| Author | Message |
|---|
ToluVolunteer developer
Send message
Joined: Aug 5 04
Posts: 173
Credit: 1,992,981
RAC: 0
|
|
version 4.22 of the sulphur cycle is available.
We\'ve improved the performance somewhat.
PS: Note there shotly be a mac release ( before the wknd)
as well as a new Advanced viz update to resolve some of the current quirks.
____________
|
|
|
|
|
|
SC 4.22 for Windows downloaded and started fine.
Download is a bit smaller (couple MBs).
Sulphur_se_4.22_windows_intelx86.zip does no longer contains 6.3MB globe.tga file. Does it mean that it is no londer needed (when older 4.19 models are finished of course) and you managed to use only globe.rgb for the excellent graphics? I would say so because former 4.19 used 60+52MB when graphics is on, 4.22 uses 59+only 22MB...
There might be a slight speed improvement (now at 3.12 sec/TS vs. 3.21-3.33 or other 4 SC 4.19) but it\'s too soon to make conslusions.
Good job!
____________
<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a> |
|
|
|
|
version 4.22 of the sulphur cycle is available.
We\'ve improved the performance somewhat.
PS: Note there shotly be a mac release ( before the wknd)
as well as a new Advanced viz update to resolve some of the current quirks.
I just hope that version doesn\'t catch signal 11 as I got in this,
this and
this result. They all crashed when they were about to be suspended. I\'m wondering if something with my PC is wrong that it crashes Sulphor WUs so often...
____________
|
|
|
|
|
|
All sulphur models on my PowerMac G5 running MacOSX 10.3.9 Panther have crashed too. I have updated to BOINC-manager 5.2.13 and hope for a better future.
____________
|
|
|
|
|
|
My first Linux FC4 downloaded SC4.22, but it runs bench 518/876 while I expected 890/1560 on this single processor PIII - 1000 MHz.
Furthermore, the visialization button is grayed out and therefore unusable.
Looks to be be caused by the earlier mentioned absence of the globe.rbg
Is there a relation between the two symptoms ?
NB: Every possible package of this Lnx fedora version is installed.
Before this the board/processor have been running as expected under XP, but it is terrible slow under these new circumstances. Also looks like it generating more disk consumption than usual.
For now, I have stopped running this pc - too sloooow
____________
|
|
|
|
|
|
Rather impressive speedup from 4.19 to 4.22 in WinXP sulphur. Default with 4.19 on a Athlon64 3400+ (2.2 GHz) was 3.18 sec/TS. Now it is 2.55 sec/TS with 4.22 which is a 20% improvement!
Kilcock...I\'m not sure about the globe, but the benchmark matters nothing in CPDN. There is no correlation between the benchmark scores and the sec/TS in CPDN. My P4 3.4 GHz had very low benchmark scores in Linux compared to XP, and yet was much faster in the sec/TS than the same PC in WinXP. |
|
|
|
|
All sulphur models on my PowerMac G5 running MacOSX 10.3.9 Panther have crashed too. I have updated to BOINC-manager 5.2.13 and hope for a better future.
Mine too.
Actually, I even see this with Slab models. Depressingly enough I now only run CPDN on Windows machines. Of course, my windows machines out number my lonesome G5 so maybe this is not a bad thing ... :)
____________
|
|
|
ToluVolunteer developer
Send message
Joined: Aug 5 04
Posts: 173
Credit: 1,992,981
RAC: 0
|
Does it mean that it is no londer needed (when older 4.19 models are
The globe.tga file was only shipped ( or intended to ) on win32 distributions. It was an extra so you could toggle between 2 reliefs.
One was really high res.
It did make the download quite large so i\'ve taken it out.
globe.rgb is now defacto.
There\'s still a standalone version sulphur_gfx_* which can be run
on its own
____________
|
|
|
|
|
|
Where can I download this globe.rgb, globe.tga or standalone version sulphur_gfx_* for Linux.
