scott
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 1 Apr 07
Posts: 662
Credit: 13,742
RAC: 0
Message 891 - Posted: 9 Jul 2007, 14:23:09 UTC

So I ran 40 random parameter files on the new and old CAMB in stand-alone mode to see the differences: here are the resulting times:

New CAMB
Old CAMB

Each time, the new CAMB's run-time is around 2/3 that of the old CAMB.

Now, it's certainly possible that 1.19 is optimized for AMD CPUs, since I compiled it on an AMD machine. I think ifort may have set optimizations for AMD as the default when I installed it, so I'll have to look through the options to see what the deal is.
Scott Kruger
Project Administrator, Cosmology@Home
ID: 891 · Report as offensive     Reply Quote
Profile Webmaster Yoda

Send message
Joined: 28 Jun 07
Posts: 21
Credit: 1,632,327
RAC: 0
Message 893 - Posted: 9 Jul 2007, 15:19:32 UTC - in response to Message 891.  
Last modified: 9 Jul 2007, 15:23:28 UTC

Now, it's certainly possible that 1.19 is optimized for AMD CPUs, since I compiled it on an AMD machine. I think ifort may have set optimizations for AMD as the default when I installed it, so I'll have to look through the options to see what the deal is.


I have a dual-booting Core 2 Duo (at 2.66GHz). In Windows it takes up to 20 minutes per result. In 64 bit Linux it's rarely more than 8 minutes.

However, my Athlon 64 3700+ (at 2.6GHz) running Windows XP takes close to 20 minutes too. My guess is that it will still be faster under Linux but I haven't got a spare HDD to try it (might run it as a DRBL node tomorrow to see if there is a big difference)
ID: 893 · Report as offensive     Reply Quote
Profile Scott
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 1 Apr 07
Posts: 662
Credit: 13,742
RAC: 0
Message 930 - Posted: 11 Jul 2007, 15:28:29 UTC
Last modified: 11 Jul 2007, 15:55:10 UTC

We're going to try another version of CAMB for Windows in a little bit to see if I've discovered the problem (which may have had to do with Windows ifc taking different optimization arguments than Linux ifc).

One of the differences that I've yet to find a way to fix is that the linux app uses static linking, while the Windows app uses dynamic linking. I would have used static linking for Windows, but ifc and cygwin don't seem to play very well with each other.

I'll keep you posted.

EDIT: Version 1.20 released. Tell me how it works out for you. I'll keep hacking away at the other problems in the mean time.
Scott Kruger
Project Administrator, Cosmology@Home
ID: 930 · Report as offensive     Reply Quote
Anchesa

Send message
Joined: 29 Jun 07
Posts: 6
Credit: 142,576
RAC: 0
Message 936 - Posted: 11 Jul 2007, 18:11:18 UTC - in response to Message 930.  

....

EDIT: Version 1.20 released. Tell me how it works out for you. I'll keep hacking away at the other problems in the mean time.


if the results are valit its working fantastic.

New Time between 450 and 1000 sec.
Old Time between 700 and 1800 sec.

The only Problem in the moment is the zero granted Credits for some Results:
some examples:

http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119398
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119397
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119391
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119390
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119389
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119387
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119386
http://cosmos.astro.uiuc.edu/cosmohome//workunit.php?wuid=119385
etc.

Greetings
Anchesa
ID: 936 · Report as offensive     Reply Quote
Honza
Volunteer tester

Send message
Joined: 21 May 07
Posts: 26
Credit: 5,222,146
RAC: 0
Message 942 - Posted: 11 Jul 2007, 19:48:43 UTC
Last modified: 11 Jul 2007, 19:49:19 UTC

Similar experience here - speed-up and no validation done.
Even when 3 results are done (out of quarum 2)
ID: 942 · Report as offensive     Reply Quote
Profile Scott
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 1 Apr 07
Posts: 662
Credit: 13,742
RAC: 0
Message 943 - Posted: 11 Jul 2007, 19:49:47 UTC

ID: 943 · Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · Next

Forums : Technical Support : Linux/Windows speed difference?