So far a globe.xx was not downloaded and if you ask me graphics data is written to harddisk instead of the display!
geophi:
not this time, cpu under xp would do 12.5 s/Ts, while present results under Linux FC4 is around 17s/Ts
(Not sure, because can\'t see the actual Ts it is on! No graphics!!
Has anybody positive experiances with FC4/SC4.22?
____________
|
|
|
|
|
geophi:
not this time, cpu under xp would do 12.5 s/Ts, while present results under Linux FC4 is around 17s/Ts
(Not sure, because can\'t see the actual Ts it is on! No graphics!!
By starting BOINC in a terminal window ( ./boinc from the BOINC directory), you should see the average s/TS in the scrolling timesteps in that window (the number after the AVG label) so you shouldn\'t need the graphic/globe to get the s/TS. |
|
|
ToluVolunteer developer
Send message
Joined: Aug 5 04
Posts: 173
Credit: 1,992,981
RAC: 0
|
Where can I download this globe.rgb, globe.tga or standalone version sulphur_gfx_* for Linux.
.....
Has anybody positive experiances with FC4/SC4.22?
Navigate to BOINC\\projects\\www.climateorediction.net\\
There, you should find sulphur_gfx & a globe.rgb and
more important sulphur_se_4.22*
If these files aren\'t present, then they weren\'t downloaded and something\'s wrong.
If they are run the standalone with ./sulphur_gfx*
I\'ve seen a bizzare similar behaviour on FC4 with the graphics window.
Nothing that could lead to a computationl error though.
____________
|
|
|
|
|
geophi:
not this time, cpu under xp would do 12.5 s/Ts, while present results under Linux FC4 is around 17s/Ts
(Not sure, because can\'t see the actual Ts it is on! No graphics!!
By starting BOINC in a terminal window ( ./boinc from the BOINC directory), you should see the average s/TS in the scrolling timesteps in that window (the number after the AVG label) so you shouldn\'t need the graphic/globe to get the s/TS.
Hi. that\'s cool! thanks Geo-phi
Don\'t like to run X anyway, to much performance loss.
____________
|
|
|
|
|
|
Well, as you surmised Kilcock, 4.22 is slower in Linux on Intel. A dual Xeon running two sulphur 4.21 models on Fedora Core 3 averaged about 2.95 sec/TS per model. Suspending one so a sulphur 4.22 model downloaded resulted in that 4.22 model running at about 4.20 sec/TS. So I have confirmed your findings about Linux sulphur 4.22 and Intel. |
|
|
|
|
Where can I download this globe.rgb, globe.tga or standalone version sulphur_gfx_* for Linux.
.....
Has anybody positive experiances with FC4/SC4.22?
Navigate to BOINC\\projects\\www.climateorediction.net\\
There, you should find sulphur_gfx & a globe.rgb and
more important sulphur_se_4.22*
If these files aren\'t present, then they weren\'t downloaded and something\'s wrong.
If they are run the standalone with ./sulphur_gfx*
I\'ve seen a bizzare similar behaviour on FC4 with the graphics window.
Nothing that could lead to a computationl error though.
Tolu,
As Geophi, and a third person has comfirmed severe slowdowns of SC4.22 on linux, I strongly advise to take this version from the server asap.
and revert back to SC4.21 till reasons of slowdowns are solved.
Here is an other thread on the issue:
http://www.climateprediction.net/board/viewtopic.php?p=31850#31850
-----------
Eric.ie
____________
|
|
|
|
|
|
Linux Fedora C4 and AMD XP, BOINC 4.13.
Shows graphics alright with the .rgb.
Well, then I started toggling around with the keyboard but when hitting 0 I got this:
sulphur_e138_000654596 - PH 1 TS 0000078 A - 02/12/1810 15:00 - H:M:S=0000:04:37 AVG= 3.55 DLT= 2.38
sulphur_e138_000654596 - PH 1 TS 0000079 A - 02/12/1810 15:30 - H:M:S=0000:04:39 AVG= 3.54 DLT= 2.38
2005-12-02 18:43:52 [climateprediction.net] Unrecoverable error for result sulphur_e138_000654596_0 (process got signal 11)2005-12-02 18:43:52 [climateprediction.net] Unrecoverable error for result sulphur_e138_000654596_0 (process got signal 11)2005-12-02 18:43:52 [---] request_reschedule_cpus: process exited
2005-12-02 18:43:52 [climateprediction.net] Computation for result sulphur_e138_000654596_0 finished
2005-12-02 18:43:52 [---] Allowing work fetch again.
2005-12-02 18:43:52 [---] Resuming round-robin CPU scheduling.
2005-12-02 18:44:56 [climateprediction.net] Sending scheduler request to http://climateapps2.oucs.ox.ac.uk/cpdnboinc_cgi/cgi
2005-12-02 18:44:56 [climateprediction.net] Reason: To fetch work
2005-12-02 18:44:56 [climateprediction.net] Requesting 8640 seconds of new work, and reporting 1 results
2005-12-02 18:45:01 [climateprediction.net] Scheduler request to http://climateapps2.oucs.ox.ac.uk/cpdnboinc_cgi/cgi succeeded
2005-12-02 18:45:01 [climateprediction.net] Message from server: No work sent
2005-12-02 18:45:01 [climateprediction.net] Message from server: No work sent
2005-12-02 18:45:01 [climateprediction.net] Message from server: (reached daily quota of 1 results)
2005-12-02 18:45:01 [climateprediction.net] Message from server: (reached daily quota of 1 results)
2005-12-02 18:45:01 [climateprediction.net] No work from project
Guess it should run a bit more to see the speed.
(toggle 6 shows the contributors, hehe :-)
)
____________
 |
|
|
|
|
Linux Fedora C4 and AMD XP, BOINC 4.13.
Shows graphic alright with the .rgb.
2005-12-02 18:45:01 [climateprediction.net] Message from server: (reached daily quota of 1 results)
2005-12-02 18:45:01 [climateprediction.net] No work from project
Guess it should run a bit more to see the speed.
(toggle 6 shows the contributors :-))
I\'m afraid you have crashed you sulphur model. and worse, because you where connected, the crash is uploaded to the server. Therefore restoring a backup would not work.
I only can hope that you just started the crashed model.
Actually on my side there are two issues:
1 Grapics
2 Slowdown
As the slowdown is severe and seems to be related to SC4.22, issue 2 has become the main priority. Need some Admin action to get this version from the server, before more people get the same slowdown !
____________
|
|
|
|
|
|
Oh, I\'m rather used to run first releases
We\'ll see tomorrow about the speed here
____________
|
|
|
|
|
Oh, I\'m rather used to run first releases
We\'ll see tomorrow about the speed here
Yes, Please do!
These could be a slight posibility, that this slowdown and maybe my globe absence is caused by an longer existing download issue. (model starts execution before the application is fully downloaded)
Hence a second model download, provided the application files are already ondisk, might perform as expected.
Lets wait and hope for the best!
Think all Admins are in Melbourne anyway.
____________
|
|
|
|
|
|
I pulled an AMD64 3700+ away from spinup for a little while and booted into FC4 x86-64. Running sulphur 4.22 gave 2.85 sec/TS, which is slightly faster than this same PC ran older sulphur versions.
I did the same thing with a P4 3.4 GHz in Mandrake 10.1. With previous versions of sulphur, it was about 2.25 sec/TS. Now it\'s about 3.5 sec/TS.
So, it\'s very repeatable. Only Intel is having the slowdowns with this version. |
|
|
|
|
I pulled an AMD64 3700+ away from spinup for a little while and booted into FC4 x86-64. Running sulphur 4.22 gave 2.85 sec/TS, which is slightly faster than this same PC ran older sulphur versions.
I did the same thing with a P4 3.4 GHz in Mandrake 10.1. With previous versions of sulphur, it was about 2.25 sec/TS. Now it\'s about 3.5 sec/TS.
So, it\'s very repeatable. Only Intel is having the slowdowns with this version.
Must have been done as a revenge for the ss2/ss3 treatment of AMD\'s by the Intel compiler ;-)
____________
|
|
|
|
|
|
AMD XP Linux Fedora C4
Now running at 3.49 compared to 3.66 before, so a 5% boost then.
One odd thing is that sulphur_um_4.22_i686-pc-linux-gnu started running at nice level 0 directly after the download. After stopping and restarting it now runs at nice 19 as it should.
____________
|
|
|
|
|
|
So, does this mean I should abort all my sulfur models and get new ones? :)
Just kidding ... :)
____________
|
|
|
|
|
So, does this mean I should abort all my sulfur models and get new ones? :)
Just kidding ... :)
Hmmm, On an AMD 2000+ XPSP2, which was on SC4.19 @ Trickle 2.5, I did and now runs 8% better on SC4.22.
On an Intel PIII - 1000MHz XPSP1, which was on SC4.19 @ Trickle 4, I initialy did but reverted back after the first checkpoint because there was a performance degradation of 14% to 18%.
On a sister Intel PIII - 1000MHz Linux FC4, I reported a performance loss of 36%.
Due to Model performance variations, it is acually not accurate to compare the different models. But lacking a accurate testmodel (and the ability to plug it into the application)one has to do with it.
Earlier I have concluded that the Linux version was not good. But having more overview now, it seems to me the performance difference is influenced by 3 factors, being:
- Processor type: Intel vs. AMD
- Processor speed: slow vs. fast CPU (includes ss2 ss3)
- Used OS: Linux vs. Windows
So a relative slow intel cpu crunching SC4.22 on Linux gets the sum of 3 dips!
In this situation we get the best performance boost out of a certain machine if there is a conditional version push or if we could choose the version application ourselfs, instead of the latest version being pushed automatically to your machine.
____________
|
|
|
ToluVolunteer developer
Send message
Joined: Aug 5 04
Posts: 173
Credit: 1,992,981
RAC: 0
|
So, it\'s very repeatable. Only Intel is having the slowdowns with this version.
I\'ll look into this
____________
|
|
|
ToluVolunteer developer
Send message
Joined: Aug 5 04
Posts: 173
Credit: 1,992,981
RAC: 0
|
|
We\'ve released a new version of the sulphur cycle application
for the mac - version 4.22.
This version resolves the crash bug reported in earlier versions.
Apologies for the delayed release.
If you\'ve had problems prior to this, PLEASE reset the project and
download the latest version.
____________
|
|
|
|
|
|
What deadlines are given for v. 4.22? The v. 4.19 wu I\'m working on has a deadline far far less than a year - which means BOINC practically only wants to run that one to meet the deadline... Other projects get no CPU time unless I force a temporary suspend of CPDN
____________
 |
|
|
|
|
|
Originally, there were 500 models with a short deadline of 4 or 5 months, because the results were needed ASAP.
Two months ago, a new batch of 50,000 were generated, with a deadline of about a year.
Mine was sent on 10 Oct 2005, and is expected back on 22 Sep 2006.
This is not set in concrete.
You appear to still be working on one of the original data sets.
If you look at this projects front page, 710 sulphur models have been completed, (although it is more than this, if you look at Project stats page).
So there is no real hurry.
|
|
|
|
|
|
thanks for the info :)
____________
|
|
|
|
|
|
Back when Sulphur was being developed, someone said that an experimental version ran much better on AMD with a certain switched turned on --
Does anyone know if moves have been made to allow AMD processors to run the projects effcienetly? The whole \"Disable SS2/SS3 if it\'s not a Pentium IV\" was quite under-handed, and takes away a lot of processing ability from an otherwise VERY powerful AMD processor.
Matthew
____________
<img src=\"http://boinc.mundayweb.com/one/stats.php?userID=2118&prj=4\"> |
|
|
|
|
Does anyone know if moves have been made to allow AMD processors to run the projects effcienetly? The whole \"Disable SS2/SS3 if it\'s not a Pentium IV\" was quite under-handed, and takes away a lot of processing ability from an otherwise VERY powerful AMD processor.
Well, it may be obscured in this thread because of all the posts about Linux performance loss by Intel, but this post showed the performance improvement by AMD in Windows with sulphur 4.22.
|
|
|
